enhance!: send OTLP metrics under OpenTelemetry semantic conventions, add runner state (#1282)

Sends the OTLP metrics under [OpenTelemetry semantic convention](https://opentelemetry.io/docs/specs/semconv/) names where the runner's data matches the definition, like `cicd.worker.count`, `cicd.system.errors` and the `go.*` and `process.*` runtime metrics, and under `gitea.runner.*` otherwise. OTel dashboards then work without remapping, as they already do for the traces from https://gitea.com/gitea/runner/pulls/1206.

Adds a `gitea_runner_state` gauge (`busy`, `idle`, `unavailable`), and `/readyz` now fails once the runner shuts down.

**Breaking**

1. OTLP metric names, units, types and attribute values change, and the info and start time metrics move to resource attributes.
1. Prometheus `gitea_runner_client_errors_total` gains a `code` label, and `gitea_runner_job_total` counts jobs stopped by `runner.timeout` as `timeout` instead of `failure`.

---------

Co-authored-by: silverwind <[email protected]>
Co-authored-by: silverwind <[email protected]>
Co-authored-by: bircni <[email protected]>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1282
Reviewed-by: silverwind <[email protected]>
Co-authored-by: wmTJc9IK0Q <[email protected]>

.gitea/workflows/release-nightly.yml

@@ -20,11 +20,10 @@ jobs:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
         with:
           fetch-depth: 0
-      # Custom publishers (the R2 mirror below) run as the very last
+      # Custom publishers (the R2 upload below) run as the very last
       # step of goreleaser's publish pipeline, after the Gitea release
-      # has already been created and every artifact already uploaded
+      # has already been created. Fail here instead, before anything
-      # to S3. Fail here instead, before anything is built or
+      # is built or published, if the R2 secrets are missing.
-      # published, if the R2 secrets are missing.
       - name: check R2 configuration
         run: sh scripts/upload-r2.sh --check-config
         env:
@@ -35,6 +34,7 @@ jobs:
       - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
         with:
           go-version-file: "go.mod"
+          check-latest: true
       - name: goreleaser
         uses: goreleaser/goreleaser-action@f06c13b6b1a9625abc9e6e439d9c05a8f2190e94 # v7
         with:
@@ -42,11 +42,6 @@ jobs:
           args: release --nightly
         env:
           GORELEASER_KEY: ${{ secrets.GORELEASER_KEY }}
-          AWS_REGION: ${{ secrets.AWS_REGION }}
-          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_KEY_ID }}
-          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
-          S3_REGION: ${{ secrets.AWS_REGION }}
-          S3_BUCKET: ${{ secrets.AWS_BUCKET }}
           R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
           R2_BUCKET: ${{ secrets.R2_BUCKET }}
           R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
@@ -79,13 +74,13 @@ jobs:
           fetch-depth: 0 # all history for all branches and tags
 
       - name: Set up QEMU
-        uses: docker/setup-qemu-action@96fe6ef7f33517b61c61be40b68a1882f3264fb8 # v4
+        uses: docker/setup-qemu-action@99012661954931238ded8c8b007157a8430204e1 # v4
 
       - name: Set up Docker BuildX
-        uses: docker/setup-buildx-action@bb05f3f5519dd87d3ba754cc423b652a5edd6d2c # v4
+        uses: docker/setup-buildx-action@f87e5991a6d7451dcb8d9637bfbc97413f497069 # v4
 
       - name: Login to DockerHub
-        uses: docker/login-action@abd2ef45e78c5afb21d64d4ca52ee8550d9572c7 # v4
+        uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4
         with:
           username: ${{ secrets.DOCKER_USERNAME }}
           password: ${{ secrets.DOCKER_PASSWORD }}
@@ -99,7 +94,7 @@ jobs:
           echo REPO_VERSION=$(git describe --tags --always | sed 's/-/+/' | sed 's/^v//') >> $GITHUB_OUTPUT
 
       - name: Build and push
-        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7
+        uses: docker/build-push-action@c3c9e263c25d99ce0380d002d59b67737d91b0dc # v7
         with:
           context: .
           file: ./Dockerfile

.gitea/workflows/release-tag.yml

@@ -12,11 +12,10 @@ jobs:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
         with:
           fetch-depth: 0 # all history for all branches and tags
-      # Custom publishers (the R2 mirror below) run as the very last
+      # Custom publishers (the R2 upload below) run as the very last
       # step of goreleaser's publish pipeline, after the Gitea release
-      # has already been created and every artifact already uploaded
+      # has already been created. Fail here instead, before anything
-      # to S3. Fail here instead, before anything is built or
+      # is built or published, if the R2 secrets are missing.
-      # published, if the R2 secrets are missing.
       - name: check R2 configuration
         run: sh scripts/upload-r2.sh --check-config
         env:
@@ -27,6 +26,7 @@ jobs:
       - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
         with:
           go-version-file: "go.mod"
+          check-latest: true
       - name: Import GPG key
         id: import_gpg
         uses: crazy-max/ghaction-import-gpg@2dc316deee8e90f13e1a351ab510b4d5bc0c82cd # v7
@@ -41,11 +41,6 @@ jobs:
           args: release
         env:
           GORELEASER_KEY: ${{ secrets.GORELEASER_KEY }}
-          AWS_REGION: ${{ secrets.AWS_REGION }}
-          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_KEY_ID }}
-          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
-          S3_REGION: ${{ secrets.AWS_REGION }}
-          S3_BUCKET: ${{ secrets.AWS_BUCKET }}
           R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
           R2_BUCKET: ${{ secrets.R2_BUCKET }}
           R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
@@ -82,13 +77,13 @@ jobs:
           fetch-depth: 0 # all history for all branches and tags
 
       - name: Set up QEMU
-        uses: docker/setup-qemu-action@96fe6ef7f33517b61c61be40b68a1882f3264fb8 # v4
+        uses: docker/setup-qemu-action@99012661954931238ded8c8b007157a8430204e1 # v4
 
       - name: Set up Docker BuildX
-        uses: docker/setup-buildx-action@bb05f3f5519dd87d3ba754cc423b652a5edd6d2c # v4
+        uses: docker/setup-buildx-action@f87e5991a6d7451dcb8d9637bfbc97413f497069 # v4
 
       - name: Login to DockerHub
-        uses: docker/login-action@abd2ef45e78c5afb21d64d4ca52ee8550d9572c7 # v4
+        uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4
         with:
           username: ${{ secrets.DOCKER_USERNAME }}
           password: ${{ secrets.DOCKER_PASSWORD }}
@@ -108,7 +103,7 @@ jobs:
             suffix=${{ matrix.variant.tag_suffix }},onlatest=true
 
       - name: Build and push
-        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7
+        uses: docker/build-push-action@c3c9e263c25d99ce0380d002d59b67737d91b0dc # v7
         with:
           context: .
           file: ./Dockerfile

.gitea/workflows/test.yml

@@ -1,38 +1,47 @@
 name: checks
 on:
-  push:
-    branches:
-      - main
   pull_request:
+  workflow_dispatch:
+
+env:
+  # The runner image ships a stale docker.io login; point docker at an empty config so
+  # image pulls go straight to anonymous instead of attempting (and failing) that auth
+  # first. The path must be a literal: the `runner` context is unavailable in workflow-level
+  # env, so `${{ runner.temp }}` would resolve to empty and config.Dir() would fall back
+  # to ~/.docker with the stale credentials.
+  DOCKER_CONFIG: /tmp/docker-noauth
 
 jobs:
   lint:
     name: check and test
     runs-on: ubuntu-latest
-    env:
-      # The runner image ships a stale docker.io login; point docker at an empty config so
-      # image pulls go straight to anonymous instead of attempting (and failing) that auth
-      # first. The path must be a literal: the `runner` context is unavailable in job-level
-      # env, so `${{ runner.temp }}` would resolve to empty and config.Dir() would fall back
-      # to ~/.docker with the stale credentials.
-      DOCKER_CONFIG: /tmp/docker-noauth
     steps:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
       - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
         with:
           go-version-file: 'go.mod'
+          check-latest: true
       - name: prepare anonymous docker config
         run: mkdir -p "$DOCKER_CONFIG" && echo '{}' > "$DOCKER_CONFIG/config.json"
       # Pre-pull act/runner's two largest base images so a slow pull can't dominate `make test`;
       # the rest (alpine/ubuntu) pull on demand, absorbed by the make-test -timeout. The host
       # daemon retains them between runs, so this is usually a fast manifest re-check.
       - name: pre-pull test images
+        env:
+          TEST_JOB_IMAGE: node:24-bookworm-slim@sha256:0e0ff40c39bc087845bfb27465a0df4ea419520094bc35842ff83dd8cbe6f9b6 # renovate: datasource=docker
+          TEST_SERVICE_IMAGE: nginx:alpine@sha256:df221db836e1754089190208cee7eeda94f233197056426eda74a43ab1abeac2 # renovate: datasource=docker
         run: |
-          for img in node:24-bookworm-slim nginx:alpine; do
+          for image in "$TEST_JOB_IMAGE" "$TEST_SERVICE_IMAGE"; do
-            for try in 1 2 3; do docker pull "$img" && break || sleep 5; done
+            for attempt in 1 2 3; do
+              docker pull "$image" && break
+              [ "$attempt" = 3 ] || sleep 5
+            done
+            docker tag "$image" "${image%@*}"
           done
       - name: lint
         run: make lint
+      - name: checks
+        run: make checks
       - name: build
         run: make build
       - name: test
@@ -42,7 +51,28 @@ jobs:
       # after `make test` so the images it needs are already present on the host daemon.
       - name: test against dind image
         run: make test-dind
+      - name: test against podman
+        run: make test-dind TARGET=podman
       - name: coverage report
         run: |
           make coverage-report
           cat .tmp/coverage.md >> "$GITHUB_STEP_SUMMARY"
+
+  e2e:
+    name: gitea compatibility (${{ matrix.gitea_image }})
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        gitea_image:
+          - gitea/gitea:latest
+          - gitea/gitea:main-nightly
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
+      - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
+        with:
+          go-version-file: 'go.mod'
+          check-latest: true
+      - name: prepare anonymous docker config
+        run: mkdir -p "$DOCKER_CONFIG" && echo '{}' > "$DOCKER_CONFIG/config.json"
+      - name: e2e compatibility
+        run: make test-e2e E2E_GITEA_IMAGE=${{ matrix.gitea_image }}

.gitignore

@@ -2,6 +2,7 @@
 .env
 !/act/runner/testdata/secrets/.env
 .runner
+.runner.lock
 coverage.txt
 .tmp/
 /config.yaml

.golangci.yml

@@ -11,6 +11,7 @@ linters:
     - dupl
     - errcheck
     - forbidigo
+    - forcetypeassert
     - gocheckcompilerdirectives
     - gocritic
     - goheader
@@ -36,12 +37,8 @@ linters:
       rules:
         main:
           deny:
-            - pkg: io/ioutil
-              desc: use os or io instead
             - pkg: golang.org/x/exp
               desc: it's experimental and unreliable
-            - pkg: github.com/pkg/errors
-              desc: use builtin errors package instead
     nolintlint:
       allow-unused: false
       require-explanation: true
@@ -49,8 +46,7 @@ linters:
     gocritic:
       enabled-checks:
         - equalFold
-      disabled-checks:
+      disabled-checks: []
-        - ifElseChain
     revive:
       severity: error
       rules:
@@ -74,10 +70,14 @@ linters:
         - name: unexported-return
         - name: var-declaration
         - name: var-naming
+          arguments:
+            - [] # AllowList - do not remove as args for the rule are positional and won't work without lists first
+            - [] # DenyList
+            - - skip-initialism-name-checks: true
     staticcheck:
       checks:
         - all
-        - -ST1005
+    testifylint: {}
     usetesting:
       os-temp-dir: true
     perfsprint:
@@ -95,13 +95,14 @@ linters:
     generated: lax
     presets:
       - comments
-      - common-false-positives
-      - legacy
       - std-error-handling
     rules:
       - linters:
           - forbidigo
         path: cmd
+      - linters:
+          - forcetypeassert
+        path: _test\.go
 issues:
   max-issues-per-linter: 0
   max-same-issues: 0
@@ -118,7 +119,8 @@ formatters:
         - blank
         - default
     gofumpt:
-      extra-rules: true
+      extra:
+        group-params: true
   exclusions:
     generated: lax
 run:

.goreleaser.yaml

@@ -83,24 +83,12 @@ builds:
       - cmd: sh .goreleaser.checksum.sh {{ .Path }}
       - cmd: sh .goreleaser.checksum.sh {{ .Path }}.xz
 
-blobs:
+# Uploads every release artifact to Cloudflare R2. The `blobs:` pipe
-  -
+# isn't usable here since it authenticates from the global AWS_* env
-    provider: s3
+# with no per-entry credentials; `publishers:` supports per-entry
-    bucket: "{{ .Env.S3_BUCKET }}"
+# `env:` instead, so it's used to invoke scripts/upload-r2.sh once per
-    region: "{{ .Env.S3_REGION }}"
+# artifact. Custom publishers inherit almost nothing from the
-    directory: "gitea-runner/{{.Version}}"
+# environment, hence the explicit R2_* forwarding below.
-    extra_files:
-      - glob: ./**.xz
-      - glob: ./**.sha256
-
-# Mirrors the S3 `blobs:` upload above into Cloudflare R2 during the
-# parallel S3+R2 period (S3 will be removed once migration completes).
-# A second `blobs:` entry is impossible here since the blob pipe
-# authenticates from the global AWS_* env with no per-entry
-# credentials; `publishers:` supports per-entry `env:` instead, so
-# it's used to invoke scripts/upload-r2.sh once per artifact. Custom
-# publishers inherit almost nothing from the environment, hence the
-# explicit R2_* forwarding below.
 #
 # This publisher fires 109 times for 73 distinct keys because
 # goreleaser's release pipe already registers `release.extra_files`
@@ -125,7 +113,7 @@ publishers:
       - R2_SECRET_ACCESS_KEY={{ index .Env "R2_SECRET_ACCESS_KEY" }}
 
 archives:
-  - format: binary
+  - formats: [binary]
     name_template: "{{ .Binary }}"
     allow_different_binary_count: true
 

AGENTS.md

@@ -1,10 +1,19 @@
+- Never assume, verify before claiming
 - Use `make help` to find available development targets
-- Run `make fmt` to format `.go` files, and run `make lint-go` to lint them
+- PR descriptions: minimal, only what and why, no task lists or file listings
-- Run `make tidy` after any `go.mod` changes
+- Reference issues and PRs by full URL, not by number
-- Run single go unit tests with `go test -run '^TestName$' ./modulepath/`
+- Use Conventional Commits for commit messages and PR titles, plus the `enhance` type for user-facing enhancements
-- Add the current year into the copyright header of new `.go` files
+- Add an `Assisted-by: AGENT_NAME:MODEL_VERSION` trailer to commit messages, never `Co-Authored-By` or `Signed-off-by`
-- Ensure no trailing whitespace in edited files
+- Attribute agent authorship on one trailing line in issue and pull request comments, never as a PR description section
 - Never force-push, amend, or squash unless asked. Use new commits and normal push for pull request updates
-- Preserve existing code comments, do not remove or rewrite comments that are still relevant
+- Comments: write almost none, short and preferably same-line, explaining why for a future reader. Never narrate code, the change or the prompt. Preserve existing ones that still apply
-- Include authorship attribution in issue and pull request comments
+- Add the current year into the copyright header of new `.go` files
-- Add `Co-Authored-By` lines to all commits, indicating name and model used
+- Read `DEVELOPMENT.md` for internals and conventions
+- Ensure no trailing whitespace in edited files
+- Run `make fmt` after `.go` edits, `make tidy` after `go.mod` edits, and `make checks` for the non-lint source checks
+- Lint what changed with `make lint-go`, and `make lint-go-windows` for Windows and platform-split files
+- Fix the cause rather than disabling a linter or weakening a test. Where unavoidable, use the narrowest scope with a trailing comment giving the reason
+- Run single go tests with `go test -run '^TestName$' ./modulepath/`. `make test` self-skips the integration tests without docker or network, `make test-dind` runs the daemon-facing tests against the built dind image
+- Write the fewest, fastest tests covering the behavior, extending an existing one where possible. Prefer unit tests where logic is testable in isolation
+- Wait on a deterministic condition rather than `sleep`
+- Update the files under `docs/` when behavior documented there changes

CLAUDE.md

@@ -1 +0,0 @@
-@AGENTS.md

DEVELOPMENT.md

@@ -0,0 +1,46 @@
+# Development
+
+## Job log line format
+
+Gitea stores one log row per line and its web UI decodes the payload, so getting the encoding
+wrong never fails a test here, it only shows up in the browser.
+
+**A row cannot contain a real newline.** `FormatLog` rewrites `\n` to a literal backslash-n and
+truncates at 64 KiB on a byte boundary.
+
+**The payload of a line starting with a recognised prefix is decoded**, with the escape set
+depending on the prefix:
+
+| prefix | decodes |
+| --- | --- |
+| `##[error]` `##[warning]` `##[notice]` `##[debug]` `##[group]` `##[endgroup]` `##[add-matcher]` | `%25` `%0D` `%0A` `%3B` `%5D` |
+| `::error::` `::warning::` `::notice::` `::debug::` (with or without ` key=value` properties), `::group::` `::endgroup::` `::add-matcher::` | `%25` `%0D` `%0A` |
+| `##[command]` `[command]`, or no recognised prefix | nothing |
+
+### Rules
+
+- **Emitting a command line?** Escape the payload with `runner.EscapeCommandData`. One escaper
+  covers both forms: it escapes `%` first, so a literal `%3B` becomes `%253B` that the extra
+  `##[…]` rules cannot match, and a raw `;` or `]` is never decoded. It is also what makes
+  multi-line work, `\n` becomes `%0A` and the UI turns it back into a line break.
+- **Forwarding a command from step output?** Leave the payload alone, it arrived escaped and is
+  decoded once. Decoding here double-decodes and destroys multi-line.
+- **No prefix?** Do not escape, and split multi-line values into one row each.
+- **Interpolating a secret?** Masking runs after escaping, so `AppendSecretMasker` registers the
+  encoded forms too.
+- Command *properties* also escape `%3A` and `%2C`, which the UI never decodes, so the reporter
+  decodes exactly those two when folding a location into an annotation.
+
+## Built-in actions
+
+Each lives in `internal/pkg/action/<name>` and is registered in `builtinActions` in
+`act/runner/step_action_builtin.go`.
+
+## End-to-end compatibility tests
+
+`make test-e2e` runs the runner against `E2E_GITEA_IMAGE`. It defaults to the nightly image.
+It requires Docker and is excluded from `make test`. CI runs stable and nightly variants in
+parallel.
+
+The suite shares one Gitea and regular runner. Cache and ephemeral scenarios use isolated
+repository runners. A run path is `<workflow>@<ref>` and a log row is `<timestamp>Z <payload>`.

Dockerfile

@@ -1,7 +1,7 @@
 ### BUILDER STAGE
 #
 #
-FROM golang:1.26-alpine3.23 AS builder
+FROM golang:1.27-alpine3.23 AS builder
 
 # Do not remove `git` here, it is required for getting runner version when executing `make build`
 RUN apk add --no-cache make git
@@ -17,7 +17,7 @@ RUN make clean && make build
 ### DIND VARIANT
 #
 #
-FROM docker:29.6.2-dind AS dind
+FROM docker:29.7.2-dind AS dind
 
 ARG VERSION=dev
 
@@ -32,12 +32,14 @@ COPY scripts/s6 /etc/s6
 
 VOLUME /data
 
+ENV TINI_SUBREAPER=1
+
 ENTRYPOINT ["s6-svscan","/etc/s6"]
 
 ### DIND-ROOTLESS VARIANT
 #
 #
-FROM docker:29.6.2-dind-rootless AS dind-rootless
+FROM docker:29.7.2-dind-rootless AS dind-rootless
 
 ARG VERSION=dev
 
@@ -56,6 +58,7 @@ VOLUME /data
 RUN mkdir -p /data && chown -R rootless:rootless /etc/s6 /data
 
 ENV DOCKER_HOST=unix:///run/user/1000/docker.sock
+ENV TINI_SUBREAPER=1
 
 USER rootless
 ENTRYPOINT ["s6-svscan","/etc/s6"]

Makefile

@@ -4,9 +4,15 @@ DIST_DIRS := $(DIST)/binaries $(DIST)/release
 GO ?= go
 SHASUM ?= shasum -a 256
 HAS_GO = $(shell hash $(GO) > /dev/null 2>&1 && echo "GO" || echo "NOGO" )
-XGO_PACKAGE ?= src.techknowlogick.com/xgo@latest
+XGO_PACKAGE ?= src.techknowlogick.com/xgo@v1.9.0 # renovate: datasource=go
-XGO_VERSION := go-1.26.x
+XGO_VERSION := go-1.27.x
-GXZ_PACKAGE ?= github.com/ulikunitz/xz/cmd/[email protected]0
+GXZ_PACKAGE ?= github.com/ulikunitz/xz/cmd/[email protected]7 # renovate: datasource=go
+
+PODMAN_TEST_IMAGE ?= quay.io/containers/podman:v5.8.4-immutable@sha256:90405e3c6a86f33c5a6f8dbdfa8e02a98143a8ba4337b84bdaec5336d2ef008a # renovate: datasource=docker
+E2E_JOB_IMAGE ?= docker.gitea.com/runner-images:ubuntu-latest@sha256:0e62e56b382ebf485bff1e51a05cbe27c708545a2e91425467fd31eb3e249521 # renovate: datasource=docker
+SERVICE_IMAGE ?= nginx:1.31.4-alpine@sha256:db35bfc6b2951e7f8a72db5db120288c127ffaeeb4a6d4b95a26fead017d5913 # renovate: datasource=docker
+E2E_GITEA_IMAGE ?= gitea/gitea:main-nightly
+E2E_K3S_IMAGE ?= rancher/k3s:v1.36.4-k3s1@sha256:edad48e12bf81c3a09ac1c05c0c0ffaaa22145980b989d6fae84543a76b83657 # renovate: datasource=docker
 
 LINUX_ARCHS ?= linux/amd64,linux/arm64
 DARWIN_ARCHS ?= darwin-12/amd64,darwin-12/arm64
@@ -18,8 +24,10 @@ DOCKER_TAG ?= nightly
 DOCKER_REF := $(DOCKER_IMAGE):$(DOCKER_TAG)
 DOCKER_ROOTLESS_REF := $(DOCKER_IMAGE):$(DOCKER_TAG)-dind-rootless
 
-GOLANGCI_LINT_PACKAGE ?= github.com/golangci/golangci-lint/v2/cmd/[email protected]2.2
+GOLANGCI_LINT_PACKAGE ?= github.com/golangci/golangci-lint/v2/cmd/[email protected]4.0 # renovate: datasource=go
-GOVULNCHECK_PACKAGE ?= golang.org/x/vuln/cmd/govulncheck@v1.3.0
+GOVULNCHECK_PACKAGE ?= golang.org/x/vuln/cmd/govulncheck@v1.8.0 # renovate: datasource=go
+
+GOTEST_FLAGS ?= -race -timeout 20m -parallel 8
 
 STATIC ?=
 EXTLDFLAGS ?=
@@ -94,13 +102,8 @@ go-check:
 	fi
 
 .PHONY: fmt-check
-fmt-check: fmt
+fmt-check: ## check the Go code formatting
-	@diff=$$(git diff --color=always); \
+	$(GO) run $(GOLANGCI_LINT_PACKAGE) fmt --diff-colored
-	if [ -n "$$diff" ]; then \
-		echo "Please run 'make fmt' and commit the result:"; \
-		printf "%s" "$${diff}"; \
-		exit 1; \
-	fi
 
 .PHONY: deps-tools
 deps-tools: ## install tool dependencies
@@ -110,6 +113,9 @@ deps-tools: ## install tool dependencies
 	$(GO) install $(GOVULNCHECK_PACKAGE) & \
 	wait
 
+.PHONY: checks
+checks: tidy-check fmt-check security-check ## run the non-lint source checks
+
 .PHONY: lint
 lint: lint-go lint-go-windows ## lint everything
 
@@ -131,25 +137,25 @@ lint-pr-title: ## lint PR title against Conventional Commits (set PR_TITLE=...)
 	@node ./tools/lint-pr-title.ts
 
 .PHONY: security-check
-security-check: deps-tools
+security-check:
-	GOEXPERIMENT= $(GO) run $(GOVULNCHECK_PACKAGE) -show color ./... || true
+	$(GO) run $(GOVULNCHECK_PACKAGE) -show color ./... || true
 
 .PHONY: tidy
 tidy: ## run go mod tidy
+	$(eval GO_TOOLCHAIN := $(shell grep -Eo '^toolchain\s+go[0-9.]+' go.mod | cut -d' ' -f2))
 	$(GO) mod tidy
-
+	@# workaround https://github.com/golang/go/issues/75331: restore toolchain if tidy dropped it
-.PHONY: tidy-check
+	@if [ -n "$(GO_TOOLCHAIN)" ] && ! grep -qE '^toolchain\s' go.mod; then \
-tidy-check: tidy
+		$(GO) mod edit -toolchain=$(GO_TOOLCHAIN); \
-	@diff=$$(git diff --color=always -- go.mod go.sum); \
-	if [ -n "$$diff" ]; then \
-		echo "Please run 'make tidy' and commit the result:"; \
-		printf "%s" "$${diff}"; \
-		exit 1; \
 	fi
 
+.PHONY: tidy-check
+tidy-check: ## check that go.mod and go.sum are tidy
+	$(GO) mod tidy -diff
+
 .PHONY: test
-test: fmt-check security-check ## test everything (integration tests self-skip without docker/network)
+test: ## test everything (integration tests self-skip without docker/network)
-	@$(GO) test -race -timeout 20m -v -cover -coverprofile coverage.txt ./... && echo "\n==>\033[32m Ok\033[m\n" || exit 1
+	@$(GO) test $(GOTEST_FLAGS) -cover -coverprofile coverage.txt ./... && echo "\n==>\033[32m Ok\033[m\n" || exit 1
 
 .PHONY: coverage-report
 coverage-report: ## turn coverage.txt from `make test` into .tmp/coverage.md
@@ -158,8 +164,14 @@ coverage-report: ## turn coverage.txt from `make test` into .tmp/coverage.md
 	@echo "Wrote .tmp/coverage.md"
 
 .PHONY: test-dind
-test-dind: ## run the daemon-facing tests against the built dind image (TARGET=dind|dind-rootless)
+test-dind: ## run the daemon-facing tests against the built dind image or podman (TARGET=dind|dind-rootless|podman)
-	@./scripts/test-dind.sh $(TARGET)
+	@PODMAN_TEST_IMAGE=$(PODMAN_TEST_IMAGE) ./scripts/test-dind.sh $(TARGET)
+
+E2E_CONCURRENCY ?= 8
+
+.PHONY: test-e2e
+test-e2e: ## run Gitea compatibility tests against E2E_GITEA_IMAGE
+	@E2E_CONCURRENCY=$(E2E_CONCURRENCY) E2E_GITEA_IMAGE=$(E2E_GITEA_IMAGE) E2E_K3S_IMAGE=$(E2E_K3S_IMAGE) E2E_JOB_IMAGE=$(E2E_JOB_IMAGE) GO=$(GO) SERVICE_IMAGE=$(SERVICE_IMAGE) ./tools/test-e2e.sh
 
 .PHONY: install
 install: $(GOFILES) ## install the runner binary via `go install`

README.md

@@ -4,7 +4,19 @@
 
 ### Prerequisites
 
-Docker Engine Community version is required for docker mode. To install Docker CE, follow the official [install instructions](https://docs.docker.com/engine/install/).
+On the runner:
+
+- `git` 2.34.1 or newer on `PATH`, to download actions and reusable workflows.
+- `docker` or `podman` with `podman.socket` enabled, for docker mode, Docker actions and `docker://` steps. To install Docker CE, follow the official [install instructions](https://docs.docker.com/engine/install/).
+
+Where steps run, in the job image or on the runner in host mode:
+
+- `bash` or `sh` for `run` steps, `pwsh` or `powershell` on a Windows host.
+- `node` for JavaScript actions and `hashFiles()`.
+- `git` 2.34.1 or newer for [`builtin:checkout`](docs/builtin-actions.md).
+- `git-lfs` and `ssh` for `builtin:checkout`, only when its `lfs` or `ssh-key` input is used.
+
+Kubernetes mode needs [a few more tools](docs/kubernetes.md) in job images.
 
 ### Download pre-built binary
 
@@ -79,77 +91,85 @@ If the registry succeed, it will run immediately. Next time, you could run the r
 
 ### Run with docker
 
-```bash
+The recommended Docker-in-Docker image runs its own Docker daemon for jobs. Jobs never get the host's Docker socket, which is equivalent to root access on the host.
-docker run -e GITEA_INSTANCE_URL=https://your_gitea.com -e GITEA_RUNNER_REGISTRATION_TOKEN=<your_token> -v /var/run/docker.sock:/var/run/docker.sock --name my_runner gitea/runner:nightly
-```
-
-Mount a volume on `/data` if you want the registration file and optional config to survive container recreation (see [scripts/run.sh](scripts/run.sh)).
-
-> **`/data` does not hold the image cache.** It is the runner's working directory and contains only the `.runner` registration file and, optionally, your config file. Images pulled for jobs live in the *Docker daemon's* data root, which for the `dind` flavours is inside the container (`/var/lib/docker`, or `/home/rootless/.local/share/docker` for `dind-rootless`). To keep the image cache across restarts, give that path its own volume as well — otherwise every new container re-pulls the job images. With the `basic` flavour the images live on whichever daemon you point the runner at, so there is nothing extra to persist.
-
-### Image flavours
-
-The image is published in three flavours, all built from the single multi-stage [Dockerfile](Dockerfile) in this repository. They differ only in how a Docker daemon is made available to the jobs the runner executes; the `gitea-runner` binary inside them is identical.
-
-| Tag | Build target | Base image | Docker daemon | Process supervisor | Runs as |
-| --- | --- | --- | --- | --- | --- |
-| `latest` (and `<version>`) | `basic` | `alpine` | none — uses an external daemon you provide | [`tini`](https://github.com/krallin/tini) | `root` |
-| `latest-dind` | `dind` | `docker:dind` | bundled, started inside the container | [`s6`](https://skarnet.org/software/s6/) | `root` (privileged) |
-| `latest-dind-rootless` | `dind-rootless` | `docker:dind-rootless` | bundled, started rootless inside the container | [`s6`](https://skarnet.org/software/s6/) | `rootless` (UID 1000) |
-
-#### `latest` — basic
-
-The default flavour ships only the runner on a minimal Alpine base. It contains **no Docker daemon of its own**: jobs that use `docker://` images need a daemon supplied from outside the container, typically by bind-mounting the host's socket:
-
-```bash
-docker run -e GITEA_INSTANCE_URL=https://your_gitea.com -e GITEA_RUNNER_REGISTRATION_TOKEN=<your_token> \
-  -v /var/run/docker.sock:/var/run/docker.sock --name my_runner gitea/runner:latest
-```
-
-`tini` is the entrypoint (it reaps zombie processes), and it just runs [`scripts/run.sh`](scripts/run.sh), which registers the runner on first start and then execs `gitea-runner daemon`. This flavour does not need `--privileged`. The trade-off is that jobs share the host's daemon, so they can see other containers and images on that daemon.
-
-#### `latest-dind` — Docker-in-Docker
-
-This flavour is based on the official `docker:dind` image and bundles its own Docker daemon, so it needs no external socket — only the `--privileged` flag that Docker-in-Docker requires:
 
 ```bash
 docker run --privileged -e GITEA_INSTANCE_URL=https://your_gitea.com -e GITEA_RUNNER_REGISTRATION_TOKEN=<your_token> \
-  --name my_runner gitea/runner:latest-dind
+  -v "$PWD/data:/data" -v runner-docker:/var/lib/docker --name my_runner gitea/runner:nightly-dind
 ```
 
-Two processes have to run side by side here (the Docker daemon and the runner), so the entrypoint is the [`s6`](https://skarnet.org/software/s6/) supervision tree under [`scripts/s6`](scripts/s6) instead of `tini`. `s6` starts `dockerd`, and the runner service waits for the daemon to come up (`s6-svwait`) before launching [`run.sh`](scripts/run.sh). Each container has a private daemon isolated from the host's, at the cost of running privileged.
+The volumes keep the runner's registration and the job image cache across container recreation.
 
-#### `latest-dind-rootless` — rootless Docker-in-Docker
+### Image flavours
 
-Same idea as `dind`, but built on `docker:dind-rootless` so the bundled daemon and the runner run as an unprivileged user (`rootless`, UID 1000) rather than `root`. `DOCKER_HOST` is preset to `unix:///run/user/1000/docker.sock` so the runner talks to the rootless daemon. This reduces the blast radius compared to the privileged `dind` flavour, but rootless Docker carries the usual rootless limitations (networking, cgroups, storage drivers, and some operations that need additional host configuration such as `/etc/subuid` / `/etc/subgid` mappings and unprivileged user-namespace support).
+| Tag | Docker daemon for jobs |
+| --- | --- |
+| `latest-dind` (recommended) | Bundled, needs `--privileged`. |
+| `latest-dind-rootless` | Bundled, runs as a non-root user (UID 1000) for stronger isolation. Needs `--privileged` and [rootless Docker](https://docs.docker.com/engine/security/rootless/) support on the host. Keeps its image cache in `/home/rootless/.local/share/docker`. |
+| `latest` | None. Mount the host's socket with `-v /var/run/docker.sock:/var/run/docker.sock`, or set `DOCKER_HOST`. |
 
-> **The UID is fixed at 1000.** It comes from the `rootless` user baked into the upstream `docker:dind-rootless` base image, and the bundled daemon always listens on `/run/user/1000/docker.sock` inside the container, so running this flavour as a different user (`--user 1001`) does not work. If you need the runner to talk to a *host* rootless daemon that runs under some other UID, use the `basic` flavour instead and bind-mount that daemon's socket (see [examples/vm/rootless-docker.md](examples/vm/rootless-docker.md)); pointing `DOCKER_HOST` at a host socket from inside `dind-rootless` will not work. Changing the UID otherwise means rebuilding the image from a base with a different `rootless` user.
+Versioned and `nightly` tags carry the same suffixes, for example `nightly-dind`. The images target Docker, not Podman.
-
-> **Note on Podman:** these images target the Docker daemon. The bundled `dind`/`dind-rootless` daemons are `dockerd`, not Podman, and the `basic` flavour expects a Docker-compatible socket. Running them under rootless Podman is not a supported configuration, though pointing the `basic` flavour at a Podman socket that emulates the Docker API may work for some workloads.
 
 ### Configuration
 
-The runner is configured with a YAML file. Generate a starting point (this matches what ships in the tree):
+The runner reads a YAML file. Without one, every option keeps its default.
 
 ```bash
-./gitea-runner generate-config > config.yaml
+./gitea-runner config init             # write config.yaml, with no option set
+./gitea-runner config generate | less  # read what the options do
+./gitea-runner -c config.yaml daemon   # -c also works on register and cache-server
 ```
 
-Pass it with `-c` / `--config` on any command that loads configuration (`register`, `daemon`, `cache-server`):
+`config generate` prints [config.example.yaml](internal/pkg/config/config.example.yaml). Every value in it is commented out, so copy the lines you want to change into your own file and uncomment them.
+
+#### Editing a config file
+
+`config` edits a file in place, which is handy in provisioning scripts:
 
 ```bash
-./gitea-runner -c config.yaml register
+./gitea-runner config set runner.capacity 4
-./gitea-runner -c config.yaml daemon
+./gitea-runner config set runner.timeout 90m  # written as 1h30m0s
-./gitea-runner -c config.yaml cache-server
+./gitea-runner config set runner.envs.MY_VAR value
+./gitea-runner config add runner.labels 'ubuntu:docker://node:22'
+./gitea-runner config remove runner.labels 'ubuntu:docker://node:22'
+./gitea-runner config get runner.labels
 ```
 
-Every option is described in [config.example.yaml](internal/pkg/config/config.example.yaml) (the same content `generate-config` prints).
+A key is its dotted YAML path. An unknown key, a value of the wrong type, or `add`/`remove` on anything but a list is refused before the file is touched. `set` replaces a whole list when you give it several values.
 
-#### Without a config file
+An edit keeps the comments and the key order of the file. Indentation becomes two spaces, and a blank line between two values is dropped.
 
-If you omit `-c`, built-in defaults apply (same as an empty YAML document).
+`config get`, `set`, `add` and `remove` use `config.yaml` (or `config.yml`) from the working directory, then from the directory of the binary, and print their choice to stderr. `config init` writes `config.yaml` in the working directory, and refuses to overwrite an existing config without `--force`. Pass `-c` for another path.
 
-Earlier releases let a small set of environment variables (`GITEA_DEBUG`, `GITEA_TRACE`, `GITEA_RUNNER_CAPACITY`, `GITEA_RUNNER_FILE`, `GITEA_RUNNER_ENVIRON`, `GITEA_RUNNER_ENV_FILE`) override parts of the default config. Those overrides have been removed — use a YAML config file for all settings instead. For the Docker images, the entrypoint still understands a separate set of variables (such as `RUNNER_STATE_FILE`); see [scripts/run.sh](scripts/run.sh) and the container documentation below.
+#### Tool cache
+
+Setup actions like `setup-go` install tools into `RUNNER_TOOL_CACHE`, which is `/opt/hostedtoolcache` inside a job. `runner.tool_cache_mode` selects what backs it:
+
+| Mode | Tool cache | Trade-off |
+| --- | --- | --- |
+| `none` (default) | Per job, provided by the job image | A version the image lacks is downloaded in every job |
+| `shared` | One volume reused by every job | Two jobs writing the same tool version at once corrupt it, so use it only with `runner.capacity: 1` |
+
+With `none`, tools must come from the job image. Install them into `/opt/hostedtoolcache/<tool>/<version>/<arch>`, with an empty `<arch>.complete` file next to the directory:
+
+```dockerfile
+RUN GO=$(curl -fsSL 'https://go.dev/dl/?mode=json' | grep -oP '"version": "\Kgo1\.26\.[0-9]*' | head -1); \
+  DIR="/opt/hostedtoolcache/go/${GO#go}/x64" && \
+  mkdir -p "$(dirname "$DIR")" && \
+  curl -fsSL "https://dl.google.com/go/${GO}.linux-amd64.tar.gz" | tar -xz -C /tmp && \
+  mv /tmp/go "$DIR" && \
+  touch "${DIR}.complete"
+```
+
+A workflow requesting a minor version, `go-version: "1.26"`, resolves to the newest matching version in the cache, so a patch update in the image still hits it.
+
+Of the [runner images](https://gitea.com/gitea/runner-images), the `-full` flavour is the one that ships tools in this layout.
+
+`gitea-runner exec` reads no config file and takes `--tool-cache-mode` instead, defaulting to `none`.
+
+#### Environment variables
+
+Earlier releases let a few environment variables (`GITEA_DEBUG`, `GITEA_TRACE`, `GITEA_RUNNER_CAPACITY`, `GITEA_RUNNER_FILE`, `GITEA_RUNNER_ENVIRON`, `GITEA_RUNNER_ENV_FILE`) override parts of the config. They are gone, use the YAML file for all settings. Config values can read the runner's environment as `${NAME}`, for example `capacity: ${RUNNER_CAPACITY}`. An unset `NAME` fails the load, and `$${NAME}` is a literal `${NAME}`. The Docker images still read their own variables, such as `RUNNER_STATE_FILE`, see [scripts/run.sh](scripts/run.sh) and the container documentation below.
 
 ### Labels
 
@@ -164,10 +184,10 @@ A label is written as:
 | Part | Meaning |
 | --- | --- |
 | `name` | The name a workflow refers to in `runs-on`, e.g. `ubuntu-latest`. |
-| `schema` | Either `docker` or `host`. Defaults to `host` when omitted. |
+| `schema` | `docker`, `kubernetes` or `host`. Defaults to `host` when omitted. |
-| `args` | Only used by the `docker` schema: the image to run the job in. |
+| `args` | Only used by the `docker` and `kubernetes` schemas: the image to run the job in. |
 
-Two schemas are supported:
+Three schemas are supported:
 
 - **`docker://<image>`** — the job runs inside a container created from `<image>`:
 
@@ -175,6 +195,12 @@ Two schemas are supported:
   ubuntu-latest:docker://docker.gitea.com/runner-images:ubuntu-latest
   ```
 
+- **`kubernetes://<image>`** — the job runs as a Kubernetes pod created from `<image>`, see **[docs/kubernetes.md](docs/kubernetes.md)**:
+
+  ```text
+  ubuntu-latest:kubernetes://docker.gitea.com/runner-images:ubuntu-latest
+  ```
+
 - **`host`** — the job's steps run directly on the machine the runner is on, using the tools installed there:
 
   ```text
@@ -189,9 +215,9 @@ ubuntu-latest:docker://docker.gitea.com/runner-images:ubuntu-latest,macos:host
 
 a workflow with `runs-on: ubuntu-latest` is executed in the `runner-images:ubuntu-latest` container, and one with `runs-on: macos` is executed directly on the host.
 
-Names may themselves contain a colon (for example `pool:e57e18d4-10d4-406f-93bf-60f127221bdd`); only `host` and `docker` are treated as schemas.
+Names may themselves contain a colon, for example `pool:e57e18d4-10d4-406f-93bf-60f127221bdd`. Only `host`, `docker` and `kubernetes` are treated as schemas.
 
-If a job's `runs-on` matches none of the runner's labels, the job still runs, in the default `docker.gitea.com/runner-images:ubuntu-latest` image. Images maintained for this purpose are listed at [gitea/runner-images](https://gitea.com/gitea/runner-images).
+Jobs run on `runner.default_image` when no runner label matches their `runs-on`: as a pod on runners with `kubernetes` labels, otherwise in a container, or on the host when docker is unavailable. Suitable images are listed at [gitea/runner-images](https://gitea.com/gitea/runner-images).
 
 Labels are chosen at registration time (`--labels`, or the interactive prompt) and can be changed afterwards by editing `runner.labels` in the config file, or in the Gitea UI under the runner's settings.
 
@@ -209,44 +235,128 @@ Whenever the resulting labels differ from the ones in the registration file, the
 
 > **Note:** A runner that only exposes `host` labels still needs access to a Docker daemon (e.g. a mounted `/var/run/docker.sock`) whenever a job uses a `docker://` action or a service container. `host` labels only change where the job's own steps run; container-based steps and actions are still executed with Docker.
 
+#### Service containers
+
+A job's `services` are started before its steps run. When a service's image or its `options` declare a healthcheck, the runner waits for it to report healthy, so a workflow does not have to poll for its own services:
+
+```yaml
+services:
+  postgres:
+    image: postgres:17
+    options: >-
+      --health-cmd pg_isready
+      --health-interval 5s
+      --health-retries 10
+```
+
+A service that reports unhealthy fails the job right away, with its container log. One that never becomes healthy fails it after `container.service_ready_timeout` (default `5m`, negative disables the wait). A service that exits without declaring a healthcheck only gets its log and a warning.
+
+A job in a container reaches a service by its id on the job network, on the port the service listens on, for example `psql -h postgres -p 5432`. The started containers also fill the `job` context: `job.container.{id,network}` and `job.services.<id>.{id,network,ports}`, where `ports` maps a container port to the host port Docker published it on, for the services that publish one.
+
+Unlike GitHub, a job whose steps run on the host (a `host` label without `container:`) starts no service containers, so `job.services` and `job.container` stay empty. Give such a job a `container:` when it needs services.
+
+#### Docker from a job (`GITEA_DOCKER_WORKSPACE`)
+
+Containers a job starts through the Docker socket can bind-mount the workspace by the job's own path, as on a host, for example `docker run -v "$PWD:/src"`. Without the Docker proxy, for example with a remote `DOCKER_HOST`, use `GITEA_DOCKER_WORKSPACE`, the path the daemon sees, as the prefix of workspace binds, with `.` as the fallback for local use:
+
+```yaml
+volumes:
+  - ${GITEA_DOCKER_WORKSPACE:-.}/data:/app/data
+```
+
+Linux container jobs use a Docker proxy when socket sharing and permissions allow it. Other setups use the daemon socket directly. Host jobs use their existing Docker access, where `.` works. The proxy removes the containers, networks and volumes a job creates, `docker compose` services included, after post steps and the completed hook. Set `container.sweep: false` when jobs deploy services to the runner's Docker daemon.
+
+#### Proxy
+
+Set these variables in the runner's environment, with systemd `Environment=`, `docker run -e`, or Kubernetes `env:`:
+
+```sh
+http_proxy=http://proxy.example:3128
+https_proxy=http://proxy.example:3128
+no_proxy=gitea.internal,.example.local
+```
+
+The runner uses them for its own requests and gives them to every job, in lower and upper case.
+
+These hosts are added to `no_proxy` for jobs, so they are always reached directly:
+
+- the built-in cache server, whose address is assigned at startup
+- `localhost`, `127.0.0.1` and `::1`
+- the job's service containers
+- the Docker daemon, when it is reached over `tcp://`
+
+Gitea is not added. Add it to `no_proxy` yourself if it should be reached directly.
+
+For one step, set it in the step's `env:`, for the rest of a job, write it to `$GITHUB_ENV`. Setting it at workflow or job level or in `container.env` has no effect. To change it for the whole runner, set it in `runner.envs`. A `no_proxy` set there is added to the list above instead of replacing it.
+
+Images are pulled by the Docker daemon, which needs its own proxy setting. In the `dind` images the daemon runs in the same container and reads the variables above. For any other daemon, see [the Docker documentation](https://docs.docker.com/engine/daemon/proxy/). The runner logs a warning at startup if it has a proxy and the daemon does not.
+
+Dockerfile actions are built with these variables as build arguments, so their `RUN` steps can reach the network.
+
+A password in a proxy URL is hidden in job logs. Any step can still read it, because the step is given the proxy URL in its environment.
+
+#### Built-in actions (`uses: builtin:<name>`)
+
+Use `uses: builtin:checkout` to check out a repository without downloading `actions/checkout` or needing Node in the job image, see **[docs/builtin-actions.md](docs/builtin-actions.md)**.
+
 #### Caching (`actions/cache`)
 
-Each runner starts its own cache server automatically. Cache entries are local to that runner — runners do not share a cache by default.
+Each runner starts its own cache server, so runners do not share cached entries. A runner in a container on the jobs' Docker daemon attaches itself to each job network, so jobs reach its cache without extra setup.
 
-**Shared cache across multiple runners**
+**Sharing a cache between runners**
 
-Run one dedicated `gitea-runner cache-server` that all runners point at.
+Run one `gitea-runner cache-server` and point every runner at it, using the same secret everywhere, for example from `openssl rand -hex 32`:
 
-1. Create a config file for the cache server host:
+```yaml
-
+# cache-server.yaml
-   ```yaml
+cache:
-   cache:
   dir: /data/actcache
   port: 8088
-     external_secret: "replace-with-a-strong-random-secret"
+  external_secret: "<secret>" # or external_secret_file: /path/to/secret
-     # external_secret_file: /path/to/secret # secret can also be passed via a file
+```
-   ```
 
-2. Start the server:
+```bash
+gitea-runner -c cache-server.yaml cache-server
+```
 
-   ```bash
+```yaml
-   gitea-runner -c cache-server-config.yaml cache-server
+# each runner's config
-   ```
+cache:
-
-3. On every runner:
-
-   ```yaml
-   cache:
   external_server: "http://<cache-server-host>:8088/"
-     external_secret: "replace-with-a-strong-random-secret"  # must match the server
+  external_secret: "<secret>"
-     # external_secret_file: /path/to/secret # secret can also be passed via a file
+```
-   ```
 
-Alternatively, mount the same NFS/CIFS share on every runner and point `cache.dir` at it — simpler, but with weaker isolation between repositories.
+`--dir`, `--host` and `--port` override the matching `cache` keys. Eviction settings take effect on the cache server, not on the runners.
 
-**S3 / MinIO** — mount object storage as a FUSE filesystem (e.g. [s3fs](https://github.com/s3fs-fuse/s3fs-fuse) or [goofys](https://github.com/kahing/goofys)) and set `cache.dir` to the mount point.
+Runners can also share one `cache.dir` on a file system with working file locks, at the cost of slower cache requests.
 
-Flags `--dir`, `--host`, and `--port` on `cache-server` override the corresponding `cache.*` YAML keys; all other settings, including `external_secret`, require the config file.
+**Storing the cache in an S3-compatible bucket**
+
+The cache can also be stored in an S3-compatible bucket such as AWS S3, MinIO or RustFS, with no cache server. Runners with the same settings use the same cache:
+
+```yaml
+cache:
+  s3:
+    endpoint: "http://minio:9000"
+    region: "us-east-1"
+    bucket: "runner-cache"
+    prefix: "gitea-runner-cache"
+    path_style: true
+    access_key_file: "/run/secrets/cache-s3-access-key"
+    secret_key_file: "/run/secrets/cache-s3-secret-key"
+```
+
+A cache saved on one runner can be restored on the others within 30 seconds. The eviction settings below do not apply to the bucket, so add lifecycle rules to it that expire old caches and abort incomplete multipart uploads.
+
+**Eviction**
+
+Entries not read or written within `retention` (default `168h`) are removed. A repository over `repo_size_limit` (default `10GB`) loses its least recently used entries, and `size_limit` (off by default) caps the whole cache the same way. Entries in use are never removed. The cache also keeps 1024 MiB free on its volume, or `health_check.min_free_disk_space_mb` when health checks are enabled. See [config.example.yaml](internal/pkg/config/config.example.yaml) for all options.
+
+**Cache service v2**
+
+`actions/cache` v3.4.0, v4.2.0 and later use the cache service v2 API, which the runner serves by default. These actions fall back to v1 on hosts they do not recognize as GitHub, so the runner removes that check from them while a job runs. This also lets the stock `actions/upload-artifact` v4.4.0 and `actions/download-artifact` v4.1.8 and later work without the `gitea-upload-artifact` fork. With `runner.patch_actions: false`, the cache stays on v1 and the artifact actions fail.
+
+With v2, artifact calls also go through the cache server, so jobs must be able to reach it. `cache.v2: false` sends them to Gitea directly, unless the Gitea URL has a path or `runner.insecure` is set with HTTPS.
 
 #### Official Docker image
 
@@ -254,11 +364,13 @@ Besides `GITEA_INSTANCE_URL` and `GITEA_RUNNER_REGISTRATION_TOKEN`, the image en
 
 For a fuller container-oriented walkthrough, see [examples/docker](examples/docker/README.md).
 
-When `container.bind_workdir` is enabled, stale task workspace directories can be cleaned while the runner is idle:
+While the runner is idle it cleans up after earlier jobs:
-- directories older than `runner.workdir_cleanup_age` are removed (default: `24h`; set `0` to disable)
+- when `container.bind_workdir` is enabled, stale task workspace directories older than `runner.workdir_cleanup_age` are removed (default: `24h`; set `0` to disable)
-- cleanup runs every `runner.idle_cleanup_interval` (default: `10m`; set `0` to disable)
 - only purely numeric subdirectories under `container.workdir_parent` are treated as task workspaces and may be removed
 - cleanup assumes `container.workdir_parent` is not shared across multiple runners
+- on runners that use docker, per-job networks left behind by jobs the runner did not live to tear down are removed, identified by the `com.gitea.runner.uuid` label carrying this runner's uuid
+- on runners with `kubernetes` labels, job pods with that label older than `runner.workdir_cleanup_age` are removed the same way, together with their Secrets
+- cleanup runs every `runner.idle_cleanup_interval` (default: `10m`; set `0` to disable), and setting either knob to `0` disables all of the above
 
 #### Post-task script (`runner.post_task_script`)
 
@@ -270,6 +382,30 @@ On Windows, use `.exe`, `.bat`, or `.cmd` paths; **PowerShell (`.ps1`) is not su
 
 See **[docs/post-task-script.md](docs/post-task-script.md)** for lifecycle details, environment variables, timeout interaction, and platform notes.
 
+#### Job hooks (`runner.hooks.job_started`, `runner.hooks.job_completed`)
+
+Optional scripts that run **inside the job environment** (the job container, or the host in host mode), before the job's first step and after its last one. They are the equivalent of GitHub's `ACTIONS_RUNNER_HOOK_JOB_STARTED` / `ACTIONS_RUNNER_HOOK_JOB_COMPLETED`, which are read when the settings are unset.
+
+Because they run where the steps run and see the job's environment, they are the place for per-job setup no workflow should have to carry: registry logins, mirror configuration, or masking runner-wide secrets with `::add-mask::`. Their output is part of the job log and is scanned for workflow commands, and they can export to the job through `$GITHUB_ENV` and `$GITHUB_PATH`.
+
+Both hooks are synchronous and block the job while they run. Either one exiting non-zero fails the job, and there is no per-hook timeout.
+
+See **[docs/job-hooks.md](docs/job-hooks.md)** for the execution order, environment, and platform notes.
+
+#### OpenTelemetry (`OTEL_*`)
+
+Set `OTEL_EXPORTER_OTLP_ENDPOINT` in the runner's environment to send a trace of every job, and the runner's metrics, to an OpenTelemetry collector. See **[docs/telemetry.md](docs/telemetry.md)**.
+
+#### Local job logs (`log.job.dir`)
+
+Set `log.job.dir` to a path and the runner writes a copy of every task's log there as `<start time>-task-<id>.log`: the rows exactly as Gitea received them, with the same secrets masked and the job's result on the last line. Off by default, and what Gitea shows does not change.
+
+#### Secret masking
+
+A job's secrets and its `::add-mask::` values are hidden from what the runner writes and uploads: the job log, the local copy above, job summaries, and the names of the containers it creates. A job output carrying one is skipped with a warning rather than sent masked, as GitHub does, so a downstream `needs.<job>.outputs.<name>` reading it is empty.
+
+`log.job.retention` (default `168h`) is how long a log is kept, expired ones being deleted as new tasks start, and `log.job.max_size` (default `1GB`) caps one log. Keep `retention` above `runner.timeout` so a long job cannot outlive its own log, and prefer local disk, the file is written while the job runs. Only the runner's own user can read it.
+
 ### Example Deployments
 
 Check out the [examples](examples) directory for sample deployment types.

act/artifactcache/handler_v2.go

@@ -0,0 +1,334 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"cmp"
+	"encoding/json/jsontext"
+	"encoding/json/v2"
+	"encoding/xml"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/julienschmidt/httprouter"
+)
+
+// The cache service v2 API. A client on this version talks twirp to
+// `github.actions.results.api.v1.CacheService` instead of the /_apis/artifactcache
+// endpoints, and uploads the archive to the returned URL with the Azure blob protocol.
+// Both API versions are served from the same store, so a repository keeps its cache
+// when a workflow moves between action versions.
+//
+// Responses carry the proto field names, which is what Gitea's own results API emits and the only
+// spelling the Go clients parse. The JavaScript toolkit accepts either.
+const (
+	cacheServiceV2Path = "/twirp/github.actions.results.api.v1.CacheService"
+
+	// blobPath authenticates by signature, because the client uploads without an
+	// Authorization header. Downloads are handed the v1 artifact URL instead.
+	blobPath = apiPath + "/blobs"
+
+	// blobUploadPurpose keeps an upload URL from being replayed to read an entry.
+	blobUploadPurpose = "upload:"
+
+	blobUploadURLTTL = time.Hour
+
+	twirpInternal        = "internal"
+	twirpUnavailable     = "unavailable"
+	twirpUnauthenticated = "unauthenticated"
+)
+
+func (h *Handler) registerV2Routes(router *httprouter.Router) {
+	router.POST(cacheServiceV2Path+"/CreateCacheEntry", h.bearerAuth(h.v2CreateCacheEntry))
+	router.POST(cacheServiceV2Path+"/FinalizeCacheEntryUpload", h.bearerAuth(h.v2FinalizeCacheEntryUpload))
+	router.POST(cacheServiceV2Path+"/GetCacheEntryDownloadURL", h.bearerAuth(h.v2GetCacheEntryDownloadURL))
+	router.PUT(blobPath+"/:id", h.signedAuth(blobUploadPurpose, h.v2UploadBlob))
+}
+
+// An entry that already exists is reported as not ok, which is how the client learns to skip
+// the upload.
+func (h *Handler) v2CreateCacheEntry(w http.ResponseWriter, r *http.Request, _ httprouter.Params) {
+	cred := credFromContext(r.Context())
+	req, err := decodeTwirpRequest[v2CreateRequest](r)
+	if err != nil {
+		h.twirpError(w, r, "malformed", err)
+		return
+	}
+	if req.Key == "" || req.Version == "" {
+		h.twirpError(w, r, "invalid_argument", errors.New("key and version are required"))
+		return
+	}
+
+	db, err := h.openDB()
+	if err != nil {
+		h.twirpError(w, r, twirpInternal, err)
+		return
+	}
+	defer db.Close()
+
+	// An exact (key, version) match means the entry is already cached; the client then skips
+	// the upload. A prefix match must not count here, or a shorter key would be reported as
+	// existing and silently never saved.
+	if existing, err := findExactCache(db, cred.Repo, req.Key, req.Version, true); err != nil {
+		h.twirpError(w, r, twirpInternal, err)
+		return
+	} else if existing != nil {
+		h.touch(db, existing) // the client skips the upload, so this is the only sign the entry is still in use
+		h.twirpNotOK(w, r)
+		return
+	}
+
+	// A second live reservation is finalized by whichever job calls last, against the other's upload.
+	owner := hashedToken(bearerToken(r))
+	if pending, err := findExactCache(db, cred.Repo, req.Key, req.Version, false); err != nil {
+		h.twirpError(w, r, twirpInternal, err)
+		return
+	} else if pending != nil && pending.UsedAt > time.Now().Add(-uploadStallTimeout).Unix() {
+		if pending.Owner != owner {
+			h.twirpNotOK(w, r)
+			return
+		}
+		h.touch(db, pending)                                // still uploading, so it must not go stale under the sweep
+		h.responseJSON(w, r, http.StatusOK, map[string]any{ // this job retrying its own reservation
+			"ok":                true,
+			"signed_upload_url": h.signedURL(cred, blobPath, blobUploadPurpose, pending.ID, time.Now().Add(blobUploadURLTTL)),
+		})
+		return
+	}
+
+	now := time.Now().Unix()
+	cache := &Cache{
+		Repo:      cred.Repo,
+		Key:       req.Key,
+		Version:   req.Version,
+		Owner:     owner,
+		Size:      -1, // the size is only known at finalize time
+		CreatedAt: now,
+		UsedAt:    now,
+	}
+	if err := h.insertCache(db, cache); err != nil {
+		h.twirpError(w, r, twirpInternal, err)
+		return
+	}
+
+	h.responseJSON(w, r, http.StatusOK, map[string]any{
+		"ok":                true,
+		"signed_upload_url": h.signedURL(cred, blobPath, blobUploadPurpose, cache.ID, time.Now().Add(blobUploadURLTTL)),
+	})
+}
+
+func (h *Handler) v2FinalizeCacheEntryUpload(w http.ResponseWriter, r *http.Request, _ httprouter.Params) {
+	cred := credFromContext(r.Context())
+	req, err := decodeTwirpRequest[v2FinalizeRequest](r)
+	if err != nil {
+		h.twirpError(w, r, "malformed", err)
+		return
+	}
+
+	db, err := h.openDB()
+	if err != nil {
+		h.twirpError(w, r, twirpInternal, err)
+		return
+	}
+	defer db.Close()
+
+	cache, err := findExactCache(db, cred.Repo, req.Key, req.Version, false)
+	if err == nil && cache != nil && cache.Owner != hashedToken(bearerToken(r)) {
+		cache = nil // not the reservation this job made, so not this job's to commit
+	}
+	if err == nil && cache == nil {
+		// A retry whose first response was lost finds the entry already committed.
+		cache, err = findExactCache(db, cred.Repo, req.Key, req.Version, true)
+	}
+	if err != nil {
+		h.twirpError(w, r, twirpInternal, err)
+		return
+	}
+	if cache == nil {
+		h.twirpNotOK(w, r)
+		return
+	}
+
+	if !cache.Complete {
+		db.Close() // commitCache needs the store closed
+		cache.Size = int64(cmp.Or(req.SizeBytes, req.SizeBytesCamel))
+		if err := h.commitCache(cache); err != nil {
+			h.logger.Errorf("finalize cache %d (%s): %v", cache.ID, cache.Key, err)
+			h.twirpNotOK(w, r)
+			return
+		}
+	}
+
+	h.responseJSON(w, r, http.StatusOK, map[string]any{
+		"ok": true,
+		// int64 fields travel as strings in the proto JSON mapping.
+		"entry_id": strconv.FormatUint(cache.ID, 10),
+	})
+}
+
+func (h *Handler) v2GetCacheEntryDownloadURL(w http.ResponseWriter, r *http.Request, _ httprouter.Params) {
+	cred := credFromContext(r.Context())
+	req, err := decodeTwirpRequest[v2DownloadRequest](r)
+	if err != nil {
+		h.twirpError(w, r, "malformed", err)
+		return
+	}
+
+	db, err := h.openDB()
+	if err != nil {
+		h.twirpError(w, r, twirpInternal, err)
+		return
+	}
+	defer db.Close()
+
+	cache, err := h.lookupCache(db, cred.Repo, req.keys(), req.Version)
+	if err != nil {
+		h.twirpError(w, r, twirpInternal, err)
+		return
+	}
+	if cache == nil {
+		h.twirpNotOK(w, r)
+		return
+	}
+
+	h.responseJSON(w, r, http.StatusOK, map[string]any{
+		"ok":                  true,
+		"signed_download_url": h.signedArtifactURL(cred, cache.ID, time.Now().Add(artifactURLTTL)),
+		"matched_key":         cache.Key,
+	})
+}
+
+// The archive arrives over the subset of the Azure blob API the toolkit uses: a small
+// cache is a single PUT, a large one is staged as blocks that a final block list puts
+// in order.
+func (h *Handler) v2UploadBlob(w http.ResponseWriter, r *http.Request, params httprouter.Params) {
+	id, err := strconv.ParseUint(params.ByName("id"), 10, 64)
+	if err != nil {
+		h.responseJSON(w, r, http.StatusBadRequest, err)
+		return
+	}
+
+	if err := h.touchCache(id, true); err != nil {
+		h.responseJSON(w, r, http.StatusBadRequest, err)
+		return
+	}
+
+	query := r.URL.Query()
+	switch strings.ToLower(query.Get("comp")) {
+	case "block":
+		blockID := query.Get("blockid")
+		if blockID == "" {
+			h.responseJSON(w, r, http.StatusBadRequest, errors.New("missing blockid"))
+			return
+		}
+		err = h.storage.WriteBlock(id, blockID, r.Body)
+	case "blocklist":
+		var list struct{ Latest []string }
+		if err := xml.NewDecoder(io.LimitReader(r.Body, 8<<20)).Decode(&list); err != nil {
+			h.responseJSON(w, r, http.StatusBadRequest, fmt.Errorf("malformed block list: %w", err))
+			return
+		}
+		err = h.storage.OrderBlocks(id, list.Latest)
+	default:
+		err = h.storage.Write(id, 0, r.Body)
+	}
+	if err != nil {
+		h.responseJSON(w, r, http.StatusInternalServerError, err)
+		return
+	}
+
+	// The Azure SDK client dereferences this without checking, so its absence panics the caller.
+	w.Header().Set("x-ms-request-id", strconv.FormatInt(time.Now().UnixNano(), 10))
+	w.WriteHeader(http.StatusCreated)
+}
+
+// twirpNotOK is the negative answer all three endpoints share: no such entry to restore, no
+// reservation to finalize, or an entry that already exists and need not be uploaded again.
+func (h *Handler) twirpNotOK(w http.ResponseWriter, r *http.Request) {
+	h.responseJSON(w, r, http.StatusOK, map[string]any{"ok": false})
+}
+
+// twirpError reports in the shape a twirp client expects, so the toolkit surfaces the message
+// instead of a parse error.
+func (h *Handler) twirpError(w http.ResponseWriter, r *http.Request, code string, err error) {
+	h.logger.Debugf("%s %s: %v", r.Method, r.URL.Path, err)
+	status := http.StatusBadRequest
+	switch code {
+	case twirpInternal:
+		status = http.StatusInternalServerError
+	case twirpUnavailable:
+		status = http.StatusServiceUnavailable
+	case twirpUnauthenticated:
+		status = http.StatusUnauthorized
+	}
+	h.responseJSON(w, r, status, map[string]any{"code": code, "msg": err.Error()})
+}
+
+// The twirp request bodies. The toolkit's client serialises with useProtoFieldName, so the proto
+// names are what arrive; the camelCase spellings of the same mapping are accepted too, as are
+// int64s sent as a bare number rather than the string the mapping prescribes.
+type (
+	v2CreateRequest struct {
+		Key     string `json:"key"`
+		Version string `json:"version"`
+	}
+
+	v2FinalizeRequest struct {
+		Key            string     `json:"key"`
+		Version        string     `json:"version"`
+		SizeBytes      twirpInt64 `json:"size_bytes"`
+		SizeBytesCamel twirpInt64 `json:"sizeBytes"`
+	}
+
+	v2DownloadRequest struct {
+		Key              string   `json:"key"`
+		Version          string   `json:"version"`
+		RestoreKeys      []string `json:"restore_keys"`
+		RestoreKeysCamel []string `json:"restoreKeys"`
+	}
+)
+
+// twirpInt64 accepts its value as the JSON string the mapping prescribes or as a bare number.
+type twirpInt64 int64
+
+func (n *twirpInt64) UnmarshalJSONFrom(dec *jsontext.Decoder) error {
+	val, err := dec.ReadValue()
+	if err != nil {
+		return err
+	}
+	digits := []byte(val)
+	switch val.Kind() {
+	case 'n': // absent, keep the zero value
+		return nil
+	case '"':
+		if digits, err = jsontext.AppendUnquote(nil, val); err != nil {
+			return err
+		}
+	}
+	parsed, err := strconv.ParseInt(string(digits), 10, 64)
+	if err != nil {
+		return err
+	}
+	*n = twirpInt64(parsed)
+	return nil
+}
+
+func (d v2DownloadRequest) keys() []string {
+	restoreKeys := d.RestoreKeys
+	if len(restoreKeys) == 0 {
+		restoreKeys = d.RestoreKeysCamel
+	}
+	return append([]string{d.Key}, restoreKeys...)
+}
+
+func decodeTwirpRequest[T any](r *http.Request) (T, error) {
+	var req T
+	err := json.UnmarshalRead(io.LimitReader(r.Body, 1<<20), &req)
+	return req, err
+}

act/artifactcache/handler_v2_test.go

@@ -0,0 +1,353 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"bytes"
+	"encoding/base64"
+	"encoding/json/v2"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"strconv"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// v2Call posts a twirp request to the cache service and returns the decoded response.
+// Field names are the proto ones, which is what the toolkit's client sends.
+func v2Call(t *testing.T, handler *Handler, client *http.Client, method string, request any) map[string]any {
+	t.Helper()
+
+	body, err := json.Marshal(request)
+	require.NoError(t, err)
+
+	resp, err := client.Post(handler.ExternalURL()+cacheServiceV2Path+"/"+method, "application/json", bytes.NewReader(body))
+	require.NoError(t, err)
+	defer resp.Body.Close()
+	require.Equal(t, http.StatusOK, resp.StatusCode)
+
+	got := map[string]any{}
+	require.NoError(t, json.UnmarshalRead(resp.Body, &got))
+	return got
+}
+
+// putBlob uploads to a signed URL and returns the status, so a test can assert a refusal.
+func putBlob(t *testing.T, url string, content []byte) int {
+	t.Helper()
+
+	req, err := http.NewRequestWithContext(t.Context(), http.MethodPut, url, bytes.NewReader(content))
+	require.NoError(t, err)
+	resp, err := http.DefaultClient.Do(req)
+	require.NoError(t, err)
+	defer resp.Body.Close()
+	if resp.StatusCode == http.StatusCreated {
+		// The Azure SDK client dereferences this header without checking, so a blob upload that
+		// omits it panics the caller rather than failing it.
+		require.NotEmpty(t, resp.Header.Get("x-ms-request-id"))
+	}
+	return resp.StatusCode
+}
+
+func getURL(t *testing.T, url string) []byte {
+	t.Helper()
+
+	req, err := http.NewRequestWithContext(t.Context(), http.MethodGet, url, nil)
+	require.NoError(t, err)
+	resp, err := http.DefaultClient.Do(req)
+	require.NoError(t, err)
+	defer resp.Body.Close()
+	body, err := io.ReadAll(resp.Body)
+	require.NoError(t, err)
+	return body
+}
+
+// saveV2 runs the reserve/upload/finalize sequence and returns the finalize response along
+// with the upload URL it used.
+func saveV2(t *testing.T, handler *Handler, key, version string, content []byte) (finalized map[string]any, uploadURL string) {
+	t.Helper()
+
+	created := v2Call(t, handler, testClient, "CreateCacheEntry", map[string]any{"key": key, "version": version})
+	require.Equal(t, true, created["ok"])
+	uploadURL, _ = created["signed_upload_url"].(string)
+	require.NotEmpty(t, uploadURL)
+	require.Equal(t, http.StatusCreated, putBlob(t, uploadURL, content))
+
+	return v2Call(t, handler, testClient, "FinalizeCacheEntryUpload", map[string]any{
+		"key": key, "version": version,
+		"size_bytes": strconv.Itoa(len(content)),
+	}), uploadURL
+}
+
+// The whole round trip an actions/cache v2 client makes, plus the guarantees on the signed
+// URLs it is handed: unsigned requests are refused, an upload URL cannot be replayed to read
+// or to replace a finalized entry.
+func TestCacheServiceV2RoundTrip(t *testing.T) {
+	handler := newTestHandler(t, Policy{})
+	content := []byte("the cached archive")
+
+	unsigned := fmt.Sprintf("%s%s/1", handler.ExternalURL(), blobPath)
+	assert.Equal(t, http.StatusUnauthorized, putBlob(t, unsigned, content))
+
+	finalized, uploadURL := saveV2(t, handler, "deps-v1", "abc123", content)
+	require.Equal(t, true, finalized["ok"])
+	assert.NotEmpty(t, finalized["entry_id"])
+
+	// The upload URL outlives the finalize call, so replaying it must not poison the entry,
+	// and it is an upload URL only: nothing reads a blob back through it.
+	assert.Equal(t, http.StatusBadRequest, putBlob(t, uploadURL, []byte("poisoned")))
+	resp, err := http.Get(uploadURL) //nolint:noctx // the URL is the server under test
+	require.NoError(t, err)
+	defer resp.Body.Close()
+	assert.Equal(t, http.StatusMethodNotAllowed, resp.StatusCode)
+
+	got := v2Call(t, handler, testClient, "GetCacheEntryDownloadURL", map[string]any{"key": "deps-v1", "version": "abc123"})
+	require.Equal(t, true, got["ok"])
+	assert.Equal(t, "deps-v1", got["matched_key"])
+	downloadURL, _ := got["signed_download_url"].(string)
+	require.NotEmpty(t, downloadURL)
+	assert.Equal(t, content, getURL(t, downloadURL))
+}
+
+// A large archive is staged as blocks and only put in order by the final block list, so
+// blocks that arrive out of order must still be assembled the way the client asked.
+func TestCacheServiceV2BlockUpload(t *testing.T) {
+	handler := newTestHandler(t, Policy{})
+
+	created := v2Call(t, handler, testClient, "CreateCacheEntry", map[string]any{"key": "blocks", "version": "v1"})
+	uploadURL, _ := created["signed_upload_url"].(string)
+	require.NotEmpty(t, uploadURL)
+
+	blocks := map[string][]byte{}
+	var order []string
+	for i, part := range []string{"hello ", "world", "!"} {
+		blockID := base64.StdEncoding.EncodeToString(fmt.Appendf(nil, "block-%d", i))
+		blocks[blockID] = []byte(part)
+		order = append(order, blockID)
+	}
+	// Upload in an order that is not the block list order.
+	for _, blockID := range []string{order[2], order[0], order[1]} {
+		require.Equal(t, http.StatusCreated, putBlob(t, uploadURL+"&comp=block&blockid="+blockID, blocks[blockID]))
+	}
+
+	var list bytes.Buffer
+	list.WriteString(`<?xml version="1.0" encoding="utf-8"?><BlockList>`)
+	for _, blockID := range order {
+		fmt.Fprintf(&list, "<Latest>%s</Latest>", blockID)
+	}
+	list.WriteString(`</BlockList>`)
+	require.Equal(t, http.StatusCreated, putBlob(t, uploadURL+"&comp=blocklist", list.Bytes()))
+	require.Equal(t, http.StatusCreated, putBlob(t, uploadURL+"&comp=blocklist", list.Bytes()),
+		"a client that lost the first answer retries the list it already sent")
+
+	var reordered bytes.Buffer
+	reordered.WriteString(`<?xml version="1.0" encoding="utf-8"?><BlockList>`)
+	for _, blockID := range []string{order[1], order[0], order[2]} {
+		fmt.Fprintf(&reordered, "<Latest>%s</Latest>", blockID)
+	}
+	reordered.WriteString(`</BlockList>`)
+	assert.Equal(t, http.StatusInternalServerError, putBlob(t, uploadURL+"&comp=blocklist", reordered.Bytes()),
+		"the blocks are already assembled, so a different order would silently disagree with them")
+
+	finalized := v2Call(t, handler, testClient, "FinalizeCacheEntryUpload", map[string]any{
+		"key": "blocks", "version": "v1", "size_bytes": len("hello world!"),
+	})
+	require.Equal(t, true, finalized["ok"])
+
+	got := v2Call(t, handler, testClient, "GetCacheEntryDownloadURL", map[string]any{"key": "blocks", "version": "v1"})
+	require.Equal(t, true, got["ok"])
+	assert.Equal(t, "hello world!", string(getURL(t, got["signed_download_url"].(string))))
+}
+
+func TestCacheServiceV2Lookups(t *testing.T) {
+	handler := newTestHandler(t, Policy{})
+	saved, _ := saveV2(t, handler, "deps-abc", "v1", []byte("x"))
+	require.Equal(t, true, saved["ok"])
+
+	t.Run("reports a miss for an unknown key", func(t *testing.T) {
+		got := v2Call(t, handler, testClient, "GetCacheEntryDownloadURL", map[string]any{"key": "nothing", "version": "v1"})
+		assert.Equal(t, false, got["ok"])
+	})
+
+	// The toolkit serialises with the proto field names; the camelCase spellings of the same
+	// proto JSON mapping are accepted alongside them.
+	for _, field := range []string{"restore_keys", "restoreKeys"} {
+		t.Run("restore keys match by prefix, spelled "+field, func(t *testing.T) {
+			got := v2Call(t, handler, testClient, "GetCacheEntryDownloadURL", map[string]any{
+				"key": "deps-zzz", field: []string{"deps-"}, "version": "v1",
+			})
+			require.Equal(t, true, got["ok"])
+			assert.Equal(t, "deps-abc", got["matched_key"])
+		})
+	}
+
+	t.Run("an existing entry is not reserved twice", func(t *testing.T) {
+		again := v2Call(t, handler, testClient, "CreateCacheEntry", map[string]any{"key": "deps-abc", "version": "v1"})
+		assert.Equal(t, false, again["ok"])
+	})
+
+	// A key that is only a prefix of an existing one is a different entry, so the
+	// reservation check must be exact and not a restore-key prefix match, or the shorter
+	// key would be reported as existing and silently never saved.
+	t.Run("a prefix of an existing key is still reserved", func(t *testing.T) {
+		reserved := v2Call(t, handler, testClient, "CreateCacheEntry", map[string]any{"key": "deps", "version": "v1"})
+		require.Equal(t, true, reserved["ok"])
+		assert.NotEmpty(t, reserved["signed_upload_url"])
+	})
+
+	t.Run("a proxied job is handed the address its runner registered", func(t *testing.T) {
+		const proxy = "https://cache.example.invalid"
+		handler.RegisterJob("proxied", JobCredential{Repo: testRepo, PublicURL: proxy + "/"})
+		client := &http.Client{Transport: &bearerTransport{token: "proxied"}}
+
+		created := v2Call(t, handler, client, "CreateCacheEntry", map[string]any{"key": "proxied-key", "version": "v1"})
+		assert.True(t, strings.HasPrefix(created["signed_upload_url"].(string), proxy+blobPath+"/"))
+
+		got := v2Call(t, handler, client, "GetCacheEntryDownloadURL", map[string]any{"key": "deps-abc", "version": "v1"})
+		assert.True(t, strings.HasPrefix(got["signed_download_url"].(string), proxy+apiPath+"/artifacts/"))
+	})
+
+	t.Run("finalizing without a reservation is not ok", func(t *testing.T) {
+		got := v2Call(t, handler, testClient, "FinalizeCacheEntryUpload", map[string]any{
+			"key": "never-reserved", "version": "v1", "size_bytes": 1,
+		})
+		assert.Equal(t, false, got["ok"])
+	})
+
+	// The size the client declares is what Commit validates the assembled archive against.
+	t.Run("finalizing with the wrong size is not ok", func(t *testing.T) {
+		created := v2Call(t, handler, testClient, "CreateCacheEntry", map[string]any{"key": "wrong-size", "version": "v1"})
+		require.Equal(t, http.StatusCreated, putBlob(t, created["signed_upload_url"].(string), []byte("four")))
+
+		got := v2Call(t, handler, testClient, "FinalizeCacheEntryUpload", map[string]any{
+			"key": "wrong-size", "version": "v1", "size_bytes": 99,
+		})
+		assert.Equal(t, false, got["ok"])
+	})
+
+	// Both API versions are served from one store, so an entry written through v2 is a hit for
+	// a v1 client asking for the same key and version.
+	t.Run("a v1 client sees an entry written through v2", func(t *testing.T) {
+		resp, err := testClient.Get(fmt.Sprintf("%s%s/cache?keys=deps-abc&version=v1", handler.ExternalURL(), apiPath))
+		require.NoError(t, err)
+		defer resp.Body.Close()
+		require.Equal(t, http.StatusOK, resp.StatusCode)
+
+		got := map[string]any{}
+		require.NoError(t, json.UnmarshalRead(resp.Body, &got))
+		assert.Equal(t, "deps-abc", got["cacheKey"])
+		assert.NotEmpty(t, got["archiveLocation"])
+	})
+
+	// The cache of one repository must stay invisible to another, as it does for the v1 API.
+	t.Run("another repository sees nothing", func(t *testing.T) {
+		handler.RegisterJob("other-runtime-token", JobCredential{Repo: "other/repo"})
+		otherClient := &http.Client{Transport: &bearerTransport{token: "other-runtime-token"}}
+
+		got := v2Call(t, handler, otherClient, "GetCacheEntryDownloadURL", map[string]any{"key": "deps-abc", "version": "v1"})
+		assert.Equal(t, false, got["ok"])
+	})
+}
+
+func TestCacheServiceV2RefusalsAreTwirp(t *testing.T) {
+	handler := newTestHandler(t, Policy{})
+
+	tests := []struct {
+		name   string
+		token  string
+		body   string
+		status int
+		code   string
+	}{
+		{"no bearer at all", "", `{"key":"k","version":"v"}`, http.StatusUnauthorized, twirpUnauthenticated},
+		{"a bearer nobody registered", "not-a-job", `{"key":"k","version":"v"}`, http.StatusUnauthorized, twirpUnauthenticated},
+		{"a body that is not the request", testToken, `{`, http.StatusBadRequest, "malformed"},
+		{"a request missing its key", testToken, `{"version":"v"}`, http.StatusBadRequest, "invalid_argument"},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			req, err := http.NewRequestWithContext(t.Context(), http.MethodPost,
+				handler.ExternalURL()+cacheServiceV2Path+"/CreateCacheEntry", strings.NewReader(tt.body))
+			require.NoError(t, err)
+			if tt.token != "" {
+				req.Header.Set("Authorization", "Bearer "+tt.token)
+			}
+			resp, err := http.DefaultClient.Do(req)
+			require.NoError(t, err)
+			defer resp.Body.Close()
+
+			require.Equal(t, tt.status, resp.StatusCode)
+			got := map[string]string{}
+			require.NoError(t, json.UnmarshalRead(resp.Body, &got))
+			assert.Equal(t, tt.code, got["code"])
+			assert.NotEmpty(t, got["msg"])
+		})
+	}
+}
+
+func TestCacheServiceV2ConcurrentSaveAndRetries(t *testing.T) {
+	handler := newTestHandler(t, Policy{})
+	content := []byte("the cached archive")
+
+	const otherJob = "another-jobs-token"
+	defer handler.RegisterJob(otherJob, JobCredential{Repo: testRepo})()
+	reserve := func(client *http.Client) map[string]any {
+		return v2Call(t, handler, client, "CreateCacheEntry", map[string]any{"key": "shared", "version": "v1"})
+	}
+	first := reserve(testClient)
+	require.Equal(t, true, first["ok"])
+	retry := reserve(testClient)
+	assert.Equal(t, true, retry["ok"], "a job retrying its own reservation is not a conflict")
+	entryOf := func(reserved map[string]any) string {
+		parsed, err := url.Parse(reserved["signed_upload_url"].(string))
+		require.NoError(t, err)
+		return parsed.Path
+	}
+	assert.Equal(t, entryOf(first), entryOf(retry), "the same reservation, however freshly its URL is signed")
+	assert.Equal(t, false, reserve(&http.Client{Transport: &bearerTransport{token: otherJob}})["ok"],
+		"another job's reservation would be finalized against the first upload")
+
+	otherClient := &http.Client{Transport: &bearerTransport{token: otherJob}}
+	reserved := v2Call(t, handler, otherClient, "CreateCacheEntry", map[string]any{"key": "owned", "version": "v1"})
+	require.Equal(t, true, reserved["ok"])
+	require.Equal(t, http.StatusCreated, putBlob(t, reserved["signed_upload_url"].(string), content))
+	assert.Equal(t, false, v2Call(t, handler, testClient, "FinalizeCacheEntryUpload",
+		map[string]any{"key": "owned", "version": "v1", "size_bytes": strconv.Itoa(len(content))})["ok"],
+		"another job's upload is not this job's to finalize")
+
+	base := handler.ExternalURL() + apiPath
+	reserveV1, err := json.Marshal(&Request{Key: "v1-made", Version: "v1", Size: int64(len(content))})
+	require.NoError(t, err)
+	resp, err := testClient.Post(base+"/caches", "application/json", bytes.NewReader(reserveV1))
+	require.NoError(t, err)
+	var madeByV1 struct {
+		CacheID uint64 `json:"cacheId"`
+	}
+	require.NoError(t, json.UnmarshalRead(resp.Body, &madeByV1))
+	resp.Body.Close()
+	upload, err := http.NewRequestWithContext(t.Context(), http.MethodPatch,
+		fmt.Sprintf("%s/caches/%d", base, madeByV1.CacheID), bytes.NewReader(content))
+	require.NoError(t, err)
+	upload.Header.Set("Content-Range", fmt.Sprintf("bytes 0-%d/*", len(content)-1))
+	resp, err = testClient.Do(upload)
+	require.NoError(t, err)
+	resp.Body.Close()
+	require.Equal(t, http.StatusOK, resp.StatusCode)
+
+	assert.Equal(t, false, v2Call(t, handler, testClient, "FinalizeCacheEntryUpload",
+		map[string]any{"key": "v1-made", "version": "v1", "size_bytes": strconv.Itoa(len(content))})["ok"],
+		"a reservation this job did not make through v2 is not its to commit")
+
+	finalized, _ := saveV2(t, handler, "deps", "v1", content)
+	require.Equal(t, true, finalized["ok"])
+	again := v2Call(t, handler, testClient, "FinalizeCacheEntryUpload", map[string]any{
+		"key": "deps", "version": "v1", "size_bytes": strconv.Itoa(len(content)),
+	})
+	assert.Equal(t, true, again["ok"], "a retry whose first answer was lost must not report a failed save")
+	assert.Equal(t, finalized["entry_id"], again["entry_id"])
+}

act/artifactcache/model.go

@@ -10,29 +10,13 @@ type Request struct {
 	Size    int64  `json:"cacheSize"`
 }
 
-func (c *Request) ToCache() *Cache {
-	if c == nil {
-		return nil
-	}
-	ret := &Cache{
-		Key:     c.Key,
-		Version: c.Version,
-		Size:    c.Size,
-	}
-	if c.Size == 0 {
-		// So the request comes from old versions of actions, like `actions/cache@v2`.
-		// It doesn't send cache size. Set it to -1 to indicate that.
-		ret.Size = -1
-	}
-	return ret
-}
-
 type Cache struct {
 	ID        uint64 `json:"id" boltholdKey:"ID"`
 	Repo      string `json:"repo" boltholdIndex:"Repo"`
 	Key       string `json:"key" boltholdIndex:"Key"`
 	Version   string `json:"version" boltholdIndex:"Version"`
 	Size      int64  `json:"cacheSize"`
+	Owner     string `json:"owner"` // hashed bearer, so a job's own retry is not another job's reservation
 	Complete  bool   `json:"complete" boltholdIndex:"Complete"`
 	UsedAt    int64  `json:"usedAt" boltholdIndex:"UsedAt"`
 	CreatedAt int64  `json:"createdAt" boltholdIndex:"CreatedAt"`

act/artifactcache/results.go

@@ -0,0 +1,95 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"crypto/tls"
+	"errors"
+	"fmt"
+	"net/http"
+	"net/http/httputil"
+	"net/url"
+	"strings"
+)
+
+// The results service is one origin serving every github.actions.results.api.v1 service, and
+// Gitea implements only the artifact half of it. Forwarding that half from here makes this origin
+// the whole service, so ACTIONS_RESULTS_URL can point at it truthfully, which is what the clients
+// this runner cannot patch need, docker buildx among them.
+const artifactServicePath = "/twirp/github.actions.results.api.v1.ArtifactService/"
+
+// forwardOrNotFound is the router's fallback: the artifact service of the instance the job
+// registered with, and the 404 the router would have written otherwise.
+func (h *Handler) forwardOrNotFound(w http.ResponseWriter, r *http.Request) {
+	if !strings.HasPrefix(r.URL.Path, artifactServicePath) {
+		http.NotFound(w, r)
+		return
+	}
+	cred, ok := h.lookupCredential(bearerToken(r))
+	if !ok {
+		h.twirpError(w, r, twirpUnauthenticated, errors.New("unknown bearer token"))
+		return
+	}
+	if cred.Results == "" {
+		h.twirpError(w, r, twirpInternal, errors.New("no instance is registered for this job"))
+		return
+	}
+	target, err := url.Parse(strings.TrimSuffix(cred.Results, "/"))
+	if err == nil && (target.Hostname() == "" || (target.Scheme != "http" && target.Scheme != "https")) {
+		err = errors.New("not an absolute http or https URL the job could use")
+	}
+	if err != nil {
+		h.logger.Errorf("artifact service forward to %q: %v", cred.Results, err)
+		h.twirpError(w, r, twirpInternal, fmt.Errorf("artifact service address %q is unusable: %w", cred.Results, err))
+		return
+	}
+	if !h.mustProxy(r, cred, target) {
+		redirect := target.JoinPath(r.URL.EscapedPath())
+		redirect.RawQuery = mergeQuery(target.RawQuery, r.URL.RawQuery)
+		h.logger.Debugf("%s %s: redirecting to %s", r.Method, r.URL.Path, target)
+		http.Redirect(w, r, redirect.String(), http.StatusTemporaryRedirect)
+		return
+	}
+	h.logger.Debugf("%s %s: forwarding to %s", r.Method, r.URL.Path, target)
+
+	proxy := &httputil.ReverseProxy{
+		Rewrite: func(r *httputil.ProxyRequest) {
+			r.SetURL(target)
+			// Gitea builds the URLs it hands back from this Host, and their scheme from the
+			// connection unless a forwarded header overrides it, so artifact bodies go to Gitea
+			// directly and never through here.
+			r.Out.Host = target.Host
+		},
+		ErrorHandler: func(w http.ResponseWriter, r *http.Request, err error) {
+			h.logger.Warnf("artifact service forward to %s: %v", target, err)
+			h.twirpError(w, r, twirpInternal, fmt.Errorf("cache server cannot reach the artifact service at %s: %w", target, err))
+		},
+	}
+	if cred.InsecureTLS {
+		proxy.Transport = insecureTransport
+	}
+	proxy.ServeHTTP(w, r)
+}
+
+// A protocol switch loses the bearer or the agent, and an untrusted instance needs skipped verification.
+func (h *Handler) mustProxy(r *http.Request, cred JobCredential, target *url.URL) bool {
+	if cred.InsecureTLS && target.Scheme == "https" {
+		return true
+	}
+	scheme := r.Header.Get("X-Forwarded-Proto")
+	if scheme == "" {
+		scheme = "http"
+	}
+	return scheme != target.Scheme
+}
+
+func mergeQuery(target, request string) string {
+	if target == "" || request == "" {
+		return target + request
+	}
+	return target + "&" + request
+}
+
+// insecureTransport is shared, because a transport per request would pool no connections.
+var insecureTransport = &http.Transport{Proxy: http.ProxyFromEnvironment, TLSClientConfig: &tls.Config{InsecureSkipVerify: true}} //nolint:gosec // the runner reaches its instance on the operator's say-so

act/artifactcache/results_test.go

@@ -0,0 +1,134 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"cmp"
+	"encoding/json/v2"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestFrontResultsService(t *testing.T) {
+	var gotHost, gotPath, gotProto string
+	gitea := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		gotHost, gotPath, gotProto = r.Host, r.URL.Path, r.Header.Get("X-Forwarded-Proto")
+		_, _ = io.WriteString(w, `{"ok":true}`)
+	}))
+	defer gitea.Close()
+	selfSigned := httptest.NewTLSServer(gitea.Config.Handler)
+	defer selfSigned.Close()
+
+	handler, err := StartHandler(Options{Dir: t.TempDir(), OutboundIP: "127.0.0.1"})
+	require.NoError(t, err)
+	defer handler.Close()
+	const token, rpc = "forward-token", artifactServicePath + "CreateArtifact"
+
+	noRedirect := func(token string) *http.Client {
+		return &http.Client{
+			Transport:     &bearerTransport{token: token},
+			CheckRedirect: func(*http.Request, []*http.Request) error { return http.ErrUseLastResponse },
+		}
+	}
+	post := func(path string) (int, map[string]string) {
+		req, err := http.NewRequestWithContext(t.Context(), http.MethodPost, handler.ExternalURL()+path, nil)
+		require.NoError(t, err)
+		resp, err := noRedirect(token).Do(req)
+		require.NoError(t, err)
+		defer resp.Body.Close()
+		body := map[string]string{}
+		_ = json.UnmarshalRead(resp.Body, &body)
+		return resp.StatusCode, body
+	}
+
+	status, body := post(rpc)
+	assert.Equal(t, http.StatusUnauthorized, status, "an unregistered token is told so, not left to parse a 404 page")
+	assert.Equal(t, twirpUnauthenticated, body["code"])
+	assert.Contains(t, body["msg"], "bearer")
+	assert.NotEmpty(t, body["msg"])
+
+	defer handler.RegisterJob(token, JobCredential{Repo: "owner/repo", Results: selfSigned.URL, InsecureTLS: true})()
+
+	status, _ = post(rpc)
+	require.Equal(t, http.StatusOK, status, "an http job reaching an https instance must proxy")
+	assert.Equal(t, strings.TrimPrefix(selfSigned.URL, "https://"), gotHost, "Gitea must see the host it mints its URLs from")
+	assert.Empty(t, gotProto, "a forwarded scheme would make an https Gitea mint http URLs")
+	assert.Equal(t, rpc, gotPath)
+
+	gotPath = ""
+	status, _ = post("/twirp/github.actions.results.api.v1.OtherService/Do")
+	assert.Equal(t, http.StatusNotFound, status)
+	status, _ = post("/api/v1/repos/owner/repo")
+	assert.Equal(t, http.StatusNotFound, status)
+	assert.Empty(t, gotPath, "only the artifact service is forwarded")
+
+	tests := []struct {
+		name           string
+		cred           JobCredential
+		forwardedProto string
+		reqQuery       string
+		want           int
+		location       string
+		wantMsg        string
+	}{
+		{name: "an instance the job reaches itself", cred: JobCredential{Results: gitea.URL}, want: http.StatusTemporaryRedirect},
+		{name: "a sub-path instance keeps its prefix", cred: JobCredential{Results: gitea.URL + "/sub"}, want: http.StatusTemporaryRedirect},
+		{name: "an instance with no certificate to distrust", cred: JobCredential{Results: gitea.URL, InsecureTLS: true}, want: http.StatusTemporaryRedirect},
+		{
+			name: "both queries survive", cred: JobCredential{Results: gitea.URL + "?tenant=one"}, reqQuery: "?run=7",
+			want: http.StatusTemporaryRedirect, location: gitea.URL + rpc + "?tenant=one&run=7",
+		},
+		{
+			name: "a trusted https instance is reached by the job itself",
+			cred: JobCredential{Results: selfSigned.URL}, forwardedProto: "https", want: http.StatusTemporaryRedirect,
+		},
+		{
+			name: "the job does not share this server's disregard for the certificate",
+			cred: JobCredential{Results: selfSigned.URL, InsecureTLS: true}, forwardedProto: "https", want: http.StatusOK,
+		},
+		{name: "an http job is not redirected to an https instance", cred: JobCredential{Results: selfSigned.URL}, want: http.StatusInternalServerError},
+		{
+			name: "a forwarded https scheme onto an http instance is proxied",
+			cred: JobCredential{Results: gitea.URL}, forwardedProto: "https", want: http.StatusOK,
+		},
+		{
+			name: "a scheme the job could not use is refused before it is sent one",
+			cred: JobCredential{Results: "ftp://gitea.example"}, want: http.StatusInternalServerError, wantMsg: "unusable",
+		},
+		{
+			name: "an address with no host is refused before it is sent one",
+			cred: JobCredential{Results: "http://:3000"}, want: http.StatusInternalServerError, wantMsg: "unusable",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			defer handler.RegisterJob(tt.name, tt.cred)()
+			req, err := http.NewRequestWithContext(t.Context(), http.MethodPost, handler.ExternalURL()+rpc+tt.reqQuery, nil)
+			require.NoError(t, err)
+			if tt.forwardedProto != "" {
+				req.Header.Set("X-Forwarded-Proto", tt.forwardedProto)
+			}
+			resp, err := noRedirect(tt.name).Do(req)
+			require.NoError(t, err)
+			defer resp.Body.Close()
+
+			require.Equal(t, tt.want, resp.StatusCode)
+			switch tt.want {
+			case http.StatusTemporaryRedirect:
+				assert.Equal(t, cmp.Or(tt.location, tt.cred.Results+rpc), resp.Header.Get("Location"))
+			case http.StatusInternalServerError:
+				body := map[string]string{}
+				require.NoError(t, json.UnmarshalRead(resp.Body, &body))
+				assert.Equal(t, twirpInternal, body["code"])
+				assert.Contains(t, body["msg"], cmp.Or(tt.wantMsg, tt.cred.Results))
+			}
+		})
+	}
+}

act/artifactcache/s3_storage.go

@@ -0,0 +1,326 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"bytes"
+	"cmp"
+	"context"
+	"crypto/hmac"
+	"crypto/sha256"
+	"encoding/json/v2"
+	"encoding/xml"
+	"errors"
+	"fmt"
+	"io"
+	"io/fs"
+	"maps"
+	"net/http"
+	"net/url"
+	"os"
+	"path"
+	"slices"
+	"strconv"
+	"strings"
+	"time"
+)
+
+// S3Options configures the bucket that stores completed cache entries.
+type S3Options struct {
+	Endpoint  string
+	Region    string
+	Bucket    string
+	Prefix    string
+	PathStyle *bool // nil uses virtual-hosted requests on AWS and path-style ones elsewhere
+	AccessKey string
+	SecretKey string
+}
+
+type cacheStorage interface {
+	Exist(uint64) (bool, error)
+	Write(uint64, int64, io.Reader) error
+	WriteBlock(uint64, string, io.Reader) error
+	OrderBlocks(uint64, []string) error
+	Commit(uint64, int64) (int64, error)
+	Serve(http.ResponseWriter, *http.Request, uint64)
+	Remove(uint64) error
+}
+
+const s3PartSize = 5 << 30 // the largest part S3 accepts
+
+// s3Storage stages uploads on disk and keeps completed entries and their metadata in a bucket.
+type s3Storage struct {
+	*Storage
+	client    *http.Client
+	bucketURL *url.URL
+	opts      S3Options
+}
+
+type s3Part struct {
+	PartNumber int
+	ETag       string
+}
+
+func newS3Storage(staging *Storage, opts S3Options) (*s3Storage, error) {
+	endpoint, err := url.Parse(opts.Endpoint)
+	if err != nil || endpoint.Host == "" || (endpoint.Scheme != "http" && endpoint.Scheme != "https") {
+		return nil, fmt.Errorf("invalid S3 endpoint %q", opts.Endpoint)
+	}
+	transport, _ := http.DefaultTransport.(*http.Transport)
+	transport = transport.Clone()
+	transport.ResponseHeaderTimeout = time.Minute // a stalled endpoint must not hold the store forever
+	pathStyle := !strings.HasSuffix(endpoint.Hostname(), ".amazonaws.com")
+	if opts.PathStyle != nil {
+		pathStyle = *opts.PathStyle
+	}
+	if pathStyle {
+		endpoint.Path = "/" + opts.Bucket
+	} else {
+		endpoint.Host = opts.Bucket + "." + endpoint.Host
+		endpoint.Path = "/"
+	}
+	opts.Region = cmp.Or(opts.Region, "us-east-1")
+	return &s3Storage{Storage: staging, client: &http.Client{Transport: transport}, bucketURL: endpoint, opts: opts}, nil
+}
+
+func (s *s3Storage) blobKey(id uint64) string {
+	return path.Join(s.opts.Prefix, "blobs", strconv.FormatUint(id, 10))
+}
+
+func (s *s3Storage) metadataDir() string {
+	return path.Join(s.opts.Prefix, "metadata") + "/"
+}
+
+func (s *s3Storage) metadataKey(id uint64) string {
+	return s.metadataDir() + strconv.FormatUint(id, 10) + ".json"
+}
+
+func (s *s3Storage) url(key string, query url.Values) *url.URL {
+	u := *s.bucketURL
+	u.Path = path.Join(u.Path, key)
+	u.RawPath = strings.NewReplacer("+", "%20", "%2F", "/").Replace(url.QueryEscape(u.Path)) // SigV4 encoding
+	u.RawQuery = strings.ReplaceAll(query.Encode(), "+", "%20")
+	return &u
+}
+
+// sign authorizes req with AWS Signature Version 4, leaving the payload unsigned so bodies can stream.
+// https://docs.aws.amazon.com/AmazonS3/latest/API/sig-v4-header-based-auth.html
+func (s *s3Storage) sign(req *http.Request, now time.Time) {
+	date := now.UTC().Format("20060102T150405Z")
+	req.Header.Set("X-Amz-Date", date)
+	if req.Header.Get("X-Amz-Content-Sha256") == "" {
+		req.Header.Set("X-Amz-Content-Sha256", "UNSIGNED-PAYLOAD")
+	}
+	headers := map[string]string{"host": req.URL.Host}
+	for name, values := range req.Header {
+		if name = strings.ToLower(name); strings.HasPrefix(name, "x-amz-") {
+			headers[name] = strings.Join(values, ",")
+		}
+	}
+	names := slices.Sorted(maps.Keys(headers))
+	canonical := []string{req.Method, req.URL.EscapedPath(), req.URL.RawQuery}
+	for _, name := range names {
+		canonical = append(canonical, name+":"+headers[name])
+	}
+	canonical = append(canonical, "", strings.Join(names, ";"), req.Header.Get("X-Amz-Content-Sha256"))
+	scope := date[:8] + "/" + s.opts.Region + "/s3/aws4_request"
+	signature := []byte("AWS4" + s.opts.SecretKey)
+	for _, data := range []string{date[:8], s.opts.Region, "s3", "aws4_request", fmt.Sprintf("AWS4-HMAC-SHA256\n%s\n%s\n%x", date, scope, sha256.Sum256([]byte(strings.Join(canonical, "\n"))))} { // key derivation, the last round signs
+		mac := hmac.New(sha256.New, signature)
+		_, _ = mac.Write([]byte(data))
+		signature = mac.Sum(nil)
+	}
+	req.Header.Set("Authorization", fmt.Sprintf("AWS4-HMAC-SHA256 Credential=%s/%s,SignedHeaders=%s,Signature=%x", s.opts.AccessKey, scope, strings.Join(names, ";"), signature))
+}
+
+// do returns the response to a signed request when it succeeded, fs.ErrNotExist for a missing key.
+func (s *s3Storage) do(ctx context.Context, method, key string, query url.Values, header http.Header, body io.Reader) (*http.Response, error) {
+	var size int64
+	if sized, ok := body.(interface{ Size() int64 }); ok {
+		size = sized.Size()
+	}
+	if size == 0 {
+		body = nil
+	}
+	req, err := http.NewRequestWithContext(ctx, method, s.url(key, query).String(), body)
+	if err != nil {
+		return nil, err
+	}
+	req.ContentLength = size
+	maps.Copy(req.Header, header)
+	s.sign(req, time.Now())
+	resp, err := s.client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	if resp.StatusCode < http.StatusMultipleChoices {
+		return resp, nil
+	}
+	defer drain(resp)
+	var failure struct{ Code, Message string }
+	_ = xml.NewDecoder(resp.Body).Decode(&failure)
+	if resp.StatusCode == http.StatusNotFound && (failure.Code == "NoSuchKey" || failure.Code == "" && method == http.MethodHead) {
+		return nil, fs.ErrNotExist
+	}
+	return nil, fmt.Errorf("S3 %s %s: %s", method, key, cmp.Or(strings.TrimSpace(failure.Code+" "+failure.Message), resp.Status))
+}
+
+func (s *s3Storage) call(ctx context.Context, method, key string, query url.Values, body io.Reader, result any) (http.Header, error) {
+	resp, err := s.do(ctx, method, key, query, nil, body)
+	if err != nil {
+		return nil, err
+	}
+	defer drain(resp)
+	if result == nil {
+		return resp.Header, nil
+	}
+	return resp.Header, xml.NewDecoder(resp.Body).Decode(result)
+}
+
+// drain reads a response to its end before closing it, so its connection gets reused.
+func drain(resp *http.Response) {
+	_, _ = io.Copy(io.Discard, resp.Body)
+	_ = resp.Body.Close()
+}
+
+func (s *s3Storage) Exist(id uint64) (bool, error) {
+	_, err := s.call(context.Background(), http.MethodHead, s.blobKey(id), nil, nil, nil)
+	if errors.Is(err, fs.ErrNotExist) {
+		return false, nil
+	}
+	return err == nil, err
+}
+
+func (s *s3Storage) Commit(id uint64, size int64) (int64, error) {
+	written, err := s.Storage.Commit(id, size)
+	if err != nil {
+		return 0, err
+	}
+	defer func() { _ = s.Storage.Remove(id) }()
+	file, err := os.Open(s.filename(id))
+	if err != nil {
+		return 0, err
+	}
+	defer file.Close()
+	if err := s.upload(context.Background(), s.blobKey(id), file, written); err != nil {
+		return 0, fmt.Errorf("upload cache to S3: %w", err)
+	}
+	return written, nil
+}
+
+// upload always goes in parts, the only way S3 takes an object over 5 GiB.
+func (s *s3Storage) upload(ctx context.Context, key string, file io.ReaderAt, size int64) error {
+	var created struct {
+		UploadID string `xml:"UploadId"`
+	}
+	if _, err := s.call(ctx, http.MethodPost, key, url.Values{"uploads": {""}}, nil, &created); err != nil {
+		return err
+	}
+	var completion struct {
+		XMLName xml.Name `xml:"CompleteMultipartUpload"`
+		Parts   []s3Part `xml:"Part"`
+	}
+	for offset := int64(0); offset == 0 || offset < size; offset += s3PartSize {
+		number, length := len(completion.Parts)+1, min(s3PartSize, size-offset)
+		query := url.Values{"partNumber": {strconv.Itoa(number)}, "uploadId": {created.UploadID}}
+		header, err := s.call(ctx, http.MethodPut, key, query, io.NewSectionReader(file, offset, length), nil)
+		if err != nil {
+			return err
+		}
+		completion.Parts = append(completion.Parts, s3Part{number, header.Get("ETag")})
+	}
+	body, err := xml.Marshal(completion)
+	if err != nil {
+		return err
+	}
+	var result struct{ Code, Message string } // a completion can fail after its 200 status
+	if _, err := s.call(ctx, http.MethodPost, key, url.Values{"uploadId": {created.UploadID}}, bytes.NewReader(body), &result); err != nil {
+		return err
+	}
+	if result.Code != "" {
+		return fmt.Errorf("S3 complete upload %s: %s %s", key, result.Code, result.Message)
+	}
+	return nil
+}
+
+func (s *s3Storage) Serve(w http.ResponseWriter, r *http.Request, id uint64) {
+	header := http.Header{}
+	if byteRange := r.Header.Get("Range"); byteRange != "" {
+		header.Set("Range", byteRange)
+	}
+	resp, err := s.do(r.Context(), http.MethodGet, s.blobKey(id), nil, header, nil)
+	if errors.Is(err, fs.ErrNotExist) {
+		http.NotFound(w, r)
+		return
+	} else if err != nil {
+		http.Error(w, "read cache from S3", http.StatusBadGateway)
+		return
+	}
+	defer resp.Body.Close()
+	for _, name := range []string{"Content-Length", "Content-Range"} {
+		if value := resp.Header.Get(name); value != "" {
+			w.Header().Set(name, value)
+		}
+	}
+	w.WriteHeader(resp.StatusCode)
+	_, _ = io.Copy(w, resp.Body)
+}
+
+// Remove unpublishes an entry before deleting its blob, so no runner imports it without its bytes.
+func (s *s3Storage) Remove(id uint64) error {
+	for _, key := range []string{s.metadataKey(id), s.blobKey(id)} {
+		if _, err := s.call(context.Background(), http.MethodDelete, key, nil, nil, nil); err != nil && !errors.Is(err, fs.ErrNotExist) {
+			return fmt.Errorf("remove %s from S3: %w", key, err)
+		}
+	}
+	return s.Storage.Remove(id)
+}
+
+func (s *s3Storage) publish(cache *Cache) error {
+	data, err := json.Marshal(cache)
+	if err != nil {
+		return err
+	}
+	if _, err := s.call(context.Background(), http.MethodPut, s.metadataKey(cache.ID), nil, bytes.NewReader(data), nil); err != nil {
+		return fmt.Errorf("publish cache to S3: %w", err)
+	}
+	return nil
+}
+
+func (s *s3Storage) published(ctx context.Context) (map[uint64]bool, error) {
+	ids := make(map[uint64]bool)
+	query := url.Values{"list-type": {"2"}, "prefix": {s.metadataDir()}}
+	for {
+		var page struct {
+			Contents              []struct{ Key string }
+			IsTruncated           bool
+			NextContinuationToken string
+		}
+		if _, err := s.call(ctx, http.MethodGet, "", query, nil, &page); err != nil {
+			return nil, fmt.Errorf("list S3 cache metadata: %w", err)
+		}
+		for _, object := range page.Contents {
+			if id, err := strconv.ParseUint(strings.TrimSuffix(strings.TrimPrefix(object.Key, s.metadataDir()), ".json"), 10, 64); err == nil {
+				ids[id] = true
+			}
+		}
+		if !page.IsTruncated {
+			return ids, nil
+		}
+		query.Set("continuation-token", page.NextContinuationToken)
+	}
+}
+
+func (s *s3Storage) fetch(ctx context.Context, id uint64) (*Cache, error) {
+	resp, err := s.do(ctx, http.MethodGet, s.metadataKey(id), nil, nil, nil)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+	cache := &Cache{}
+	if err := json.UnmarshalRead(resp.Body, cache); err != nil {
+		return nil, fmt.Errorf("read S3 cache metadata %d: %w", id, err)
+	}
+	return cache, nil
+}

act/artifactcache/s3_storage_test.go

@@ -0,0 +1,46 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"testing"
+	"time"
+
+	"github.com/johannesboyne/gofakes3"
+	"github.com/johannesboyne/gofakes3/backend/s3mem"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestHandler_S3SharesEntriesAndDropsExpiredOnes(t *testing.T) {
+	backend := s3mem.New()
+	require.NoError(t, backend.CreateBucket("cache"))
+	server := httptest.NewServer(gofakes3.New(backend).Server())
+	t.Cleanup(server.Close)
+	opts := Options{S3: &S3Options{Endpoint: server.URL, Bucket: "cache", Prefix: "runners", AccessKey: "key", SecretKey: "secret"}}
+	saver, restorer := startTestHandler(t, opts), startTestHandler(t, opts)
+	entry := map[string]any{"key": "deps", "version": "v1"}
+
+	finalized, _ := saveV2(t, saver, "deps", "v1", []byte("archive"))
+	require.Equal(t, true, finalized["ok"])
+	require.NoError(t, restorer.importMetadata(t.Context()))
+	found := v2Call(t, restorer, testClient, "GetCacheEntryDownloadURL", entry)
+	require.Equal(t, true, found["ok"])
+	assert.Equal(t, []byte("archive"), getURL(t, found["signed_download_url"].(string)))
+
+	require.NoError(t, backend.ForceDeleteBucket("cache"))
+	require.NoError(t, backend.CreateBucket("cache"))
+	require.NoError(t, restorer.importMetadata(t.Context()))
+	assert.Equal(t, true, v2Call(t, restorer, testClient, "CreateCacheEntry", entry)["ok"])
+}
+
+func TestS3StorageSignsLikeAWS(t *testing.T) {
+	storage := &s3Storage{opts: S3Options{Region: "us-east-1", AccessKey: "AKIAIOSFODNN7EXAMPLE", SecretKey: "wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY"}}
+	req := httptest.NewRequest(http.MethodGet, "https://examplebucket.s3.amazonaws.com/?max-keys=2&prefix=J", nil)
+	req.Header.Set("X-Amz-Content-Sha256", "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855")
+	storage.sign(req, time.Date(2013, 5, 24, 0, 0, 0, 0, time.UTC))
+	assert.Equal(t, "AWS4-HMAC-SHA256 Credential=AKIAIOSFODNN7EXAMPLE/20130524/us-east-1/s3/aws4_request,SignedHeaders=host;x-amz-content-sha256;x-amz-date,Signature=34b48302e7b5fa45bde8084f4b7868a86f0a534bc59db6670ed5711ef69dc6f7", req.Header.Get("Authorization"))
+}

act/artifactcache/storage.go

@@ -5,12 +5,17 @@
 package artifactcache
 
 import (
+	"bytes"
+	"crypto/sha256"
+	"encoding/hex"
+	"encoding/json/v2"
 	"fmt"
 	"io"
 	"net/http"
 	"os"
 	"path/filepath"
 	"strconv"
+	"strings"
 )
 
 type Storage struct {
@@ -37,7 +42,10 @@ func (s *Storage) Exist(id uint64) (bool, error) {
 }
 
 func (s *Storage) Write(id uint64, offset int64, reader io.Reader) error {
-	name := s.tempName(id, offset)
+	return s.writeFile(s.tempName(id, offset), reader)
+}
+
+func (s *Storage) writeFile(name string, reader io.Reader) error {
 	if err := os.MkdirAll(filepath.Dir(name), 0o755); err != nil {
 		return err
 	}
@@ -51,6 +59,38 @@ func (s *Storage) Write(id uint64, offset int64, reader io.Reader) error {
 	return err
 }
 
+func (s *Storage) WriteBlock(id uint64, blockID string, reader io.Reader) error {
+	return s.writeFile(s.blockName(id, blockID), reader)
+}
+
+// OrderBlocks renames the staged blocks into the order the block list gives. A block the list
+// does not name keeps its staged name, which is how Commit leaves it out, as Azure drops it. One
+// rename pass is safe because a staged name always carries blockFilePrefix and a target name
+// never does, so no rename can collide with a block not yet moved.
+func (s *Storage) OrderBlocks(id uint64, blockIDs []string) error {
+	wanted, err := json.Marshal(blockIDs) // a block id is client-chosen, so it cannot be a separator
+	if err != nil {
+		return err
+	}
+	list := filepath.Join(s.tempDir(id), blockFilePrefix+"list")
+	if recorded, err := os.ReadFile(list); err == nil {
+		// A retry repeats the list it already sent; another one would reorder what is now staged.
+		if !bytes.Equal(recorded, wanted) {
+			return fmt.Errorf("cache %d was already assembled from a different block list", id)
+		}
+		return nil
+	}
+	for i, blockID := range blockIDs {
+		if err := os.Rename(s.blockName(id, blockID), s.tempName(id, int64(i))); err != nil {
+			if os.IsNotExist(err) {
+				return fmt.Errorf("block %q of cache %d was never uploaded: %w", blockID, id, err)
+			}
+			return err
+		}
+	}
+	return os.WriteFile(list, wanted, 0o600)
+}
+
 func (s *Storage) Commit(id uint64, size int64) (int64, error) {
 	defer func() {
 		_ = os.RemoveAll(s.tempDir(id))
@@ -65,6 +105,31 @@ func (s *Storage) Commit(id uint64, size int64) (int64, error) {
 	if err := os.MkdirAll(filepath.Dir(name), 0o755); err != nil {
 		return 0, err
 	}
+	written, err := assemble(name, tempNames)
+	if err != nil {
+		return 0, err
+	}
+	// If size is less than 0, it means the size is unknown.
+	// We can't check the size of the file, just skip the check.
+	// It happens when the request comes from old versions of actions, like `actions/cache@v2`.
+	if size >= 0 && written != size {
+		_ = os.Remove(name)
+		return 0, fmt.Errorf("broken file: %v != %v", written, size)
+	}
+	return written, nil
+}
+
+// assemble concatenates the uploaded parts into name. A single part, which is what the v2 API
+// produces below the client's block threshold, is already the whole archive and is moved.
+func assemble(name string, tempNames []string) (int64, error) {
+	if len(tempNames) == 1 {
+		info, err := os.Stat(tempNames[0])
+		if err != nil {
+			return 0, err
+		}
+		return info.Size(), os.Rename(tempNames[0], name)
+	}
+
 	file, err := os.Create(name)
 	if err != nil {
 		return 0, err
@@ -84,16 +149,6 @@ func (s *Storage) Commit(id uint64, size int64) (int64, error) {
 		}
 		written += n
 	}
-
-	// If size is less than 0, it means the size is unknown.
-	// We can't check the size of the file, just skip the check.
-	// It happens when the request comes from old versions of actions, like `actions/cache@v2`.
-	if size >= 0 && written != size {
-		_ = file.Close()
-		_ = os.Remove(name)
-		return 0, fmt.Errorf("broken file: %v != %v", written, size)
-	}
-
 	return written, nil
 }
 
@@ -102,9 +157,13 @@ func (s *Storage) Serve(w http.ResponseWriter, r *http.Request, id uint64) {
 	http.ServeFile(w, r, name)
 }
 
-func (s *Storage) Remove(id uint64) {
+// Remove deletes an entry's blob and any staged parts. It reports failure so the caller can
-	_ = os.Remove(s.filename(id))
+// keep the entry and retry, rather than dropping the only reference to bytes on disk.
-	_ = os.RemoveAll(s.tempDir(id))
+func (s *Storage) Remove(id uint64) error {
+	if err := os.Remove(s.filename(id)); err != nil && !os.IsNotExist(err) {
+		return err
+	}
+	return os.RemoveAll(s.tempDir(id))
 }
 
 func (s *Storage) filename(id uint64) string {
@@ -119,6 +178,17 @@ func (s *Storage) tempName(id uint64, offset int64) string {
 	return filepath.Join(s.tempDir(id), fmt.Sprintf("%016x", offset))
 }
 
+// blockFilePrefix marks a staged, not yet ordered block, so that tempNames can keep it out of
+// Commit's name-ordered concatenation.
+const blockFilePrefix = "block-"
+
+func (s *Storage) blockName(id uint64, blockID string) string {
+	// The block id is client-chosen (base64), so it is hashed rather than trusted as a
+	// path element.
+	sum := sha256.Sum256([]byte(blockID))
+	return filepath.Join(s.tempDir(id), blockFilePrefix+hex.EncodeToString(sum[:]))
+}
+
 func (s *Storage) tempNames(id uint64) ([]string, error) {
 	dir := s.tempDir(id)
 	files, err := os.ReadDir(dir)
@@ -127,7 +197,7 @@ func (s *Storage) tempNames(id uint64) ([]string, error) {
 	}
 	var names []string
 	for _, v := range files {
-		if !v.IsDir() {
+		if !v.IsDir() && !strings.HasPrefix(v.Name(), blockFilePrefix) {
 			names = append(names, filepath.Join(dir, v.Name()))
 		}
 	}

act/artifacts/server.go

@@ -6,11 +6,12 @@ package artifacts
 
 import (
 	"context"
-	"encoding/json"
+	"encoding/json/v2"
 	"errors"
 	"fmt"
 	"io"
 	"io/fs"
+	"net"
 	"net/http"
 	"os"
 	"path/filepath"
@@ -50,65 +51,29 @@ type ResponseMessage struct {
 	Message string `json:"message"`
 }
 
-type WritableFile interface {
-	io.WriteCloser
-}
-
-type WriteFS interface {
-	OpenWritable(name string) (WritableFile, error)
-	OpenAppendable(name string) (WritableFile, error)
-}
-
-type readWriteFSImpl struct{}
-
-func (fwfs readWriteFSImpl) Open(name string) (fs.File, error) {
-	return os.Open(name)
-}
-
-func (fwfs readWriteFSImpl) OpenWritable(name string) (WritableFile, error) {
-	if err := os.MkdirAll(filepath.Dir(name), os.ModePerm); err != nil {
-		return nil, err
-	}
-	return os.OpenFile(name, os.O_CREATE|os.O_RDWR|os.O_TRUNC, 0o644)
-}
-
-func (fwfs readWriteFSImpl) OpenAppendable(name string) (WritableFile, error) {
-	if err := os.MkdirAll(filepath.Dir(name), os.ModePerm); err != nil {
-		return nil, err
-	}
-	file, err := os.OpenFile(name, os.O_CREATE|os.O_RDWR, 0o644)
-	if err != nil {
-		return nil, err
-	}
-
-	_, err = file.Seek(0, io.SeekEnd)
-	if err != nil {
-		return nil, err
-	}
-	return file, nil
-}
-
 var gzipExtension = ".gz__"
 
 func safeResolve(baseDir, relPath string) string {
 	return filepath.Join(baseDir, filepath.Clean(filepath.Join(string(os.PathSeparator), relPath)))
 }
 
-func uploads(router *httprouter.Router, baseDir string, fsys WriteFS) {
+func writeJSON(w http.ResponseWriter, value any) {
+	data, err := json.Marshal(value)
+	if err != nil {
+		panic(err)
+	}
+	if _, err := w.Write(data); err != nil {
+		panic(err)
+	}
+}
+
+func uploads(router *httprouter.Router, baseDir string) {
 	router.POST("/_apis/pipelines/workflows/:runId/artifacts", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
 		runID := params.ByName("runId")
 
-		json, err := json.Marshal(FileContainerResourceURL{
+		writeJSON(w, FileContainerResourceURL{
 			FileContainerResourceURL: fmt.Sprintf("http://%s/upload/%s", req.Host, runID),
 		})
-		if err != nil {
-			panic(err)
-		}
-
-		_, err = w.Write(json)
-		if err != nil {
-			panic(err)
-		}
 	})
 
 	router.PUT("/upload/:runId", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
@@ -122,67 +87,47 @@ func uploads(router *httprouter.Router, baseDir string, fsys WriteFS) {
 		safeRunPath := safeResolve(baseDir, runID)
 		safePath := safeResolve(safeRunPath, itemPath)
 
-		file, err := func() (WritableFile, error) {
+		if err := os.MkdirAll(filepath.Dir(safePath), os.ModePerm); err != nil {
-			contentRange := req.Header.Get("Content-Range")
+			panic(err)
-			if contentRange != "" && !strings.HasPrefix(contentRange, "bytes 0-") {
-				return fsys.OpenAppendable(safePath)
 		}
-			return fsys.OpenWritable(safePath)
+		flags := os.O_CREATE | os.O_WRONLY | os.O_TRUNC
-		}()
+		appendUpload := req.Header.Get("Content-Range")
+		if appendUpload != "" && !strings.HasPrefix(appendUpload, "bytes 0-") {
+			flags = os.O_CREATE | os.O_WRONLY | os.O_APPEND
+		}
+		file, err := os.OpenFile(safePath, flags, 0o644)
 		if err != nil {
 			panic(err)
 		}
 		defer file.Close()
-
-		writer, ok := file.(io.Writer)
-		if !ok {
-			panic(errors.New("File is not writable"))
-		}
-
 		if req.Body == nil {
-			panic(errors.New("No body given"))
+			panic(errors.New("no body given"))
 		}
 
-		_, err = io.Copy(writer, req.Body)
+		_, err = io.Copy(file, req.Body)
 		if err != nil {
 			panic(err)
 		}
 
-		json, err := json.Marshal(ResponseMessage{
+		writeJSON(w, ResponseMessage{
 			Message: "success",
 		})
-		if err != nil {
-			panic(err)
-		}
-
-		_, err = w.Write(json)
-		if err != nil {
-			panic(err)
-		}
 	})
 
 	router.PATCH("/_apis/pipelines/workflows/:runId/artifacts", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
-		json, err := json.Marshal(ResponseMessage{
+		writeJSON(w, ResponseMessage{
 			Message: "success",
 		})
-		if err != nil {
-			panic(err)
-		}
-
-		_, err = w.Write(json)
-		if err != nil {
-			panic(err)
-		}
 	})
 }
 
-func downloads(router *httprouter.Router, baseDir string, fsys fs.FS) {
+func downloads(router *httprouter.Router, baseDir string) {
 	router.GET("/_apis/pipelines/workflows/:runId/artifacts", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
 		runID := params.ByName("runId")
 
 		safePath := safeResolve(baseDir, runID)
 
-		entries, err := fs.ReadDir(fsys, safePath)
+		entries, err := os.ReadDir(safePath)
 		if err != nil {
 			panic(err)
 		}
@@ -195,18 +140,10 @@ func downloads(router *httprouter.Router, baseDir string, fsys fs.FS) {
 			})
 		}
 
-		json, err := json.Marshal(NamedFileContainerResourceURLResponse{
+		writeJSON(w, NamedFileContainerResourceURLResponse{
 			Count: len(list),
 			Value: list,
 		})
-		if err != nil {
-			panic(err)
-		}
-
-		_, err = w.Write(json)
-		if err != nil {
-			panic(err)
-		}
 	})
 
 	router.GET("/download/:container", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
@@ -215,7 +152,7 @@ func downloads(router *httprouter.Router, baseDir string, fsys fs.FS) {
 		safePath := safeResolve(baseDir, filepath.Join(container, itemPath))
 
 		var files []ContainerItem
-		err := fs.WalkDir(fsys, safePath, func(path string, entry fs.DirEntry, err error) error {
+		err := filepath.WalkDir(safePath, func(path string, entry fs.DirEntry, err error) error {
 			if !entry.IsDir() {
 				rel, err := filepath.Rel(safePath, path)
 				if err != nil {
@@ -241,17 +178,9 @@ func downloads(router *httprouter.Router, baseDir string, fsys fs.FS) {
 			panic(err)
 		}
 
-		json, err := json.Marshal(ContainerItemResponse{
+		writeJSON(w, ContainerItemResponse{
 			Value: files,
 		})
-		if err != nil {
-			panic(err)
-		}
-
-		_, err = w.Write(json)
-		if err != nil {
-			panic(err)
-		}
 	})
 
 	router.GET("/artifact/*path", func(w http.ResponseWriter, req *http.Request, params httprouter.Params) {
@@ -259,15 +188,16 @@ func downloads(router *httprouter.Router, baseDir string, fsys fs.FS) {
 
 		safePath := safeResolve(baseDir, path)
 
-		file, err := fsys.Open(safePath)
+		file, err := os.Open(safePath)
 		if err != nil {
 			// try gzip file
-			file, err = fsys.Open(safePath + gzipExtension)
+			file, err = os.Open(safePath + gzipExtension)
 			if err != nil {
 				panic(err)
 			}
 			w.Header().Add("Content-Encoding", "gzip")
 		}
+		defer file.Close()
 
 		_, err = io.Copy(w, file)
 		if err != nil {
@@ -287,19 +217,18 @@ func Serve(ctx context.Context, artifactPath, addr, port string) context.CancelF
 	router := httprouter.New()
 
 	logger.Debugf("Artifacts base path '%s'", artifactPath)
-	fsys := readWriteFSImpl{}
+	uploads(router, artifactPath)
-	uploads(router, artifactPath, fsys)
+	downloads(router, artifactPath)
-	downloads(router, artifactPath, fsys)
 
 	server := &http.Server{
-		Addr:              fmt.Sprintf("%s:%s", addr, port),
+		Addr:              net.JoinHostPort(addr, port),
 		ReadHeaderTimeout: 2 * time.Second,
 		Handler:           router,
 	}
 
 	// run server
 	go func() {
-		logger.Infof("Start server on http://%s:%s", addr, port)
+		logger.Infof("Start server on http://%s", server.Addr)
 		if err := server.ListenAndServe(); err != nil && err != http.ErrServerClosed {
 			logger.Fatal(err)
 		}

act/artifacts/server_test.go

@@ -7,8 +7,7 @@ package artifacts
 import (
 	"bytes"
 	"compress/gzip"
-	"encoding/json"
+	"encoding/json/v2"
-	"fmt"
 	"io"
 	"maps"
 	"net/http"
@@ -18,238 +17,18 @@ import (
 	"path/filepath"
 	"strings"
 	"testing"
-	"testing/fstest"
 	"time"
 
 	"github.com/julienschmidt/httprouter"
-	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
 
-type writableMapFile struct {
-	fstest.MapFile
-}
-
-func (f *writableMapFile) Write(data []byte) (int, error) {
-	f.Data = data
-	return len(data), nil
-}
-
-func (f *writableMapFile) Close() error {
-	return nil
-}
-
-type writeMapFS struct {
-	fstest.MapFS
-}
-
-func (fsys writeMapFS) OpenWritable(name string) (WritableFile, error) {
-	file := &writableMapFile{
-		MapFile: fstest.MapFile{
-			Data: []byte("content2"),
-		},
-	}
-	fsys.MapFS[name] = &file.MapFile
-
-	return file, nil
-}
-
-func (fsys writeMapFS) OpenAppendable(name string) (WritableFile, error) {
-	file := &writableMapFile{
-		MapFile: fstest.MapFile{
-			Data: []byte("content2"),
-		},
-	}
-	fsys.MapFS[name] = &file.MapFile
-
-	return file, nil
-}
-
-func TestNewArtifactUploadPrepare(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{})
-
-	router := httprouter.New()
-	uploads(router, "artifact/server/path", writeMapFS{memfs})
-
-	req, _ := http.NewRequest(http.MethodPost, "http://localhost/_apis/pipelines/workflows/1/artifacts", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.Fail("Wrong status")
-	}
-
-	response := FileContainerResourceURL{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Equal("http://localhost/upload/1", response.FileContainerResourceURL)
-}
-
-func TestArtifactUploadBlob(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{})
-
-	router := httprouter.New()
-	uploads(router, "artifact/server/path", writeMapFS{memfs})
-
-	req, _ := http.NewRequest(http.MethodPut, "http://localhost/upload/1?itemPath=some/file", strings.NewReader("content"))
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.Fail("Wrong status")
-	}
-
-	response := ResponseMessage{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Equal("success", response.Message)
-	assert.Equal("content", string(memfs["artifact/server/path/1/some/file"].Data))
-}
-
-func TestFinalizeArtifactUpload(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{})
-
-	router := httprouter.New()
-	uploads(router, "artifact/server/path", writeMapFS{memfs})
-
-	req, _ := http.NewRequest(http.MethodPatch, "http://localhost/_apis/pipelines/workflows/1/artifacts", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.Fail("Wrong status")
-	}
-
-	response := ResponseMessage{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Equal("success", response.Message)
-}
-
-func TestListArtifacts(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{
-		"artifact/server/path/1/file.txt": {
-			Data: []byte(""),
-		},
-	})
-
-	router := httprouter.New()
-	downloads(router, "artifact/server/path", memfs)
-
-	req, _ := http.NewRequest(http.MethodGet, "http://localhost/_apis/pipelines/workflows/1/artifacts", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.FailNow(fmt.Sprintf("Wrong status: %d", status))
-	}
-
-	response := NamedFileContainerResourceURLResponse{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Equal(1, response.Count)
-	assert.Equal("file.txt", response.Value[0].Name)
-	assert.Equal("http://localhost/download/1", response.Value[0].FileContainerResourceURL)
-}
-
-func TestListArtifactContainer(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{
-		"artifact/server/path/1/some/file": {
-			Data: []byte(""),
-		},
-	})
-
-	router := httprouter.New()
-	downloads(router, "artifact/server/path", memfs)
-
-	req, _ := http.NewRequest(http.MethodGet, "http://localhost/download/1?itemPath=some/file", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.FailNow(fmt.Sprintf("Wrong status: %d", status))
-	}
-
-	response := ContainerItemResponse{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Len(response.Value, 1)
-	assert.Equal("some/file", response.Value[0].Path)
-	assert.Equal("file", response.Value[0].ItemType)
-	assert.Equal("http://localhost/artifact/1/some/file/.", response.Value[0].ContentLocation)
-}
-
-func TestDownloadArtifactFile(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{
-		"artifact/server/path/1/some/file": {
-			Data: []byte("content"),
-		},
-	})
-
-	router := httprouter.New()
-	downloads(router, "artifact/server/path", memfs)
-
-	req, _ := http.NewRequest(http.MethodGet, "http://localhost/artifact/1/some/file", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.FailNow(fmt.Sprintf("Wrong status: %d", status))
-	}
-
-	data := rr.Body.Bytes()
-
-	assert.Equal("content", string(data))
-}
-
-// TestArtifactFlow drives the real Serve() artifact server over a loopback socket, exercising
-// the same upload -> finalize -> list -> download protocol the upload-artifact/download-artifact
-// actions speak. Running it in-process (rather than from a job container) keeps it network-free
-// and reachable everywhere, including when the CI job is itself a container.
 func TestArtifactFlow(t *testing.T) {
 	artifactPath := t.TempDir()
 
-	// Serve the exact routes Serve() wires up, on a real loopback socket via httptest. httptest
-	// picks a free port and Close() tears the server down synchronously — avoiding both the
-	// port-rebind race and Serve()'s detached ListenAndServe goroutine, which logger.Fatal()s
-	// (process exit) on a bind error and can outlive the test's temp-dir cleanup.
 	router := httprouter.New()
-	fsys := readWriteFSImpl{}
+	uploads(router, artifactPath)
-	uploads(router, artifactPath, fsys)
+	downloads(router, artifactPath)
-	downloads(router, artifactPath, fsys)
 	server := httptest.NewServer(router)
 	defer server.Close()
 
@@ -257,8 +36,6 @@ func TestArtifactFlow(t *testing.T) {
 	client := server.Client()
 	client.Timeout = 5 * time.Second
 
-	// request performs one HTTP call and returns the status and body. The default transport adds
-	// Accept-Encoding: gzip and transparently decompresses, so gzipped downloads come back plain.
 	request := func(t *testing.T, method, rawURL string, body io.Reader, header http.Header) (int, []byte) {
 		t.Helper()
 		req, err := http.NewRequest(method, rawURL, body)
@@ -289,6 +66,8 @@ func TestArtifactFlow(t *testing.T) {
 
 		status, data = request(t, http.MethodPatch, baseURL+"/_apis/pipelines/workflows/"+runID+"/artifacts", nil, nil)
 		require.Equal(t, http.StatusOK, status, string(data))
+		require.NoError(t, json.Unmarshal(data, &msg))
+		require.Equal(t, "success", msg.Message)
 
 		status, data = request(t, http.MethodGet, baseURL+"/_apis/pipelines/workflows/"+runID+"/artifacts", nil, nil)
 		require.Equal(t, http.StatusOK, status, string(data))
@@ -314,6 +93,21 @@ func TestArtifactFlow(t *testing.T) {
 		require.Equal(t, content, string(stored))
 	})
 
+	t.Run("content-range", func(t *testing.T) {
+		const rawURL = "/upload/4?itemPath=chunks.txt"
+		status, data := request(t, http.MethodPut, baseURL+rawURL, strings.NewReader("first"),
+			http.Header{"Content-Range": []string{"bytes 0-4/11"}})
+		require.Equal(t, http.StatusOK, status, string(data))
+
+		status, data = request(t, http.MethodPut, baseURL+rawURL, strings.NewReader("-second"),
+			http.Header{"Content-Range": []string{"bytes 5-11/11"}})
+		require.Equal(t, http.StatusOK, status, string(data))
+
+		stored, err := os.ReadFile(filepath.Join(artifactPath, "4", "chunks.txt"))
+		require.NoError(t, err)
+		require.Equal(t, "first-second", string(stored))
+	})
+
 	t.Run("gzip-roundtrip", func(t *testing.T) {
 		const runID, item, content = "2", "logs/app.log", "compressed payload\n"
 
@@ -365,9 +159,7 @@ func TestArtifactFlow(t *testing.T) {
 	})
 }
 
-func TestMkdirFsImplSafeResolve(t *testing.T) {
+func TestSafeResolve(t *testing.T) {
-	assert := assert.New(t)
-
 	baseDir := "/foo/bar"
 
 	tests := map[string]struct {
@@ -385,97 +177,13 @@ func TestMkdirFsImplSafeResolve(t *testing.T) {
 
 	for name, tc := range tests {
 		t.Run(name, func(t *testing.T) {
-			assert.Equal(tc.want, safeResolve(baseDir, tc.input))
+			require.Equal(t, tc.want, safeResolve(baseDir, tc.input))
 		})
 	}
 }
 
-func TestReadWriteFSWritableAndAppendable(t *testing.T) {
-	fsys := readWriteFSImpl{}
-	name := filepath.Join(t.TempDir(), "nested", "artifact.txt")
-
-	w, err := fsys.OpenWritable(name)
-	require.NoError(t, err)
-	_, err = w.Write([]byte("first"))
-	require.NoError(t, err)
-	require.NoError(t, w.Close())
-
-	w, err = fsys.OpenAppendable(name)
-	require.NoError(t, err)
-	_, err = w.Write([]byte("-second"))
-	require.NoError(t, err)
-	require.NoError(t, w.Close())
-
-	got, err := os.ReadFile(name)
-	require.NoError(t, err)
-	require.Equal(t, "first-second", string(got))
-
-	w, err = fsys.OpenWritable(name)
-	require.NoError(t, err)
-	_, err = w.Write([]byte("replaced"))
-	require.NoError(t, err)
-	require.NoError(t, w.Close())
-
-	got, err = os.ReadFile(name)
-	require.NoError(t, err)
-	require.Equal(t, "replaced", string(got))
-}
-
 func TestServeEmptyArtifactPathReturnsCancelableNoop(t *testing.T) {
 	cancel := Serve(t.Context(), "", "127.0.0.1", "0")
 	require.NotNil(t, cancel)
 	cancel()
 }
-
-func TestDownloadArtifactFileUnsafePath(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{
-		"artifact/server/path/some/file": {
-			Data: []byte("content"),
-		},
-	})
-
-	router := httprouter.New()
-	downloads(router, "artifact/server/path", memfs)
-
-	req, _ := http.NewRequest(http.MethodGet, "http://localhost/artifact/2/../../some/file", nil)
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.FailNow(fmt.Sprintf("Wrong status: %d", status))
-	}
-
-	data := rr.Body.Bytes()
-
-	assert.Equal("content", string(data))
-}
-
-func TestArtifactUploadBlobUnsafePath(t *testing.T) {
-	assert := assert.New(t)
-
-	memfs := fstest.MapFS(map[string]*fstest.MapFile{})
-
-	router := httprouter.New()
-	uploads(router, "artifact/server/path", writeMapFS{memfs})
-
-	req, _ := http.NewRequest(http.MethodPut, "http://localhost/upload/1?itemPath=../../some/file", strings.NewReader("content"))
-	rr := httptest.NewRecorder()
-
-	router.ServeHTTP(rr, req)
-
-	if status := rr.Code; status != http.StatusOK {
-		assert.Fail("Wrong status")
-	}
-
-	response := ResponseMessage{}
-	err := json.Unmarshal(rr.Body.Bytes(), &response)
-	if err != nil {
-		panic(err)
-	}
-
-	assert.Equal("success", response.Message)
-	assert.Equal("content", string(memfs["artifact/server/path/1/some/file"].Data))
-}

act/common/cartesian.go

@@ -1,60 +0,0 @@
-// Copyright 2026 The Gitea Authors. All rights reserved.
-// Copyright 2020 The nektos/act Authors. All rights reserved.
-// SPDX-License-Identifier: MIT
-
-package common
-
-import "slices"
-
-// CartesianProduct takes map of lists and returns list of unique tuples
-func CartesianProduct(mapOfLists map[string][]any) []map[string]any {
-	listNames := make([]string, 0)
-	lists := make([][]any, 0)
-	for k, v := range mapOfLists {
-		listNames = append(listNames, k)
-		lists = append(lists, v)
-	}
-
-	listCart := cartN(lists...)
-
-	rtn := make([]map[string]any, 0)
-	for _, list := range listCart {
-		vMap := make(map[string]any)
-		for i, v := range list {
-			vMap[listNames[i]] = v
-		}
-		rtn = append(rtn, vMap)
-	}
-	return rtn
-}
-
-func cartN(a ...[]any) [][]any {
-	c := 1
-	for _, a := range a {
-		c *= len(a)
-	}
-	if c == 0 || len(a) == 0 {
-		return nil
-	}
-	p := make([][]any, c)
-	b := make([]any, c*len(a))
-	n := make([]int, len(a))
-	s := 0
-	for i := range p {
-		e := s + len(a)
-		pi := b[s:e]
-		p[i] = pi
-		s = e
-		for j, n := range n {
-			pi[j] = a[j][n]
-		}
-		for j := range slices.Backward(n) {
-			n[j]++
-			if n[j] < len(a[j]) {
-				break
-			}
-			n[j] = 0
-		}
-	}
-	return p
-}

act/common/cartesian_test.go

@@ -1,43 +0,0 @@
-// Copyright 2026 The Gitea Authors. All rights reserved.
-// Copyright 2020 The nektos/act Authors. All rights reserved.
-// SPDX-License-Identifier: MIT
-
-package common
-
-import (
-	"testing"
-
-	"github.com/stretchr/testify/assert"
-)
-
-func TestCartesianProduct(t *testing.T) {
-	assert := assert.New(t)
-	input := map[string][]any{
-		"foo": {1, 2, 3, 4},
-		"bar": {"a", "b", "c"},
-		"baz": {false, true},
-	}
-
-	output := CartesianProduct(input)
-	assert.Len(output, 24)
-
-	for _, v := range output {
-		assert.Len(v, 3)
-
-		assert.Contains(v, "foo")
-		assert.Contains(v, "bar")
-		assert.Contains(v, "baz")
-	}
-
-	input = map[string][]any{
-		"foo": {1, 2, 3, 4},
-		"bar": {},
-		"baz": {false, true},
-	}
-	output = CartesianProduct(input)
-	assert.Empty(output)
-
-	input = map[string][]any{}
-	output = CartesianProduct(input)
-	assert.Empty(output)
-}

act/common/executor.go

@@ -54,22 +54,6 @@ func NewPipelineExecutor(executors ...Executor) Executor {
 	return rtn
 }
 
-// NewConditionalExecutor creates a new executor based on conditions
-func NewConditionalExecutor(conditional Conditional, trueExecutor, falseExecutor Executor) Executor {
-	return func(ctx context.Context) error {
-		if conditional(ctx) {
-			if trueExecutor != nil {
-				return trueExecutor(ctx)
-			}
-		} else {
-			if falseExecutor != nil {
-				return falseExecutor(ctx)
-			}
-		}
-		return nil
-	}
-}
-
 // NewErrorExecutor creates a new executor that always errors out
 func NewErrorExecutor(err error) Executor {
 	return func(ctx context.Context) error {
@@ -187,15 +171,8 @@ func (e Executor) Finally(finally Executor) Executor {
 		err := e(ctx)
 		err2 := finally(ctx)
 		if err2 != nil {
-			return fmt.Errorf("Error occurred running finally: %v (original error: %v)", err2, err)
+			return fmt.Errorf("error occurred running finally: %v (original error: %v)", err2, err)
 		}
 		return err
 	}
 }
-
-// Not return an inverted conditional
-func (c Conditional) Not() Conditional {
-	return func(ctx context.Context) bool {
-		return !c(ctx)
-	}
-}

act/common/executor_max_parallel_test.go

@@ -1,89 +0,0 @@
-// Copyright 2026 The Gitea Authors. All rights reserved.
-// SPDX-License-Identifier: MIT
-
-package common
-
-import (
-	"context"
-	"sync"
-	"sync/atomic"
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/assert"
-)
-
-// Simple fast test that verifies max-parallel: 2 limits concurrency
-func TestMaxParallel2Quick(t *testing.T) {
-	ctx := context.Background()
-
-	var currentRunning atomic.Int32
-	var maxSimultaneous atomic.Int32
-
-	executors := make([]Executor, 4)
-	for i := range 4 {
-		executors[i] = func(ctx context.Context) error {
-			current := currentRunning.Add(1)
-
-			// Update max if needed
-			for {
-				maxValue := maxSimultaneous.Load()
-				if current <= maxValue || maxSimultaneous.CompareAndSwap(maxValue, current) {
-					break
-				}
-			}
-
-			time.Sleep(10 * time.Millisecond)
-			currentRunning.Add(-1)
-			return nil
-		}
-	}
-
-	err := NewParallelExecutor(2, executors...)(ctx)
-
-	assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.LessOrEqual(t, maxSimultaneous.Load(), int32(2),
-		"Should not exceed max-parallel: 2")
-}
-
-// Test that verifies max-parallel: 1 enforces sequential execution
-func TestMaxParallel1Sequential(t *testing.T) {
-	ctx := context.Background()
-
-	var currentRunning atomic.Int32
-	var maxSimultaneous atomic.Int32
-	var executionOrder []int
-	var orderMutex sync.Mutex
-
-	executors := make([]Executor, 5)
-	for i := range 5 {
-		taskID := i
-		executors[i] = func(ctx context.Context) error {
-			current := currentRunning.Add(1)
-
-			// Track execution order
-			orderMutex.Lock()
-			executionOrder = append(executionOrder, taskID)
-			orderMutex.Unlock()
-
-			// Update max if needed
-			for {
-				maxValue := maxSimultaneous.Load()
-				if current <= maxValue || maxSimultaneous.CompareAndSwap(maxValue, current) {
-					break
-				}
-			}
-
-			time.Sleep(20 * time.Millisecond)
-			currentRunning.Add(-1)
-			return nil
-		}
-	}
-
-	err := NewParallelExecutor(1, executors...)(ctx)
-
-	assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.Equal(t, int32(1), maxSimultaneous.Load(),
-		"max-parallel: 1 should only run 1 task at a time")
-	assert.Len(t, executionOrder, 5, "All 5 tasks should have executed")
-}

act/common/executor_parallel_advanced_test.go

@@ -1,221 +0,0 @@
-// Copyright 2026 The Gitea Authors. All rights reserved.
-// SPDX-License-Identifier: MIT
-
-package common
-
-import (
-	"context"
-	"sync"
-	"sync/atomic"
-	"testing"
-	"time"
-
-	"github.com/stretchr/testify/assert"
-)
-
-// TestMaxParallelJobExecution tests actual job execution with max-parallel
-func TestMaxParallelJobExecution(t *testing.T) {
-	t.Run("MaxParallel=1 Sequential", func(t *testing.T) {
-		var currentRunning atomic.Int32
-		var maxConcurrent int32
-		var executionOrder []int
-		var mu sync.Mutex
-
-		executors := make([]Executor, 5)
-		for i := range 5 {
-			taskID := i
-			executors[i] = func(ctx context.Context) error {
-				current := currentRunning.Add(1)
-
-				// Track max concurrent
-				for {
-					maxValue := atomic.LoadInt32(&maxConcurrent)
-					if current <= maxValue || atomic.CompareAndSwapInt32(&maxConcurrent, maxValue, current) {
-						break
-					}
-				}
-
-				mu.Lock()
-				executionOrder = append(executionOrder, taskID)
-				mu.Unlock()
-
-				time.Sleep(10 * time.Millisecond)
-				currentRunning.Add(-1)
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		err := NewParallelExecutor(1, executors...)(ctx)
-		assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		assert.Equal(t, int32(1), maxConcurrent, "Should never exceed 1 concurrent execution")
-		assert.Len(t, executionOrder, 5, "All tasks should execute")
-	})
-
-	t.Run("MaxParallel=3 Limited", func(t *testing.T) {
-		var currentRunning atomic.Int32
-		var maxConcurrent int32
-
-		executors := make([]Executor, 10)
-		for i := range 10 {
-			executors[i] = func(ctx context.Context) error {
-				current := currentRunning.Add(1)
-
-				for {
-					maxValue := atomic.LoadInt32(&maxConcurrent)
-					if current <= maxValue || atomic.CompareAndSwapInt32(&maxConcurrent, maxValue, current) {
-						break
-					}
-				}
-
-				time.Sleep(20 * time.Millisecond)
-				currentRunning.Add(-1)
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		err := NewParallelExecutor(3, executors...)(ctx)
-		assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		assert.LessOrEqual(t, int(maxConcurrent), 3, "Should never exceed 3 concurrent executions")
-		assert.GreaterOrEqual(t, int(maxConcurrent), 1, "Should have at least 1 concurrent execution")
-	})
-
-	t.Run("MaxParallel=0 Uses1Worker", func(t *testing.T) {
-		var maxConcurrent int32
-		var currentRunning atomic.Int32
-
-		executors := make([]Executor, 5)
-		for i := range 5 {
-			executors[i] = func(ctx context.Context) error {
-				current := currentRunning.Add(1)
-
-				for {
-					maxValue := atomic.LoadInt32(&maxConcurrent)
-					if current <= maxValue || atomic.CompareAndSwapInt32(&maxConcurrent, maxValue, current) {
-						break
-					}
-				}
-
-				time.Sleep(10 * time.Millisecond)
-				currentRunning.Add(-1)
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		// When maxParallel is 0 or negative, it defaults to 1
-		err := NewParallelExecutor(0, executors...)(ctx)
-		assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		assert.Equal(t, int32(1), maxConcurrent, "Should use 1 worker when max-parallel is 0")
-	})
-}
-
-// TestMaxParallelWithErrors tests error handling with max-parallel
-func TestMaxParallelWithErrors(t *testing.T) {
-	t.Run("OneTaskFailsOthersContinue", func(t *testing.T) {
-		var successCount int32
-
-		executors := make([]Executor, 5)
-		for i := range 5 {
-			taskID := i
-			executors[i] = func(ctx context.Context) error {
-				if taskID == 2 {
-					return assert.AnError
-				}
-				atomic.AddInt32(&successCount, 1)
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		err := NewParallelExecutor(2, executors...)(ctx)
-
-		// Should return the error from task 2
-		assert.Error(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		// Other tasks should still execute
-		assert.Equal(t, int32(4), successCount, "4 tasks should succeed")
-	})
-
-	t.Run("ContextCancellation", func(t *testing.T) {
-		ctx, cancel := context.WithCancel(context.Background())
-
-		var startedCount int32
-		executors := make([]Executor, 10)
-		for i := range 10 {
-			executors[i] = func(ctx context.Context) error {
-				atomic.AddInt32(&startedCount, 1)
-				time.Sleep(100 * time.Millisecond)
-				return nil
-			}
-		}
-
-		// Cancel after a short delay
-		go func() {
-			time.Sleep(30 * time.Millisecond)
-			cancel()
-		}()
-
-		err := NewParallelExecutor(3, executors...)(ctx)
-		assert.Error(t, err)                     //nolint:testifylint // pre-existing issue from nektos/act
-		assert.ErrorIs(t, err, context.Canceled) //nolint:testifylint // pre-existing issue from nektos/act
-
-		// Not all tasks should start due to cancellation (but timing may vary)
-		// Just verify cancellation occurred
-		t.Logf("Started %d tasks before cancellation", startedCount)
-	})
-}
-
-// TestMaxParallelResourceSharing tests resource sharing scenarios
-func TestMaxParallelResourceSharing(t *testing.T) {
-	t.Run("SharedResourceWithMutex", func(t *testing.T) {
-		var sharedCounter int
-		var mu sync.Mutex
-
-		executors := make([]Executor, 100)
-		for i := range 100 {
-			executors[i] = func(ctx context.Context) error {
-				mu.Lock()
-				sharedCounter++
-				mu.Unlock()
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		err := NewParallelExecutor(10, executors...)(ctx)
-		assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		assert.Equal(t, 100, sharedCounter, "All tasks should increment counter")
-	})
-
-	t.Run("ChannelCommunication", func(t *testing.T) {
-		resultChan := make(chan int, 50)
-
-		executors := make([]Executor, 50)
-		for i := range 50 {
-			taskID := i
-			executors[i] = func(ctx context.Context) error {
-				resultChan <- taskID
-				return nil
-			}
-		}
-
-		ctx := context.Background()
-		err := NewParallelExecutor(5, executors...)(ctx)
-		assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
-
-		close(resultChan)
-
-		results := make(map[int]bool)
-		for result := range resultChan {
-			results[result] = true
-		}
-
-		assert.Len(t, results, 50, "All task IDs should be received")
-	})
-}

act/common/executor_test.go

@@ -9,9 +9,9 @@ import (
 	"errors"
 	"reflect"
 	"strings"
+	"sync"
 	"sync/atomic"
 	"testing"
-	"time"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -45,81 +45,45 @@ func TestNewWorkflow(t *testing.T) {
 	assert.Equal(2, runcount)
 }
 
-func TestNewConditionalExecutor(t *testing.T) {
+// concurrencyProbe returns an executor recording the peak number of concurrent copies. Copies
-	assert := assert.New(t)
+// block until wantActive are in flight so the peak is exact without sleeping, and later copies
+// find the gate already open so the last one still finishes with no partner left.
+func concurrencyProbe(wantActive int32) (exec Executor, count, maxActive *atomic.Int32) {
+	var counted, active, peak atomic.Int32
+	var once sync.Once
+	reached := make(chan struct{})
 
-	ctx := context.Background()
+	return func(ctx context.Context) error {
-
+		counted.Add(1)
-	trueCount := 0
+		running := active.Add(1)
-	falseCount := 0
-
-	err := NewConditionalExecutor(func(ctx context.Context) bool {
-		return false
-	}, func(ctx context.Context) error {
-		trueCount++
-		return nil
-	}, func(ctx context.Context) error {
-		falseCount++
-		return nil
-	})(ctx)
-
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.Equal(0, trueCount)
-	assert.Equal(1, falseCount)
-
-	err = NewConditionalExecutor(func(ctx context.Context) bool {
-		return true
-	}, func(ctx context.Context) error {
-		trueCount++
-		return nil
-	}, func(ctx context.Context) error {
-		falseCount++
-		return nil
-	})(ctx)
-
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.Equal(1, trueCount)
-	assert.Equal(1, falseCount)
-}
-
-func TestNewParallelExecutor(t *testing.T) {
-	assert := assert.New(t)
-
-	ctx := context.Background()
-
-	var count, activeCount, maxCount atomic.Int32
-	emptyWorkflow := NewPipelineExecutor(func(ctx context.Context) error {
-		count.Add(1)
-
-		active := activeCount.Add(1)
 		for {
-			m := maxCount.Load()
+			seen := peak.Load()
-			if active <= m || maxCount.CompareAndSwap(m, active) {
+			if running <= seen || peak.CompareAndSwap(seen, running) {
 				break
 			}
 		}
-		time.Sleep(2 * time.Second)
+		if running >= wantActive {
-		activeCount.Add(-1)
+			once.Do(func() { close(reached) })
-
+		}
+		<-reached
+		active.Add(-1)
 		return nil
-	})
+	}, &counted, &peak
+}
 
-	err := NewParallelExecutor(2, emptyWorkflow, emptyWorkflow, emptyWorkflow)(ctx)
+func TestNewParallelExecutor(t *testing.T) {
+	ctx := context.Background()
 
-	assert.Equal(int32(3), count.Load(), "should run all 3 executors")
+	exec, count, maxActive := concurrencyProbe(2)
-	assert.Equal(int32(2), maxCount.Load(), "should run at most 2 executors in parallel")
+	require.NoError(t, NewParallelExecutor(2, exec, exec, exec)(ctx))
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
+	assert.Equal(t, int32(3), count.Load(), "should run all 3 executors")
+	assert.Equal(t, int32(2), maxActive.Load(), "should run at most 2 executors in parallel")
 
-	// Reset to test running the executor with 0 parallelism
+	// parallelism below 1 falls back to a single worker
-	count.Store(0)
+	exec, count, maxActive = concurrencyProbe(1)
-	activeCount.Store(0)
+	require.NoError(t, NewParallelExecutor(0, exec, exec, exec)(ctx))
-	maxCount.Store(0)
+	assert.Equal(t, int32(3), count.Load(), "should run all 3 executors")
-
+	assert.Equal(t, int32(1), maxActive.Load(), "should run at most 1 executor in parallel")
-	errSingle := NewParallelExecutor(0, emptyWorkflow, emptyWorkflow, emptyWorkflow)(ctx)
-
-	assert.Equal(int32(3), count.Load(), "should run all 3 executors")
-	assert.Equal(int32(1), maxCount.Load(), "should run at most 1 executors in parallel")
-	assert.NoError(errSingle)
 }
 
 func TestNewParallelExecutorEmpty(t *testing.T) {
@@ -173,6 +137,23 @@ func TestNewParallelExecutorCanceled(t *testing.T) {
 	assert.Error(errExpected, err) //nolint:testifylint // pre-existing issue from nektos/act
 }
 
+func TestNewParallelExecutorRunsRemainingAfterFailure(t *testing.T) {
+	var successCount atomic.Int32
+	executors := make([]Executor, 5)
+	for i := range executors {
+		executors[i] = func(ctx context.Context) error {
+			if i == 2 {
+				return errors.New("fake error")
+			}
+			successCount.Add(1)
+			return nil
+		}
+	}
+
+	require.Error(t, NewParallelExecutor(2, executors...)(context.Background()))
+	assert.Equal(t, int32(4), successCount.Load(), "a failing executor must not stop the others")
+}
+
 func TestExecutorConditionalsAndFinally(t *testing.T) {
 	ctx := context.Background()
 	var calls []string
@@ -205,10 +186,3 @@ func TestExecutorFinallyReturnsFinallyErrorWithOriginal(t *testing.T) {
 		t.Fatalf("finally error = %q, want both cleanup and original error", err)
 	}
 }
-
-func TestConditionalNot(t *testing.T) {
-	cond := Conditional(func(context.Context) bool { return false })
-	if !cond.Not()(context.Background()) {
-		t.Fatal("inverted conditional should be true")
-	}
-}

act/common/git/git_test.go

@@ -6,7 +6,10 @@ package git
 
 import (
 	"context"
+	"encoding/base64"
 	"fmt"
+	"net/http"
+	"net/http/httptest"
 	"os"
 	"os/exec"
 	"path/filepath"
@@ -46,9 +49,7 @@ func TestFindGitSlug(t *testing.T) {
 	}
 
 	for _, tt := range slugTests {
-		provider, slug, err := findGitSlug(tt.url, "github.com")
+		provider, slug := findGitSlug(tt.url, "github.com")
-
-		assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
 		assert.Equal(tt.provider, provider)
 		assert.Equal(tt.slug, slug)
 	}
@@ -82,48 +83,37 @@ func cleanGitHooks(dir string) error {
 	return nil
 }
 
-func TestFindGitRemoteURL(t *testing.T) {
+func TestFindGitMetadataOfSHA256RepositoryUsesOrigin(t *testing.T) {
-	assert := assert.New(t)
+	t.Parallel()
-
 	basedir := t.TempDir()
-	err := gitCmd("init", basedir)
+	const remoteURL = "https://github.com/owner/repo.git"
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
+	require.NoError(t, gitCmd("init", "--object-format=sha256", "--initial-branch=main", basedir))
-	err = cleanGitHooks(basedir)
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-
-	remoteURL := "https://git-codecommit.us-east-1.amazonaws.com/v1/repos/my-repo-name"
-	err = gitCmd("-C", basedir, "remote", "add", "origin", remoteURL)
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-
-	u, err := findGitRemoteURL(context.Background(), basedir, "origin")
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.Equal(remoteURL, u)
-
-	remoteURL = "[email protected]/AwesomeOwner/MyAwesomeRepo.git"
-	err = gitCmd("-C", basedir, "remote", "add", "upstream", remoteURL)
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-	u, err = findGitRemoteURL(context.Background(), basedir, "upstream")
-	assert.NoError(err) //nolint:testifylint // pre-existing issue from nektos/act
-	assert.Equal(remoteURL, u)
-}
-
-func TestFindGithubRepoUsesOriginAndCustomRemote(t *testing.T) {
-	basedir := t.TempDir()
-	require.NoError(t, gitCmd("init", basedir))
 	require.NoError(t, cleanGitHooks(basedir))
-	require.NoError(t, gitCmd("-C", basedir, "remote", "add", "origin", "https://github.com/owner/repo.git"))
+	require.NoError(t, gitCmd("-C", basedir, "commit", "--allow-empty", "-m", "init"))
-	require.NoError(t, gitCmd("-C", basedir, "remote", "add", "ghe", "[email protected]:team/project.git"))
+	require.NoError(t, gitCmd("-C", basedir, "remote", "add", "origin", remoteURL))
+	require.NoError(t, gitCmd("-C", basedir, "config", "--add", "remote.origin.url", "https://github.com/other/repo.git"))
 
-	slug, err := FindGithubRepo(context.Background(), basedir, "github.com", "")
+	shortSHA, sha, err := FindGitRevision(context.Background(), basedir)
+	require.NoError(t, err)
+	require.Equal(t, gitRevParse(t, basedir, "HEAD"), sha)
+	require.Len(t, sha, 64)
+	require.Equal(t, sha[:7], shortSHA)
+
+	ref, err := FindGitRef(context.Background(), basedir)
+	require.NoError(t, err)
+	require.Equal(t, "refs/heads/main", ref)
+
+	url, err := findGitRemoteURL(context.Background(), basedir)
+	require.NoError(t, err)
+	require.Equal(t, remoteURL, url)
+
+	slug, err := FindGithubRepo(context.Background(), basedir, "github.com")
 	require.NoError(t, err)
 	require.Equal(t, "owner/repo", slug)
-
-	slug, err = FindGithubRepo(context.Background(), basedir, "git.example.com", "ghe")
-	require.NoError(t, err)
-	require.Equal(t, "team/project", slug)
 }
 
 func TestGitFindRef(t *testing.T) {
+	t.Parallel()
 	basedir := t.TempDir()
 
 	for name, tt := range map[string]struct {
@@ -210,7 +200,9 @@ func TestGitCloneExecutor(t *testing.T) {
 	workDir := t.TempDir()
 	require.NoError(t, gitCmd("clone", remoteDir, workDir))
 	require.NoError(t, gitCmd("-C", workDir, "checkout", "-b", "main"))
-	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "initial"))
+	require.NoError(t, os.WriteFile(filepath.Join(workDir, "action.sh"), []byte("#!/bin/sh\necho hi\n"), 0o755))
+	require.NoError(t, gitCmd("-C", workDir, "add", "action.sh"))
+	require.NoError(t, gitCmd("-C", workDir, "commit", "-m", "initial"))
 	require.NoError(t, gitCmd("-C", workDir, "tag", "v2"))
 	require.NoError(t, gitCmd("-C", workDir, "push", "-u", "origin", "main"))
 	require.NoError(t, gitCmd("-C", workDir, "push", "origin", "v2"))
@@ -224,6 +216,13 @@ func TestGitCloneExecutor(t *testing.T) {
 	require.NoError(t, err)
 	fullSha := strings.TrimSpace(string(out))
 
+	require.NoError(t, gitCmd("-C", workDir, "push", "origin", "main:refs/heads/"+fullSha[:4]))
+
+	hostConfig := filepath.Join(t.TempDir(), "gitconfig")
+	require.NoError(t, os.WriteFile(hostConfig, []byte("[core]\n\tautocrlf = true\n[url \""+remoteDir+"\"]\n\tinsteadOf = https://example.invalid/action\n"), 0o644))
+	t.Setenv("GIT_CONFIG_GLOBAL", hostConfig)
+	t.Setenv("GIT_DIR", t.TempDir())
+
 	for name, tt := range map[string]struct {
 		Err error
 		Ref string
@@ -244,26 +243,42 @@ func TestGitCloneExecutor(t *testing.T) {
 			Err: &Error{ErrShortRef, fullSha},
 			Ref: fullSha[:7],
 		},
+		"HEAD": {
+			Ref: "HEAD",
+		},
+		"hex branch named like its own commit": {
+			Ref: fullSha[:4],
+		},
 	} {
 		t.Run(name, func(t *testing.T) {
+			dir := t.TempDir()
 			clone := NewGitCloneExecutor(NewGitCloneExecutorInput{
-				URL: remoteDir,
+				URL: "https://example.invalid/action",
 				Ref: tt.Ref,
-				Dir: t.TempDir(),
+				Dir: dir,
 			})
 
 			err := clone(context.Background())
 			if tt.Err != nil {
 				assert.Error(t, err) //nolint:testifylint // pre-existing issue from nektos/act
 				assert.Equal(t, tt.Err, err)
-			} else {
+				assert.Equal(t, tt.Err, clone(context.Background()), "a retry on the cache keeps the error")
-				assert.Empty(t, err) //nolint:testifylint // pre-existing issue from nektos/act
+				return
 			}
+			assert.Empty(t, err) //nolint:testifylint // pre-existing issue from nektos/act
+			marker := filepath.Join(dir, "marker")
+			require.NoError(t, os.WriteFile(marker, nil, 0o644))
+			require.NoError(t, clone(context.Background()))
+			assert.FileExists(t, marker, "an insteadOf rewrite must not evict the cache")
+			script, err := os.ReadFile(filepath.Join(dir, "action.sh"))
+			require.NoError(t, err)
+			assert.Equal(t, "#!/bin/sh\necho hi\n", string(script), "host core.autocrlf must not convert action files")
 		})
 	}
 }
 
 func TestGitCloneExecutorReclonesWhenOriginURLChanges(t *testing.T) {
+	t.Parallel()
 	createRemote := func(message string) string {
 		remoteDir := t.TempDir()
 		require.NoError(t, gitCmd("init", "--bare", "--initial-branch=main", remoteDir))
@@ -287,6 +302,11 @@ func TestGitCloneExecutorReclonesWhenOriginURLChanges(t *testing.T) {
 		Dir: cacheDir,
 	})(t.Context()))
 
+	nested := filepath.Join(cacheDir, "nested")
+	require.NoError(t, os.Mkdir(nested, 0o755))
+	require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{URL: oldRemoteDir, Ref: "main", Dir: nested})(t.Context()))
+	assert.DirExists(t, filepath.Join(nested, ".git"), "an empty cache dir inside a clone of the same URL gets its own repository")
+
 	markerPath := filepath.Join(cacheDir, "stale-marker")
 	require.NoError(t, os.WriteFile(markerPath, []byte("stale"), 0o644))
 
@@ -296,7 +316,7 @@ func TestGitCloneExecutorReclonesWhenOriginURLChanges(t *testing.T) {
 		Dir: cacheDir,
 	})(t.Context()))
 
-	originURL, err := findGitRemoteURL(t.Context(), cacheDir, "origin")
+	originURL, err := findGitRemoteURL(t.Context(), cacheDir)
 	require.NoError(t, err)
 	assert.Equal(t, newRemoteDir, originURL)
 
@@ -308,61 +328,36 @@ func TestGitCloneExecutorReclonesWhenOriginURLChanges(t *testing.T) {
 	require.True(t, os.IsNotExist(err), "stale cached directory should be removed before recloning")
 }
 
-func TestGitCloneExecutorNonFastForwardRef(t *testing.T) {
+func TestGitCloneExecutorFollowsRemoteRefChanges(t *testing.T) {
-	// Simulate the scenario where a remote ref (e.g. a GitHub PR head ref) changes
+	t.Parallel()
-	// non-fast-forward between two fetches. Before the fix, the fetch used Force=false,
-	// causing go-git to return ErrForceNeeded and short-circuit the checkout.
-
-	// Create a bare "remote" repo with an initial commit on main and a feature branch.
 	remoteDir := t.TempDir()
 	require.NoError(t, gitCmd("init", "--bare", "--initial-branch=main", remoteDir))
-
-	// We need a working clone to push commits from.
 	workDir := t.TempDir()
 	require.NoError(t, gitCmd("clone", remoteDir, workDir))
 	require.NoError(t, gitCmd("-C", workDir, "checkout", "-b", "main"))
 	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "initial"))
+	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "feature-1"))
 	require.NoError(t, gitCmd("-C", workDir, "push", "-u", "origin", "main"))
 
-	// Create a feature branch (simulates refs/pull/N/head).
+	dir := t.TempDir()
-	require.NoError(t, gitCmd("-C", workDir, "checkout", "-b", "feature"))
+	clone := NewGitCloneExecutor(NewGitCloneExecutorInput{URL: remoteDir, Ref: "main", Dir: dir})
-	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "feature-1"))
-	require.NoError(t, gitCmd("-C", workDir, "push", "origin", "feature"))
-
-	// First clone via the executor — should succeed and cache the repo.
-	cloneDir := t.TempDir()
-	clone := NewGitCloneExecutor(NewGitCloneExecutorInput{
-		URL: remoteDir,
-		Ref: "main",
-		Dir: cloneDir,
-	})
 	require.NoError(t, clone(context.Background()))
 
-	// Now force-push the feature branch to a non-fast-forward commit (simulates
+	require.NoError(t, gitCmd("-C", workDir, "reset", "--hard", "HEAD~1"))
-	// a PR rebase). This makes refs/heads/feature non-fast-forward.
-	require.NoError(t, gitCmd("-C", workDir, "checkout", "main"))
-	require.NoError(t, gitCmd("-C", workDir, "branch", "-D", "feature"))
-	require.NoError(t, gitCmd("-C", workDir, "checkout", "-b", "feature"))
 	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "feature-rewritten"))
-	require.NoError(t, gitCmd("-C", workDir, "push", "--force", "origin", "feature"))
+	require.NoError(t, gitCmd("-C", workDir, "push", "--force", "origin", "main"))
+	require.NoError(t, clone(context.Background()), "a non-fast-forward ref must update")
+	assert.Equal(t, "feature-rewritten", gitHeadSubject(t, dir))
 
-	// Also advance main so we can verify the clone picks up the new commit.
+	tagClone := NewGitCloneExecutor(NewGitCloneExecutorInput{URL: remoteDir, Ref: "v9", Dir: t.TempDir()})
-	require.NoError(t, gitCmd("-C", workDir, "checkout", "main"))
+	require.Error(t, tagClone(context.Background()))
-	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "second"))
+	require.NoError(t, gitCmd("-C", workDir, "tag", "v9"))
-	require.NoError(t, gitCmd("-C", workDir, "push", "origin", "main"))
+	require.NoError(t, gitCmd("-C", workDir, "push", "origin", "v9"))
-
+	require.NoError(t, tagClone(context.Background()), "a tag published after a failed lookup must be fetched")
-	// Second clone to the same directory — before the fix this returned ErrForceNeeded
-	// and left the working tree at the old commit.
-	err := clone(context.Background())
-	require.NoError(t, err, "fetch with non-fast-forward refs must not fail when Force=true")
-
-	// Verify the working tree was actually updated to the latest main commit.
-	out, err := exec.Command("git", "-C", cloneDir, "log", "--oneline", "-1", "--format=%s").Output()
-	require.NoError(t, err)
-	assert.Equal(t, "second", strings.TrimSpace(string(out)), "working tree should be at the latest commit")
 }
 
 func TestGitCloneExecutorOfflineMode(t *testing.T) {
+	t.Parallel()
 	// Build a local "remote" with a single commit on main.
 	remoteDir := t.TempDir()
 	require.NoError(t, gitCmd("init", "--bare", "--initial-branch=main", remoteDir))
@@ -409,6 +404,7 @@ func TestGitCloneExecutorOfflineMode(t *testing.T) {
 }
 
 func TestGitCloneExecutorQuietDemotesCloneLine(t *testing.T) {
+	t.Parallel()
 	remoteDir := t.TempDir()
 	require.NoError(t, gitCmd("init", "--bare", "--initial-branch=main", remoteDir))
 	workDir := t.TempDir()
@@ -447,6 +443,7 @@ func TestGitCloneExecutorQuietDemotesCloneLine(t *testing.T) {
 }
 
 func TestGitCloneExecutorShallow(t *testing.T) {
+	t.Parallel()
 	// Build a local "remote" with several commits on main plus a tag, so a full clone would pull noticeably more history than a shallow one.
 	remoteDir := t.TempDir()
 	require.NoError(t, gitCmd("init", "--bare", "--initial-branch=main", remoteDir))
@@ -457,7 +454,7 @@ func TestGitCloneExecutorShallow(t *testing.T) {
 		require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", m))
 	}
 	require.NoError(t, gitCmd("-C", workDir, "tag", "v1"))
-	sha := gitRevParse(t, workDir, "HEAD~1") // c2, a SHA that go-git cannot shallow-clone
+	unadvertisedSHA := gitRevParse(t, workDir, "HEAD~1")
 	require.NoError(t, gitCmd("-C", workDir, "push", "-u", "origin", "main"))
 	require.NoError(t, gitCmd("-C", workDir, "push", "origin", "v1"))
 
@@ -483,34 +480,105 @@ func TestGitCloneExecutorShallow(t *testing.T) {
 		assert.Equal(t, "c3", gitHeadSubject(t, dir))
 	})
 
-	t.Run("SHA falls back to a full clone", func(t *testing.T) {
+	unadvertisedRemote := func(t *testing.T, allowUnadvertised bool) string {
+		remote := filepath.Join(t.TempDir(), "remote.git")
+		require.NoError(t, gitCmd("clone", "--bare", remoteDir, remote))
+		require.NoError(t, gitCmd("-C", remote, "config", "uploadpack.allowAnySHA1InWant", strconv.FormatBool(allowUnadvertised)))
+		return remote
+	}
+
+	t.Run("commit hash resolves when the remote accepts a direct shallow fetch", func(t *testing.T) {
 		dir := t.TempDir()
 		require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
-			URL: remoteDir, Ref: sha, Dir: dir, Depth: 1,
+			URL: unadvertisedRemote(t, false), Ref: unadvertisedSHA, Dir: dir, Depth: 1,
 		})(t.Context()))
-		// go-git cannot shallow-clone a raw SHA, so it falls back to a full clone; the absence of a shallow marker proves the fallback happened.
+		assert.Equal(t, unadvertisedSHA, gitRevParse(t, dir, "HEAD"))
-		assert.NoFileExists(t, shallowMarker(dir), "a SHA ref must not produce a shallow clone")
-		assert.Equal(t, sha, gitRevParse(t, dir, "HEAD"))
 	})
 
-	t.Run("moving branch updates while staying shallow", func(t *testing.T) {
+	t.Run("commit hash is fetched shallowly, moved to another hash, then reused without reaching the remote", func(t *testing.T) {
+		remote := unadvertisedRemote(t, true)
 		dir := t.TempDir()
-		require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+		cloneAt := func(ref string) error {
-			URL: remoteDir, Ref: "main", Dir: dir, Depth: 1,
+			return NewGitCloneExecutor(NewGitCloneExecutorInput{
-		})(t.Context()))
+				URL: remote, Ref: ref, Dir: dir, Depth: 1,
+			})(t.Context())
+		}
+
+		require.NoError(t, cloneAt(unadvertisedSHA))
+		assert.FileExists(t, shallowMarker(dir))
+		assert.Equal(t, 1, gitRevCount(t, dir))
+		assert.Equal(t, unadvertisedSHA, gitRevParse(t, dir, "HEAD"))
+
+		olderSHA := gitRevParse(t, workDir, "HEAD~2")
+		require.NoError(t, cloneAt(olderSHA))
+		assert.Equal(t, olderSHA, gitRevParse(t, dir, "HEAD"))
+
+		require.NoError(t, os.RemoveAll(remote))
+		require.NoError(t, cloneAt(olderSHA))
+		assert.Equal(t, olderSHA, gitRevParse(t, dir, "HEAD"))
+	})
+
+	t.Run("moving hexadecimal-named branch updates while staying shallow", func(t *testing.T) {
+		require.NoError(t, gitCmd("-C", workDir, "push", "origin", "main:deadbeef"))
+		dir := t.TempDir()
+		clone := NewGitCloneExecutor(NewGitCloneExecutorInput{
+			URL: remoteDir, Ref: "deadbeef", Dir: dir, Depth: 1,
+		})
+		require.NoError(t, clone(t.Context()))
 		require.Equal(t, "c3", gitHeadSubject(t, dir))
 
-		// Advance main on the remote, then reuse the existing shallow clone.
 		require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "c4"))
-		require.NoError(t, gitCmd("-C", workDir, "push", "origin", "main"))
+		require.NoError(t, gitCmd("-C", workDir, "push", "origin", "main:deadbeef"))
 
-		require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+		require.NoError(t, clone(t.Context()))
-			URL: remoteDir, Ref: "main", Dir: dir, Depth: 1,
-		})(t.Context()))
 		assert.Equal(t, "c4", gitHeadSubject(t, dir), "reused shallow clone should update to the new tip")
 		assert.FileExists(t, shallowMarker(dir), "repo should remain shallow after update")
 		assert.Equal(t, 1, gitRevCount(t, dir))
 	})
+
+	t.Run("sha256 full hash is cloned shallowly, reused without the remote, and its 40-character prefix is short", func(t *testing.T) {
+		remote := filepath.Join(t.TempDir(), "remote.git")
+		require.NoError(t, gitCmd("init", "--bare", "--object-format=sha256", "--initial-branch=main", remote))
+		work := t.TempDir()
+		require.NoError(t, gitCmd("init", "--object-format=sha256", "--initial-branch=main", work))
+		require.NoError(t, gitCmd("-C", work, "remote", "add", "origin", remote))
+		require.NoError(t, gitCmd("-C", work, "commit", "--allow-empty", "-m", "pinned"))
+		require.NoError(t, gitCmd("-C", work, "push", "origin", "main"))
+		sha := gitRevParse(t, work, "HEAD")
+		dir := t.TempDir()
+		clone := NewGitCloneExecutor(NewGitCloneExecutorInput{URL: remote, Ref: sha, Dir: dir, Depth: 1})
+		require.NoError(t, clone(t.Context()))
+		assert.FileExists(t, shallowMarker(dir))
+		require.ErrorIs(t, NewGitCloneExecutor(NewGitCloneExecutorInput{URL: remote, Ref: sha[:40], Dir: dir, Depth: 1, OfflineMode: true})(t.Context()), ErrShortRef)
+		require.NoError(t, os.RemoveAll(remote))
+		require.NoError(t, clone(t.Context()))
+		assert.Equal(t, sha, gitRevParse(t, dir, "HEAD"))
+	})
+}
+
+func TestGitCloneExecutorPinnedHashIgnoresShadowingRef(t *testing.T) {
+	t.Parallel()
+	remoteDir := t.TempDir()
+	require.NoError(t, gitCmd("init", "--bare", "--initial-branch=main", remoteDir))
+	workDir := t.TempDir()
+	require.NoError(t, gitCmd("clone", remoteDir, workDir))
+	require.NoError(t, gitCmd("-C", workDir, "checkout", "-b", "main"))
+	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "pinned"))
+	pinned := gitRevParse(t, workDir, "HEAD")
+	require.NoError(t, gitCmd("-C", workDir, "commit", "--allow-empty", "-m", "decoy"))
+	require.NoError(t, gitCmd("-C", workDir, "tag", pinned))
+	require.NoError(t, gitCmd("-C", workDir, "push", "-u", "origin", "main"))
+	require.NoError(t, gitCmd("-C", workDir, "push", "origin", pinned))
+
+	for name, depth := range map[string]int{"shallow": 1, "full clone": 0} {
+		t.Run(name, func(t *testing.T) {
+			dir := t.TempDir()
+			require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+				URL: remoteDir, Ref: pinned, Dir: dir, Depth: depth,
+			})(t.Context()))
+			assert.Equal(t, pinned, gitRevParse(t, dir, "HEAD"))
+		})
+	}
 }
 
 func gitRevParse(t *testing.T, dir, rev string) string {
@@ -610,12 +678,51 @@ func TestAcquireCloneLock(t *testing.T) {
 			t.Fatal("acquire on a different directory must not block")
 		}
 	})
+}
 
-	t.Run("same directory reuses the same mutex", func(t *testing.T) {
+// An unresponsive remote must not pin a job: the refresh has to be interruptible.
-		dir := t.TempDir()
+func TestNewGitCloneExecutorFetchSendsTokenAndHonoursContext(t *testing.T) {
+	t.Parallel()
+	block := make(chan struct{})
+	reached := make(chan struct{})
+	var once sync.Once
+	var authorization string
+	server := httptest.NewServer(http.HandlerFunc(func(_ http.ResponseWriter, request *http.Request) {
+		once.Do(func() {
+			authorization = request.Header.Get("Authorization")
+			close(reached)
+		})
+		<-block
+	}))
+	t.Cleanup(func() {
+		close(block)
+		server.Close()
+	})
+
+	dir := filepath.Join(t.TempDir(), "cached-action")
+	require.NoError(t, gitCmd("init", dir))
+	require.NoError(t, gitCmd("-C", dir, "remote", "add", "origin", server.URL))
+
+	ctx, cancel := context.WithCancel(t.Context())
+	defer cancel()
 
-		v1, _ := cloneLocks.LoadOrStore(dir, &sync.Mutex{})
+	done := make(chan error, 1)
-		v2, _ := cloneLocks.LoadOrStore(dir, &sync.Mutex{})
+	go func() {
-		require.Same(t, v1, v2)
+		done <- NewGitCloneExecutor(NewGitCloneExecutorInput{URL: server.URL, Ref: "main", Dir: dir, Token: "secret"})(ctx)
-	})
+	}()
+
+	select {
+	case <-reached:
+	case <-time.After(10 * time.Second):
+		t.Fatal("the executor never reached the remote")
+	}
+	assert.Equal(t, "Basic "+base64.StdEncoding.EncodeToString([]byte("token:secret")), authorization)
+	cancel()
+
+	select {
+	case err := <-done:
+		require.Error(t, err)
+	case <-time.After(10 * time.Second):
+		t.Fatal("fetch ignored context cancellation")
+	}
 }

act/common/line_writer.go

@@ -46,14 +46,14 @@ func (lw *lineWriter) Write(p []byte) (n int, err error) {
 		line, err := pBuf.ReadString('\n')
 		w, _ := lw.buffer.WriteString(line)
 		written += w
-		if err == nil {
+		if err != nil {
-			lw.handleLine(lw.buffer.String())
+			if err == io.EOF {
-			lw.buffer.Reset()
-		} else if err == io.EOF {
 				break
-		} else {
+			}
 			return written, err
 		}
+		lw.handleLine(lw.buffer.String())
+		lw.buffer.Reset()
 	}
 
 	return written, nil

act/common/outbound_ip.go

@@ -10,6 +10,11 @@ import (
 	"strings"
 )
 
+type interfaceIP struct {
+	net.IP
+	net.Interface
+}
+
 // GetOutboundIP returns an outbound IP address of this machine.
 // It tries to access the internet and returns the local IP address of the connection.
 // If the machine cannot access the internet, it returns a preferred IP address from network interfaces.
@@ -19,16 +24,14 @@ func GetOutboundIP() net.IP {
 	conn, err := net.Dial("udp", "8.8.8.8:80")
 	if err == nil {
 		defer conn.Close()
-		return conn.LocalAddr().(*net.UDPAddr).IP
+		if addr, ok := conn.LocalAddr().(*net.UDPAddr); ok {
+			return addr.IP
+		}
 	}
 
 	// So the machine cannot access the internet. Pick an IP address from network interfaces.
 	if ifs, err := net.Interfaces(); err == nil {
-		type IP struct {
+		var ips []interfaceIP
-			net.IP
-			net.Interface
-		}
-		var ips []IP
 		for _, i := range ifs {
 			if addrs, err := i.Addrs(); err == nil {
 				for _, addr := range addrs {
@@ -40,12 +43,21 @@ func GetOutboundIP() net.IP {
 						ip = v.IP
 					}
 					if ip.IsGlobalUnicast() {
-						ips = append(ips, IP{ip, i})
+						ips = append(ips, interfaceIP{ip, i})
 					}
 				}
 			}
 		}
-		if len(ips) > 1 {
+		return preferredIP(ips)
+	}
+
+	return nil
+}
+
+func preferredIP(ips []interfaceIP) net.IP {
+	if len(ips) == 0 {
+		return nil
+	}
 	sort.Slice(ips, func(i, j int) bool {
 		ifi := ips[i].Interface
 		ifj := ips[j].Interface
@@ -72,8 +84,4 @@ func GetOutboundIP() net.IP {
 		return ipi.String() < ipj.String()
 	})
 	return ips[0].IP
-		}
-	}
-
-	return nil
 }

act/common/outbound_ip_test.go

@@ -0,0 +1,20 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package common
+
+import (
+	"net"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestPreferredIP(t *testing.T) {
+	assert.Nil(t, preferredIP(nil))
+	assert.Equal(t, net.ParseIP("2001:db8::1"), preferredIP([]interfaceIP{{IP: net.ParseIP("2001:db8::1")}}))
+	assert.Equal(t, net.ParseIP("2001:db8::1"), preferredIP([]interfaceIP{
+		{IP: net.ParseIP("192.0.2.1"), Name: "wlan0"},
+		{IP: net.ParseIP("2001:db8::1"), Name: "eth0"},
+	}))
+}

act/container/container_types.go

@@ -6,12 +6,18 @@ package container
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"io"
+	"path"
+	"strings"
+	"sync"
+	"sync/atomic"
 
 	"gitea.com/gitea/runner/act/common"
 
 	"github.com/docker/go-connections/nat"
+	"github.com/moby/moby/api/types/container"
 )
 
 // ExitCodeError reports a non-zero process exit code from a container command.
@@ -39,7 +45,8 @@ type NewContainerInput struct {
 	Privileged      bool
 	UsernsMode      string
 	Platform        string
-	Options        string
+	RunnerOptions   string // container options the runner was configured with, trusted
+	WorkflowOptions string // container options the workflow asked for, untrusted
 	NetworkAliases  []string
 	ExposedPorts    nat.PortSet
 	PortBindings    nat.PortMap
@@ -57,14 +64,72 @@ type FileEntry struct {
 	Body string
 }
 
+// Container and healthcheck states, as plain strings so a caller of Info needs no docker
+// SDK of its own.
+const (
+	StateRunning = string(container.StateRunning)
+
+	HealthNone      = string(container.NoHealthcheck)
+	HealthStarting  = string(container.Starting)
+	HealthHealthy   = string(container.Healthy)
+	HealthUnhealthy = string(container.Unhealthy)
+)
+
+// marks what a runner creates for its jobs, telling its leftovers from another runner's
+const runnerUUIDLabel = "com.gitea.runner.uuid"
+
+// ErrContainerNotFound reports a container the daemon no longer knows. Its text is a
+// fragment, missingContainerError composes it into the message every operation shares.
+var ErrContainerNotFound = errors.New("does not exist")
+
+// DockerProxy is a job's docker socket, fronting the daemon's for the job's lifetime.
+type DockerProxy struct {
+	Socket    string
+	close     func(context.Context) error
+	closeOnce sync.Once
+	closeErr  error
+	mounts    atomic.Value
+}
+
+func (p *DockerProxy) SetMounts(mounts map[string]string) {
+	p.mounts.Store(mounts)
+}
+
+func (p *DockerProxy) Close(ctx context.Context) error {
+	p.closeOnce.Do(func() { p.closeErr = p.close(ctx) })
+	return p.closeErr
+}
+
+// Info is a snapshot of a container, as of one inspect.
+type Info struct {
+	ID       string
+	State    string // the docker container state: "created", "running", "exited", ...
+	ExitCode int
+	Health   string // one of the Health* constants
+	// HealthOutput is the last healthcheck probe's output.
+	HealthOutput string
+	Ports        map[string]string // container port ("5432") to the host port it is published on
+	Mounts       map[string]string // container path to its source on the daemon
+}
+
+// DaemonPath maps target to its daemon path through the nearest mount with a source, empty without one.
+func (info *Info) DaemonPath(target string) string {
+	for dir := target; dir != "/" && dir != "."; dir = path.Dir(dir) {
+		if source := info.Mounts[dir]; source != "" {
+			return path.Join(source, strings.TrimPrefix(target, dir))
+		}
+	}
+	return ""
+}
+
 // Container for managing docker run containers
 type Container interface {
 	Create(capAdd, capDrop []string) common.Executor
-	ConnectToNetwork(name string) common.Executor
 	Copy(destPath string, files ...*FileEntry) common.Executor
-	CopyTarStream(ctx context.Context, destPath string, tarStream io.Reader) error
+	CopyDir(destPath, srcPath string, useGitIgnore, skipGitDir bool) common.Executor
-	CopyDir(destPath, srcPath string, useGitIgnore bool) common.Executor
 	GetContainerArchive(ctx context.Context, srcPath string) (io.ReadCloser, error)
+	Inspect(ctx context.Context) (*Info, error)
+	DumpLogs(ctx context.Context) error
 	Pull(forcePull bool) common.Executor
 	Start(attach bool) common.Executor
 	Exec(command []string, env map[string]string, user, workdir string) common.Executor
@@ -82,12 +147,14 @@ type NewDockerBuildExecutorInput struct {
 	BuildContext io.Reader
 	ImageTag     string
 	Platform     string
+	BuildArgs    map[string]*string
 }
 
 // NewDockerNetworkCreateExecutorInput the input for the NewDockerNetworkCreateExecutor function
 type NewDockerNetworkCreateExecutorInput struct {
 	EnableIPv4 *bool
 	EnableIPv6 *bool
+	RunnerUUID string
 }
 
 // NewDockerPullExecutorInput the input for the NewDockerPullExecutor function

act/container/docker_build.go

@@ -49,6 +49,7 @@ func NewDockerBuildExecutor(input NewDockerBuildExecutorInput) common.Executor {
 			Remove:      true,
 			AuthConfigs: LoadDockerAuthConfigs(ctx),
 			Dockerfile:  input.Dockerfile,
+			BuildArgs:   input.BuildArgs,
 		}
 		platform, err := parsePlatform(input.Platform)
 		if err != nil {

act/container/docker_cli.go

@@ -4,8 +4,11 @@
 
 //go:build !(WITHOUT_DOCKER || !(linux || darwin || windows || netbsd))
 
-// This file is exact copy of https://github.com/docker/cli/blob/9a471180cb7d39c236d090399a9d362c3f5a8ebd/cli/command/container/opts.go
+// This file is exact copy of https://github.com/docker/cli/blob/4a63305d74332de5ceba7fcbccbc3cbb7412f5ba/cli/command/container/opts.go with:
-// appended with license information.
+// * appended with license information
+// * regexp and loader.ParseVolume in place of the import-restricted internal/lazyregexp and internal/volumespec
+// * jsontext in place of json.Compact for the seccomp profile
+// * invalidParameter from the package's errors.go, and convertPortSet/convertPortMap for the callers in docker_run.go
 //
 // docker/cli is licensed under the Apache License, Version 2.0.
 // See DOCKER_LICENSE for the full license text.
@@ -15,8 +18,7 @@
 package container
 
 import (
-	"bytes"
+	"encoding/json/jsontext"
-	"encoding/json"
 	"errors"
 	"fmt"
 	"net"
@@ -30,6 +32,7 @@ import (
 	"strings"
 	"time"
 
+	cerrdefs "github.com/containerd/errdefs"
 	"github.com/docker/cli/cli/compose/loader"
 	"github.com/docker/cli/opts"
 	"github.com/docker/go-connections/nat"
@@ -150,6 +153,7 @@ type containerOptions struct {
 	runtime             string
 	autoRemove          bool
 	init                bool
+	umask               opts.UmaskOpt
 	annotations         *opts.MapOpts
 
 	Image string
@@ -324,6 +328,9 @@ func addFlags(flags *pflag.FlagSet) *containerOptions {
 	flags.Var(&copts.shmSize, "shm-size", "Size of /dev/shm")
 	flags.StringVar(&copts.utsMode, "uts", "", "UTS namespace to use")
 	flags.StringVar(&copts.runtime, "runtime", "", "Runtime to use for this container")
+	flags.Var(&copts.umask, "umask", "Set umask for the container")
+	flags.SetAnnotation("umask", "version", []string{"1.56"})
+	flags.SetAnnotation("umask", "ostype", []string{"linux"})
 
 	flags.BoolVar(&copts.init, "init", false, "Run an init inside the container that forwards signals and reaps processes")
 	flags.SetAnnotation("init", "version", []string{"1.25"})
@@ -348,7 +355,7 @@ type containerConfig struct {
 // parse parses the args for the specified command and generates a Config,
 // a HostConfig and returns them with the specified command.
 // If the specified args are not valid, it will return an error.
-func parse(flags *pflag.FlagSet, copts *containerOptions, serverOS string) (*containerConfig, error) { //nolint:gocyclo // verbatim copy from docker/cli
+func parse(flags *pflag.FlagSet, copts *containerOptions, serverOS string) (*containerConfig, error) { //nolint:gocyclo,unparam // verbatim copy from docker/cli
 	var (
 		attachStdin  = copts.attach.Get("stdin")
 		attachStdout = copts.attach.Get("stdout")
@@ -380,7 +387,7 @@ func parse(flags *pflag.FlagSet, copts *containerOptions, serverOS string) (*con
 	var binds []string
 	volumes := copts.volumes.GetMap()
 	// add any bind targets to the list of container volumes
-	for bind := range copts.volumes.GetMap() {
+	for bind := range volumes {
 		parsed, err := loader.ParseVolume(bind)
 		if err != nil {
 			return nil, err
@@ -515,13 +522,13 @@ func parse(flags *pflag.FlagSet, copts *containerOptions, serverOS string) (*con
 	// collect all the environment variables for the container
 	envVariables, err := opts.ReadKVEnvStrings(copts.envFile.GetSlice(), copts.env.GetSlice())
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("--env-file: %w", err)
 	}
 
 	// collect all the labels for the container
 	labels, err := opts.ReadKVStrings(copts.labelsFile.GetSlice(), copts.labels.GetSlice())
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("--label-file: %w", err)
 	}
 
 	pidMode := container.PidMode(copts.pidMode)
@@ -719,6 +726,7 @@ func parse(flags *pflag.FlagSet, copts *containerOptions, serverOS string) (*con
 		MaskedPaths:    maskedPaths,
 		ReadonlyPaths:  readonlyPaths,
 		Annotations:    copts.annotations.GetAll(),
+		Umask:          copts.umask.Value(),
 	}
 
 	if copts.autoRemove && !hostConfig.RestartPolicy.IsNone() {
@@ -956,11 +964,11 @@ func parseSecurityOpts(securityOpts []string) ([]string, error) {
 				if err != nil {
 					return securityOpts, fmt.Errorf("opening seccomp profile (%s) failed: %w", v, err)
 				}
-				var b bytes.Buffer
+				profile := jsontext.Value(f)
-				if err := json.Compact(&b, f); err != nil {
+				if err := profile.Compact(); err != nil {
 					return securityOpts, fmt.Errorf("compacting json for seccomp profile (%s) failed: %w", v, err)
 				}
-				securityOpts[key] = "seccomp=" + b.String()
+				securityOpts[key] = "seccomp=" + string(profile)
 			}
 		}
 	}
@@ -1164,16 +1172,17 @@ func toNetipAddrSlice(ips []string) []netip.Addr {
 }
 
 // invalidParameter wraps an error to indicate it was caused by invalid input.
-// This is a local replacement for docker/docker/errdefs.InvalidParameter.
+// This is a copy of docker/cli's cli/command/container/errors.go, which is not importable.
-type invalidParameterError struct{ error }
+type invalidParameterErr struct{ error }
 
-func (e invalidParameterError) InvalidParameter() {}
+func (invalidParameterErr) InvalidParameter() {}
+func (e invalidParameterErr) Unwrap() error   { return e.error }
 
 func invalidParameter(err error) error {
-	if err == nil {
+	if err == nil || cerrdefs.IsInvalidArgument(err) {
-		return nil
+		return err
 	}
-	return invalidParameterError{err}
+	return invalidParameterErr{err}
 }
 
 func convertPortSet(ports nat.PortSet) (network.PortSet, error) {

act/container/docker_create_flags.go

@@ -10,7 +10,9 @@ import (
 	"fmt"
 	"io"
 	"slices"
+	"strings"
 
+	"github.com/docker/cli/opts"
 	"github.com/kballard/go-shellquote"
 	"github.com/spf13/pflag"
 )
@@ -51,15 +53,16 @@ func parseContainerOptions(options string) (*pflag.FlagSet, *containerOptions, *
 	flags := pflag.NewFlagSet("container_flags", pflag.ContinueOnError)
 	flags.SetOutput(io.Discard)
 	copts := addFlags(flags)
+	copts.env = opts.NewListOpts(validateEnv) // addFlags registered this field's address, so the swap takes effect
 	cf := registerCreateFlags(flags)
 
 	args, err := shellquote.Split(options)
 	if err != nil {
-		return flags, copts, cf, fmt.Errorf("Cannot split container options: '%s': '%w'", options, err)
+		return flags, copts, cf, fmt.Errorf("cannot split container options: '%s': '%w'", options, err)
 	}
 
 	if err := flags.Parse(args); err != nil {
-		return flags, copts, cf, fmt.Errorf("Cannot parse container options: '%s': '%w'", options, err)
+		return flags, copts, cf, fmt.Errorf("cannot parse container options: '%s': '%w'", options, err)
 	}
 
 	return flags, copts, cf, nil
@@ -73,6 +76,30 @@ func createFlagsFromOptions(options string) *createFlags {
 	return cf
 }
 
+// validateEnv is opts.ValidateEnv without its lookup of a bare name in the runner's environment.
+func validateEnv(val string) (string, error) {
+	if name, _, _ := strings.Cut(val, "="); name == "" {
+		return "", errors.New("invalid environment variable: " + val)
+	}
+	return val, nil
+}
+
+// rejectHostReadingOptions refuses the flags naming files that are read here, on the
+// runner, rather than in the container.
+func rejectHostReadingOptions(options string) error {
+	flags, _, _, err := parseContainerOptions(options)
+	if err != nil {
+		return err
+	}
+
+	for _, name := range []string{"env-file", "label-file"} {
+		if flags.Changed(name) {
+			return fmt.Errorf("container option --%s reads files from the runner and is not allowed in a workflow", name)
+		}
+	}
+	return nil
+}
+
 func (cf *createFlags) validate() error {
 	if !slices.Contains(pullPolicies, cf.pull) {
 		return fmt.Errorf("invalid --pull option %q: must be one of %q", cf.pull, pullPolicies)

act/container/docker_create_flags_test.go

@@ -50,12 +50,13 @@ func TestCreateFlagsValidate(t *testing.T) {
 }
 
 func TestNewContainerAppliesCreateFlags(t *testing.T) {
-	input := &NewContainerInput{Platform: "linux/amd64", Options: "--platform linux/arm64 --pull never"}
+	input := &NewContainerInput{Platform: "linux/amd64", RunnerOptions: "--pull never", WorkflowOptions: "--platform linux/arm64"}
-	cr := NewContainer(input).(*containerReference)
+	cr, ok := NewContainer(input).(*containerReference)
+	require.True(t, ok)
 	assert.Equal(t, "linux/arm64", input.Platform)
 	assert.Equal(t, pullPolicyNever, cr.pullPolicy)
 
-	kept := &NewContainerInput{Platform: "linux/amd64", Options: "--privileged"}
+	kept := &NewContainerInput{Platform: "linux/amd64", RunnerOptions: "--privileged"}
 	NewContainer(kept)
 	assert.Equal(t, "linux/amd64", kept.Platform)
 }

act/container/docker_images.go

@@ -46,14 +46,13 @@ func RemoveImage(ctx context.Context, imageName string, force, pruneChildren boo
 	}
 	defer cli.Close()
 
-	inspectImage, err := cli.ImageInspect(ctx, imageName)
+	if _, err := cli.ImageInspect(ctx, imageName); cerrdefs.IsNotFound(err) {
-	if cerrdefs.IsNotFound(err) {
 		return false, nil
 	} else if err != nil {
 		return false, err
 	}
 
-	if _, err = cli.ImageRemove(ctx, inspectImage.ID, client.ImageRemoveOptions{
+	if _, err = cli.ImageRemove(ctx, imageName, client.ImageRemoveOptions{
 		Force:         force,
 		PruneChildren: pruneChildren,
 	}); err != nil {

act/container/docker_logger.go

@@ -8,7 +8,7 @@ package container
 
 import (
 	"bufio"
-	"encoding/json"
+	"encoding/json/v2"
 	"errors"
 	"io"
 
@@ -20,8 +20,8 @@ type dockerMessage struct {
 	Stream      string `json:"stream"`
 	Error       string `json:"error"`
 	ErrorDetail struct {
-		Message string
+		Message string `json:"message"`
-	}
+	} `json:"errorDetail"`
 	Status   string `json:"status"`
 	Progress string `json:"progress"`
 }
@@ -57,18 +57,19 @@ func logDockerResponse(logger logrus.FieldLogger, dockerResponse io.ReadCloser,
 
 		if msg.ErrorDetail.Message != "" {
 			writeLog(logger, isError, "%s", msg.ErrorDetail.Message)
-			return errors.New(msg.Error)
+			return errors.New(msg.ErrorDetail.Message)
 		}
 
-		if msg.Status != "" {
+		switch {
+		case msg.Status != "":
 			if msg.Progress != "" {
 				writeLog(logger, isError, "%s :: %s :: %s\n", msg.Status, msg.ID, msg.Progress)
 			} else {
 				writeLog(logger, isError, "%s :: %s\n", msg.Status, msg.ID)
 			}
-		} else if msg.Stream != "" {
+		case msg.Stream != "":
 			writeLog(logger, isError, "%s", msg.Stream)
-		} else {
+		default:
 			writeLog(logger, false, "Unable to handle line: %s", string(line))
 		}
 	}

act/container/docker_network.go

@@ -8,12 +8,78 @@ package container
 
 import (
 	"context"
+	"errors"
+	"fmt"
+	"net/netip"
+	"strings"
+	"time"
 
 	"gitea.com/gitea/runner/act/common"
 
+	cerrdefs "github.com/containerd/errdefs"
+	"github.com/moby/moby/api/types/network"
 	"github.com/moby/moby/client"
 )
 
+const (
+	networkCreateAttempts   = 3
+	networkCreateRetryDelay = time.Second
+)
+
+// RemoveOrphanNetworks removes the networks this runner created for jobs whose teardown did
+// not get to them: the runner died with the job, the teardown timed out, or the network still
+// had an endpoint on it at the time. Each one holds a subnet of the daemon's address pool
+// until it is removed. Networks created after createdBefore are left alone, so a job starting
+// while this runs cannot lose the network it has created but not yet attached a container to.
+func RemoveOrphanNetworks(ctx context.Context, runnerUUID, detach string, createdBefore time.Time) error {
+	cli, err := GetDockerClient(ctx)
+	if err != nil {
+		return fmt.Errorf("failed to connect to the docker daemon: %w", err)
+	}
+	defer cli.Close()
+
+	return removeOrphanNetworks(ctx, cli, runnerUUID, detach, createdBefore)
+}
+
+func removeOrphanNetworks(ctx context.Context, cli client.APIClient, runnerUUID, detach string, createdBefore time.Time) error {
+	networks, err := cli.NetworkList(ctx, client.NetworkListOptions{
+		Filters: make(client.Filters).Add("label", runnerUUIDLabel+"="+runnerUUID),
+	})
+	if err != nil {
+		return err
+	}
+
+	var errs []error
+networks:
+	for _, n := range networks.Items {
+		result, err := cli.NetworkInspect(ctx, n.ID, client.NetworkInspectOptions{})
+		if err != nil {
+			errs = append(errs, fmt.Errorf("failed to inspect network %s: %w", n.Name, err))
+			continue
+		}
+		// the emptiness check, not the label, is what keeps a live job of another process
+		// sharing this registration safe
+		if len(result.Network.Containers) > 1 || result.Network.Created.After(createdBefore) {
+			continue
+		}
+		for containerID := range result.Network.Containers {
+			if detach == "" || !strings.HasPrefix(containerID, detach) {
+				continue networks
+			}
+			if err := disconnectNetwork(ctx, cli, n.ID, containerID); err != nil {
+				errs = append(errs, fmt.Errorf("failed to disconnect container from network %s: %w", n.Name, err))
+				continue networks
+			}
+		}
+		if _, err := cli.NetworkRemove(ctx, n.ID, client.NetworkRemoveOptions{}); err != nil {
+			errs = append(errs, fmt.Errorf("failed to remove network %s: %w", n.Name, err))
+			continue
+		}
+		common.Logger(ctx).Infof("removed docker network %s left behind by an earlier job", n.Name)
+	}
+	return errors.Join(errs...)
+}
+
 func NewDockerNetworkCreateExecutor(name string, opts NewDockerNetworkCreateExecutorInput) common.Executor {
 	return func(ctx context.Context) error {
 		cli, err := GetDockerClient(ctx)
@@ -21,12 +87,17 @@ func NewDockerNetworkCreateExecutor(name string, opts NewDockerNetworkCreateExec
 			return err
 		}
 		defer cli.Close()
+		return createDockerNetwork(ctx, cli, name, opts)
+	}
+}
 
+func createDockerNetwork(ctx context.Context, cli client.APIClient, name string, opts NewDockerNetworkCreateExecutorInput) error {
 	// Only create the network if it doesn't exist
 	networks, err := cli.NetworkList(ctx, client.NetworkListOptions{})
 	if err != nil {
 		return err
 	}
+	var options map[string]string
 	// For Gitea, reduce log noise
 	// common.Logger(ctx).Debugf("%v", networks)
 	for _, n := range networks.Items {
@@ -34,22 +105,97 @@ func NewDockerNetworkCreateExecutor(name string, opts NewDockerNetworkCreateExec
 			common.Logger(ctx).Debugf("Network %v exists", name)
 			return nil
 		}
+		// the daemon's mtu only reaches its default bridge, Docker's default is left out to keep default-network-opts
+		if mtu := n.Options["com.docker.network.driver.mtu"]; n.Name == network.NetworkBridge && mtu != "" && mtu != "1500" {
+			options = map[string]string{"com.docker.network.driver.mtu": mtu}
+		}
 	}
 
-		_, err = cli.NetworkCreate(ctx, name, client.NetworkCreateOptions{
+	for i := range networkCreateAttempts {
+		if i > 0 {
+			common.Logger(ctx).Infof("Waiting for a free docker address pool to create network %s", name)
+			select {
+			case <-ctx.Done():
+				return ctx.Err()
+			case <-time.After(time.Duration(i) * networkCreateRetryDelay):
+			}
+		}
+		if _, err = cli.NetworkCreate(ctx, name, client.NetworkCreateOptions{
 			Driver:     "bridge",
 			Scope:      "local",
 			EnableIPv4: opts.EnableIPv4,
 			EnableIPv6: opts.EnableIPv6,
-		})
+			Options:    options,
+			Labels:     runnerLabels(opts.RunnerUUID),
+		}); err == nil {
+			return nil
+		}
+		if !isAddressPoolExhausted(err) {
+			return err
+		}
+	}
+	return fmt.Errorf("docker has no address pool left for this job's network, lower runner.capacity or widen default-address-pools in the docker daemon config: %w", err)
+}
+
+func runnerLabels(runnerUUID string) map[string]string {
+	if runnerUUID == "" {
+		return nil
+	}
+	return map[string]string{runnerUUIDLabel: runnerUUID}
+}
+
+// The daemon reports this as a plain invalid-parameter error, the same kind it uses for every
+// malformed request, so the message is the only discriminator.
+func isAddressPoolExhausted(err error) bool {
+	msg := err.Error()
+	return strings.Contains(msg, "all predefined address pools have been fully subnetted") ||
+		strings.Contains(msg, "could not find an available, non-overlapping IPv4 address pool among the defaults") // docker 24 and older
+}
+
+func NewDockerNetworkConnectExecutor(networkName, container string) common.Executor {
+	return func(ctx context.Context) error {
+		cli, err := GetDockerClient(ctx)
 		if err != nil {
 			return err
 		}
-
+		defer cli.Close()
+		if err := connectNetwork(ctx, cli, networkName, container); err != nil {
+			common.Logger(ctx).Warnf("cannot connect %s to network %s, so jobs on it cannot reach the cache: %v", container, networkName, err)
+		}
 		return nil
 	}
 }
 
+func connectNetwork(ctx context.Context, cli client.APIClient, networkName, container string) error {
+	_, err := cli.NetworkConnect(ctx, networkName, client.NetworkConnectOptions{
+		Container:      container,
+		EndpointConfig: &network.EndpointSettings{Aliases: []string{container}},
+	})
+	if err != nil && (strings.Contains(err.Error(), "already exists in network") || strings.Contains(err.Error(), "already connected")) {
+		return nil
+	}
+	return err
+}
+
+func NewDockerNetworkDisconnectExecutor(networkName, container string) common.Executor {
+	return func(ctx context.Context) error {
+		cli, err := GetDockerClient(ctx)
+		if err != nil {
+			return err
+		}
+		defer cli.Close()
+		return disconnectNetwork(ctx, cli, networkName, container)
+	}
+}
+
+func disconnectNetwork(ctx context.Context, cli client.APIClient, networkName, container string) error {
+	_, err := cli.NetworkDisconnect(ctx, networkName, client.NetworkDisconnectOptions{Container: container, Force: true})
+	if cerrdefs.IsNotFound(err) || (err != nil && strings.Contains(err.Error(), "is not connected")) {
+		return nil
+	}
+	return err
+}
+
 func NewDockerNetworkRemoveExecutor(name string) common.Executor {
 	return func(ctx context.Context) error {
 		cli, err := GetDockerClient(ctx)
@@ -66,6 +212,7 @@ func NewDockerNetworkRemoveExecutor(name string) common.Executor {
 		}
 		// For Gitea, reduce log noise
 		// common.Logger(ctx).Debugf("%v", networks)
+		var errs []error
 		for _, n := range networks.Items {
 			if n.Name == name {
 				result, err := cli.NetworkInspect(ctx, n.ID, client.NetworkInspectOptions{})
@@ -73,16 +220,63 @@ func NewDockerNetworkRemoveExecutor(name string) common.Executor {
 					return err
 				}
 
-				if len(result.Network.Containers) == 0 {
+				// it holds a subnet out of the daemon's pool until something reclaims it
-					if _, err = cli.NetworkRemove(ctx, n.ID, client.NetworkRemoveOptions{}); err != nil {
+				if len(result.Network.Containers) != 0 {
-						common.Logger(ctx).Debugf("%v", err)
+					common.Logger(ctx).Warnf("Refusing to remove network %s because it still has active endpoints, the idle cleanup reclaims it once they are gone", name)
+					continue
 				}
-				} else {
+				if _, err = cli.NetworkRemove(ctx, n.ID, client.NetworkRemoveOptions{}); err != nil {
-					common.Logger(ctx).Debugf("Refusing to remove network %v because it still has active endpoints", name)
+					errs = append(errs, fmt.Errorf("failed to remove network %s: %w", name, err))
 				}
 			}
 		}
 
-		return err
+		return errors.Join(errs...)
 	}
 }
+
+// IsolatedCacheContainer returns the container holding addr, when jobs on jobNetwork cannot reach it there.
+func IsolatedCacheContainer(ctx context.Context, addr netip.Addr, jobNetwork string) (string, error) {
+	cli, err := GetDockerClient(ctx)
+	if err != nil {
+		return "", err
+	}
+	defer cli.Close()
+	return isolatedCacheContainer(ctx, cli, addr, jobNetwork)
+}
+
+func isolatedCacheContainer(ctx context.Context, cli client.APIClient, addr netip.Addr, jobNetwork string) (string, error) {
+	if jobNetwork == "host" || strings.HasPrefix(jobNetwork, "container:") {
+		return "", nil
+	}
+	containers, err := cli.ContainerList(ctx, client.ContainerListOptions{})
+	if err != nil {
+		return "", err
+	}
+	var holderID, holderContainer string
+	for _, summary := range containers.Items {
+		if summary.NetworkSettings == nil {
+			continue
+		}
+		for _, endpoint := range summary.NetworkSettings.Networks {
+			if endpoint != nil && (endpoint.IPAddress == addr || endpoint.GlobalIPv6Address == addr) {
+				holderID = endpoint.NetworkID
+				holderContainer = summary.ID[:12]
+			}
+		}
+	}
+	if holderID == "" {
+		return "", nil
+	}
+	holder, err := cli.NetworkInspect(ctx, holderID, client.NetworkInspectOptions{})
+	if err != nil || holder.Network.Driver != "bridge" {
+		return "", err
+	}
+	if jobNetwork != "" {
+		job, err := cli.NetworkInspect(ctx, jobNetwork, client.NetworkInspectOptions{})
+		if err != nil || job.Network.ID == holderID {
+			return "", err
+		}
+	}
+	return holderContainer, nil
+}

act/container/docker_network_test.go

@@ -0,0 +1,154 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package container
+
+import (
+	"context"
+	"errors"
+	"net/netip"
+	"testing"
+	"time"
+
+	cerrdefs "github.com/containerd/errdefs"
+	"github.com/moby/moby/api/types/container"
+	"github.com/moby/moby/api/types/network"
+	mobyclient "github.com/moby/moby/client"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestIsolatedCacheContainer(t *testing.T) {
+	ctx := context.Background()
+	cache, lan := netip.MustParseAddr("172.18.0.3"), netip.MustParseAddr("192.168.1.20")
+	client := &mockDockerClient{}
+	client.On("ContainerList", ctx, mobyclient.ContainerListOptions{}).Return(mobyclient.ContainerListResult{Items: []container.Summary{{}, {
+		ID: "0123456789abcdef",
+		NetworkSettings: &container.NetworkSettingsSummary{Networks: map[string]*network.EndpointSettings{
+			"compose": {NetworkID: "c0ffee", IPAddress: cache},
+			"lan":     {NetworkID: "beef", IPAddress: lan},
+		}},
+	}}}, nil)
+	for ref, inspected := range map[string]network.Network{
+		"c0ffee": {ID: "c0ffee", Name: "compose", Driver: "bridge"},
+		"beef":   {ID: "beef", Name: "lan", Driver: "macvlan"},
+		"c0f":    {ID: "c0ffee"},
+		"jobs":   {ID: "f00d"},
+	} {
+		client.On("NetworkInspect", ctx, ref, mobyclient.NetworkInspectOptions{}).
+			Return(mobyclient.NetworkInspectResult{Network: network.Inspect{Network: inspected}}, nil)
+	}
+	check := func(addr netip.Addr, jobNetwork, want string) {
+		got, err := isolatedCacheContainer(ctx, client, addr, jobNetwork)
+		require.NoError(t, err)
+		assert.Equal(t, want, got, jobNetwork)
+	}
+	check(cache, "", "0123456789ab")
+	check(cache, "jobs", "0123456789ab")
+	check(cache, "c0f", "")
+	check(cache, "host", "")
+	check(lan, "", "")
+	check(netip.MustParseAddr("10.0.0.5"), "", "")
+}
+
+func TestIsAddressPoolExhausted(t *testing.T) {
+	assert.True(t, isAddressPoolExhausted(cerrdefs.ErrInvalidArgument.WithMessage("Error response from daemon: all predefined address pools have been fully subnetted")))
+	assert.True(t, isAddressPoolExhausted(errors.New("could not find an available, non-overlapping IPv4 address pool among the defaults to assign to the network")))
+	assert.False(t, isAddressPoolExhausted(cerrdefs.ErrInvalidArgument.WithMessage("invalid subnet 10.0.0.0/8: it overlaps with an existing network")))
+}
+
+func TestCreateDockerNetworkInheritsDaemonMTU(t *testing.T) {
+	ctx := context.Background()
+	for _, tc := range []struct {
+		bridgeOptions, want map[string]string
+	}{
+		{
+			map[string]string{"com.docker.network.bridge.host_binding_ipv4": "127.0.0.1", "com.docker.network.driver.mtu": "1450"},
+			map[string]string{"com.docker.network.driver.mtu": "1450"},
+		},
+		{map[string]string{"com.docker.network.driver.mtu": "1500"}, nil},
+	} {
+		client := &mockDockerClient{}
+		client.On("NetworkList", ctx, mobyclient.NetworkListOptions{}).
+			Return(mobyclient.NetworkListResult{Items: []network.Summary{{Name: network.NetworkBridge, Options: tc.bridgeOptions}}}, nil)
+		client.On("NetworkCreate", ctx, "job-network", mobyclient.NetworkCreateOptions{Driver: "bridge", Scope: "local", Options: tc.want}).
+			Return(mobyclient.NetworkCreateResult{}, nil)
+		require.NoError(t, createDockerNetwork(ctx, client, "job-network", NewDockerNetworkCreateExecutorInput{}))
+		client.AssertExpectations(t)
+	}
+}
+
+func (m *mockDockerClient) NetworkConnect(ctx context.Context, id string, opts mobyclient.NetworkConnectOptions) (mobyclient.NetworkConnectResult, error) {
+	return mobyclient.NetworkConnectResult{}, m.Called(ctx, id, opts).Error(0)
+}
+
+func (m *mockDockerClient) NetworkDisconnect(ctx context.Context, id string, opts mobyclient.NetworkDisconnectOptions) (mobyclient.NetworkDisconnectResult, error) {
+	return mobyclient.NetworkDisconnectResult{}, m.Called(ctx, id, opts).Error(0)
+}
+
+func TestConnectAndDisconnectNetwork(t *testing.T) {
+	ctx := context.Background()
+	connect := func(answer error) error {
+		client := &mockDockerClient{}
+		client.On("NetworkConnect", ctx, "jobs", mobyclient.NetworkConnectOptions{
+			Container:      "0123456789ab",
+			EndpointConfig: &network.EndpointSettings{Aliases: []string{"0123456789ab"}},
+		}).Return(answer).Once()
+		return connectNetwork(ctx, client, "jobs", "0123456789ab")
+	}
+	require.NoError(t, connect(nil))
+	require.NoError(t, connect(errors.New("endpoint with name runner already exists in network jobs")))
+	require.NoError(t, connect(errors.New("container runner already connected to network jobs")))
+	require.ErrorIs(t, connect(context.Canceled), context.Canceled)
+
+	disconnect := func(answer error) error {
+		client := &mockDockerClient{}
+		client.On("NetworkDisconnect", ctx, "jobs", mobyclient.NetworkDisconnectOptions{
+			Container: "0123456789ab", Force: true,
+		}).Return(answer).Once()
+		return disconnectNetwork(ctx, client, "jobs", "0123456789ab")
+	}
+	require.NoError(t, disconnect(cerrdefs.ErrNotFound))
+	require.NoError(t, disconnect(errors.New("container runner is not connected to the network jobs")))
+	require.ErrorIs(t, disconnect(context.Canceled), context.Canceled)
+}
+
+// Of this runner's networks, only the ones nothing but its own cache container is attached to
+// and old enough to predate any job now starting are the runner's to reclaim. An unexpected
+// NetworkRemove fails the test on its own, since testify has no expectation to match it against.
+func TestRemoveOrphanNetworks(t *testing.T) {
+	ctx := context.Background()
+	cutoff := time.Date(2026, time.April, 29, 20, 0, 0, 0, time.UTC)
+	client := &mockDockerClient{}
+	client.On("NetworkList", ctx, mobyclient.NetworkListOptions{
+		Filters: make(mobyclient.Filters).Add("label", runnerUUIDLabel+"=runner-1"),
+	}).Return(mobyclient.NetworkListResult{Items: []network.Summary{
+		{ID: "orphan"},
+		{ID: "busy"},
+		{ID: "starting"},
+		{ID: "cached"},
+		{ID: "shared"},
+	}}, nil)
+	client.On("NetworkInspect", ctx, "orphan", mobyclient.NetworkInspectOptions{}).
+		Return(mobyclient.NetworkInspectResult{}, nil)
+	client.On("NetworkInspect", ctx, "busy", mobyclient.NetworkInspectOptions{}).
+		Return(mobyclient.NetworkInspectResult{Network: network.Inspect{Containers: map[string]network.EndpointResource{"c": {}}}}, nil)
+	client.On("NetworkInspect", ctx, "starting", mobyclient.NetworkInspectOptions{}).
+		Return(mobyclient.NetworkInspectResult{Network: network.Inspect{
+			Network:    network.Network{Created: cutoff.Add(time.Second)},
+			Containers: map[string]network.EndpointResource{"0123456789abcdef": {}},
+		}}, nil)
+	client.On("NetworkInspect", ctx, "cached", mobyclient.NetworkInspectOptions{}).
+		Return(mobyclient.NetworkInspectResult{Network: network.Inspect{Containers: map[string]network.EndpointResource{"0123456789abcdef": {}}}}, nil)
+	client.On("NetworkInspect", ctx, "shared", mobyclient.NetworkInspectOptions{}).
+		Return(mobyclient.NetworkInspectResult{Network: network.Inspect{Containers: map[string]network.EndpointResource{"0123456789abcdef": {}, "job": {}}}}, nil)
+	client.On("NetworkRemove", ctx, "orphan", mobyclient.NetworkRemoveOptions{}).
+		Return(mobyclient.NetworkRemoveResult{}, nil)
+	client.On("NetworkDisconnect", ctx, "cached", mobyclient.NetworkDisconnectOptions{Container: "0123456789abcdef", Force: true}).
+		Return(nil).Once()
+	client.On("NetworkRemove", ctx, "cached", mobyclient.NetworkRemoveOptions{}).
+		Return(mobyclient.NetworkRemoveResult{}, nil).Once()
+
+	require.NoError(t, removeOrphanNetworks(ctx, client, "runner-1", "0123456789ab", cutoff))
+	client.AssertExpectations(t)
+}

act/container/docker_proxy_unix.go

@@ -0,0 +1,48 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+//go:build !WITHOUT_DOCKER && (linux || darwin || netbsd)
+
+package container
+
+import (
+	"encoding/binary"
+	"errors"
+	"io/fs"
+	"math"
+	"os"
+	"path/filepath"
+	"syscall"
+
+	"golang.org/x/sys/unix"
+)
+
+func copyDockerSocketPermissions(daemonSocket, socket string, info os.FileInfo) error {
+	stat, ok := info.Sys().(*syscall.Stat_t)
+	if !ok {
+		return errors.New("docker socket ownership is unavailable")
+	}
+	groupErr := os.Chown(socket, int(stat.Uid), int(stat.Gid))
+	if groupErr != nil && (!errors.Is(groupErr, fs.ErrPermission) || int(stat.Uid) != os.Geteuid()) {
+		return groupErr
+	}
+	if err := os.Chown(filepath.Dir(socket), int(stat.Uid), -1); err != nil {
+		return err
+	}
+	mode := uint16(info.Mode().Perm())
+	if err := os.Chmod(socket, fs.FileMode(mode)); err != nil || groupErr == nil {
+		return err
+	}
+	if size, err := unix.Getxattr(daemonSocket, "system.posix_acl_access", nil); err == nil && size > 0 {
+		return groupErr
+	}
+	owner, group, other := mode>>6, mode>>3&7, mode&7
+	acl := binary.LittleEndian.AppendUint32(nil, 2)
+	for _, entry := range []struct {
+		tag, perm uint16
+		id        uint32
+	}{{1, owner, math.MaxUint32}, {4, other, math.MaxUint32}, {8, group, stat.Gid}, {16, group | other, math.MaxUint32}, {32, other, math.MaxUint32}} {
+		acl = binary.LittleEndian.AppendUint32(binary.LittleEndian.AppendUint16(binary.LittleEndian.AppendUint16(acl, entry.tag), entry.perm), entry.id)
+	}
+	return unix.Setxattr(socket, "system.posix_acl_access", acl, 0) // names the group a rootless runner cannot chown to, its own group keeps what others had
+}

act/container/docker_proxy_windows.go

@@ -0,0 +1,15 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+//go:build !WITHOUT_DOCKER
+
+package container
+
+import (
+	"errors"
+	"os"
+)
+
+func copyDockerSocketPermissions(_, _ string, _ os.FileInfo) error {
+	return errors.New("docker socket ownership cannot be preserved on Windows")
+}

act/container/docker_pull.go

@@ -30,22 +30,18 @@ func NewDockerPullExecutor(input NewDockerPullExecutorInput) common.Executor {
 			return nil
 		}
 
-		pull := input.ForcePull
+		// skip the pull when the image is already here: either none was forced, or a digest
-		if !pull {
+		// pins the content so a forced pull could only fetch the same bytes again
+		if !input.ForcePull || isPinnedImage(input.Image) {
 			imageExists, err := ImageExistsLocally(ctx, input.Image, input.Platform)
 			logger.Debugf("Image exists? %v", imageExists)
 			if err != nil {
 				return fmt.Errorf("unable to determine if image already exists for image '%s' (%s): %w", input.Image, input.Platform, err)
 			}
-
+			if imageExists {
-			if !imageExists {
-				pull = true
-			}
-		}
-
-		if !pull {
 				return nil
 			}
+		}
 
 		imageRef := cleanImage(ctx, input.Image)
 		logger.Debugf("pulling image '%v' (%s)", imageRef, input.Platform)
@@ -61,23 +57,33 @@ func NewDockerPullExecutor(input NewDockerPullExecutorInput) common.Executor {
 			return err
 		}
 
-		reader, err := cli.ImagePull(ctx, imageRef, imagePullOptions)
+		// the daemon reports a failure that happens after the first progress line in the
-
+		// stream rather than on the call itself, so both have to be checked
-		_ = logDockerResponse(logger, reader, err != nil)
+		pullOnce := func(opts client.ImagePullOptions) error {
+			reader, err := cli.ImagePull(ctx, imageRef, opts)
+			streamErr := logDockerResponse(logger, reader, err != nil)
 			if err != nil {
-			if imagePullOptions.RegistryAuth != "" && strings.Contains(err.Error(), "unauthorized") {
+				return err
+			}
+			return streamErr
+		}
+
+		err = pullOnce(imagePullOptions)
+		if err != nil && imagePullOptions.RegistryAuth != "" && strings.Contains(err.Error(), "unauthorized") {
 			logger.Errorf("pulling image '%v' (%s) failed with credentials %s retrying without them, please check for stale docker config files", imageRef, input.Platform, err.Error())
 			imagePullOptions.RegistryAuth = ""
-				reader, err = cli.ImagePull(ctx, imageRef, imagePullOptions)
+			err = pullOnce(imagePullOptions)
-
-				_ = logDockerResponse(logger, reader, err != nil)
-			}
-			if err != nil {
-				return fmt.Errorf("failed to pull image '%s' (%s): %w", imageRef, input.Platform, err)
-			}
 		}
+		if err == nil {
 			return nil
 		}
+		// a registry that is down should not fail a job whose image is already here
+		if exists, existsErr := ImageExistsLocally(ctx, input.Image, input.Platform); existsErr == nil && exists {
+			logger.Warnf("could not update image '%s' (%s), continuing with the local copy: %v", imageRef, input.Platform, err)
+			return nil
+		}
+		return fmt.Errorf("failed to pull image '%s' (%s): %w", imageRef, input.Platform, err)
+	}
 }
 
 func getImagePullOptions(ctx context.Context, input NewDockerPullExecutorInput) (client.ImagePullOptions, error) {
@@ -122,6 +128,15 @@ func getImagePullOptions(ctx context.Context, input NewDockerPullExecutorInput)
 	return imagePullOptions, nil
 }
 
+func isPinnedImage(image string) bool {
+	ref, err := reference.ParseAnyReference(image)
+	if err != nil {
+		return false
+	}
+	_, pinned := ref.(reference.Canonical)
+	return pinned
+}
+
 func cleanImage(ctx context.Context, imageName string) string {
 	ref, err := reference.ParseAnyReference(imageName)
 	if err != nil {