fix: bound blocking calls and stop failing silently (#1174)

Jobs occasionally go silent ([example](https://gitea.com/gitea/runner/actions/runs/805045/jobs/1055123)) mid-run and Gitea reaped them after `ZOMBIE_TASK_TIMEOUT`, with no error in the log. This contains a number of related fixes, all with full test coverage:

1. Bound every RPC to Gitea with a timeout, a stalled report otherwise parked logs and heartbeats for the whole job.
2. Cap `runner.fetch_timeout` at that ceiling.
3. Let only the daemon loop close its own channel, the race panicked the process.
4. Stop the job on any terminal server result, not just `RESULT_CANCELLED`.
5. Report that result instead of relabelling it as cancelled.
6. Log reporting failures once at each end of an outage instead of discarding them.
7. Clamp the acknowledged log index, a too-large ack panicked on a slice bound.
8. Stop reading server health from a `FetchTask` deadline, it marked the runner healthy and reset the error backoff on a timeout.
9. Return an error from the Docker version probe instead of a `logrus` panic.
10. Pass the context to go-git's fetch and pull.
11. Fail the clone when a refresh dies on a cancelled context.
12. Set `terminationGracePeriodSeconds` in the Kubernetes examples.

Also contains a deprecation fix for goreleaser.

Reviewed-on: https://gitea.com/gitea/runner/pulls/1174
Reviewed-by: bircni <[email protected]>
Co-authored-by: silverwind <[email protected]>

.gitea/workflows/pull-pr-title.yml

@@ -18,8 +18,8 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 5
     steps:
-    - uses: actions/checkout@v6
+    - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
-    - uses: actions/setup-node@v6
+    - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7
       with:
         node-version: 24
     - run: make lint-pr-title

.gitea/workflows/release-nightly.yml

@@ -17,24 +17,35 @@ jobs:
   goreleaser:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
         with:
           fetch-depth: 0
-      - uses: actions/setup-go@v6
+      # 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. Fail here instead, before anything
+      # is built or published, if the R2 secrets are missing.
+      - name: check R2 configuration
+        run: sh scripts/upload-r2.sh --check-config
+        env:
+          R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
+          R2_BUCKET: ${{ secrets.R2_BUCKET }}
+          R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
+          R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
+      - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
         with:
           go-version-file: "go.mod"
+          check-latest: true
       - name: goreleaser
-        uses: goreleaser/goreleaser-action@v7
+        uses: goreleaser/goreleaser-action@f06c13b6b1a9625abc9e6e439d9c05a8f2190e94 # v7
         with:
           distribution: goreleaser-pro
           args: release --nightly
         env:
           GORELEASER_KEY: ${{ secrets.GORELEASER_KEY }}
-          AWS_REGION: ${{ secrets.AWS_REGION }}
+          R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
-          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_KEY_ID }}
+          R2_BUCKET: ${{ secrets.R2_BUCKET }}
-          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
+          R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
-          S3_REGION: ${{ secrets.AWS_REGION }}
+          R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
-          S3_BUCKET: ${{ secrets.AWS_BUCKET }}
           GORELEASER_FORCE_TOKEN: "gitea"
           GITEA_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
@@ -43,27 +54,33 @@ jobs:
     strategy:
       matrix:
         variant:
+          # The basic image is built from source and can target any arch the
+          # toolchain supports. The dind variants are limited to the arches the
+          # docker:dind base image publishes.
           - target: basic
             tag_suffix: ""
+            platforms: linux/amd64,linux/arm64,linux/riscv64,linux/s390x
           - target: dind
             tag_suffix: "-dind"
+            platforms: linux/amd64,linux/arm64
           - target: dind-rootless
             tag_suffix: "-dind-rootless"
+            platforms: linux/amd64,linux/arm64
 
     steps:
       - name: Checkout
-        uses: actions/checkout@v6
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
         with:
           fetch-depth: 0 # all history for all branches and tags
 
       - name: Set up QEMU
-        uses: docker/setup-qemu-action@v4
+        uses: docker/setup-qemu-action@96fe6ef7f33517b61c61be40b68a1882f3264fb8 # v4
 
       - name: Set up Docker BuildX
-        uses: docker/setup-buildx-action@v4
+        uses: docker/setup-buildx-action@bb05f3f5519dd87d3ba754cc423b652a5edd6d2c # v4
 
       - name: Login to DockerHub
-        uses: docker/login-action@v4
+        uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4
         with:
           username: ${{ secrets.DOCKER_USERNAME }}
           password: ${{ secrets.DOCKER_PASSWORD }}
@@ -77,14 +94,12 @@ jobs:
           echo REPO_VERSION=$(git describe --tags --always | sed 's/-/+/' | sed 's/^v//') >> $GITHUB_OUTPUT
 
       - name: Build and push
-        uses: docker/build-push-action@v7
+        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7
         with:
           context: .
           file: ./Dockerfile
           target: ${{ matrix.variant.target }}
-          platforms: |
+          platforms: ${{ matrix.variant.platforms }}
-            linux/amd64
-            linux/arm64
           push: true
           tags: |
             ${{ env.DOCKER_ORG }}/runner:nightly${{ matrix.variant.tag_suffix }}

.gitea/workflows/release-tag.yml

@@ -9,31 +9,42 @@ jobs:
   goreleaser:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
         with:
           fetch-depth: 0 # all history for all branches and tags
-      - uses: actions/setup-go@v6
+      # 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. Fail here instead, before anything
+      # is built or published, if the R2 secrets are missing.
+      - name: check R2 configuration
+        run: sh scripts/upload-r2.sh --check-config
+        env:
+          R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
+          R2_BUCKET: ${{ secrets.R2_BUCKET }}
+          R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
+          R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
+      - 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@v7
+        uses: crazy-max/ghaction-import-gpg@2dc316deee8e90f13e1a351ab510b4d5bc0c82cd # v7
         with:
           gpg_private_key: ${{ secrets.GPG_PRIVATE_KEY }}
           passphrase: ${{ secrets.PASSPHRASE }}
           fingerprint: CC64B1DB67ABBEECAB24B6455FC346329753F4B0
       - name: goreleaser
-        uses: goreleaser/goreleaser-action@v7
+        uses: goreleaser/goreleaser-action@f06c13b6b1a9625abc9e6e439d9c05a8f2190e94 # v7
         with:
           distribution: goreleaser-pro
           args: release
         env:
           GORELEASER_KEY: ${{ secrets.GORELEASER_KEY }}
-          AWS_REGION: ${{ secrets.AWS_REGION }}
+          R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
-          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_KEY_ID }}
+          R2_BUCKET: ${{ secrets.R2_BUCKET }}
-          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
+          R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
-          S3_REGION: ${{ secrets.AWS_REGION }}
+          R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
-          S3_BUCKET: ${{ secrets.AWS_BUCKET }}
           GORELEASER_FORCE_TOKEN: "gitea"
           GITEA_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           GPG_FINGERPRINT: ${{ steps.import_gpg.outputs.fingerprint }}
@@ -42,12 +53,18 @@ jobs:
     strategy:
       matrix:
         variant:
+          # The basic image is built from source and can target any arch the
+          # toolchain supports. The dind variants are limited to the arches the
+          # docker:dind base image publishes.
           - target: basic
             tag_suffix: ""
+            platforms: linux/amd64,linux/arm64,linux/riscv64,linux/s390x
           - target: dind
             tag_suffix: "-dind"
+            platforms: linux/amd64,linux/arm64
           - target: dind-rootless
             tag_suffix: "-dind-rootless"
+            platforms: linux/amd64,linux/arm64
     container:
       image: catthehacker/ubuntu:act-latest
     env:
@@ -55,25 +72,25 @@ jobs:
       DOCKER_LATEST: latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v6
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
         with:
           fetch-depth: 0 # all history for all branches and tags
 
       - name: Set up QEMU
-        uses: docker/setup-qemu-action@v4
+        uses: docker/setup-qemu-action@96fe6ef7f33517b61c61be40b68a1882f3264fb8 # v4
 
       - name: Set up Docker BuildX
-        uses: docker/setup-buildx-action@v4
+        uses: docker/setup-buildx-action@bb05f3f5519dd87d3ba754cc423b652a5edd6d2c # v4
 
       - name: Login to DockerHub
-        uses: docker/login-action@v4
+        uses: docker/login-action@dbcb813823bdd20940b903addbd779551569679f # v4
         with:
           username: ${{ secrets.DOCKER_USERNAME }}
           password: ${{ secrets.DOCKER_PASSWORD }}
 
       - name: "Docker meta"
         id: docker_meta
-        uses: docker/metadata-action@v6
+        uses: docker/metadata-action@dc802804100637a589fabce1cb79ff13a1411302 # v6
         with:
           images: |
             ${{ env.DOCKER_ORG }}/runner
@@ -86,14 +103,12 @@ jobs:
             suffix=${{ matrix.variant.tag_suffix }},onlatest=true
 
       - name: Build and push
-        uses: docker/build-push-action@v7
+        uses: docker/build-push-action@53b7df96c91f9c12dcc8a07bcb9ccacbed38856a # v7
         with:
           context: .
           file: ./Dockerfile
           target: ${{ matrix.variant.target }}
-          platforms: |
+          platforms: ${{ matrix.variant.platforms }}
-            linux/amd64
-            linux/arm64
           push: true
           tags: ${{ steps.docker_meta.outputs.tags }}
           build-args: |

.gitea/workflows/test.yml

@@ -17,10 +17,11 @@ jobs:
       # to ~/.docker with the stale credentials.
       DOCKER_CONFIG: /tmp/docker-noauth
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
-      - uses: actions/setup-go@v6
+      - 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`;
@@ -33,6 +34,8 @@ jobs:
           done
       - name: lint
         run: make lint
+      - name: checks
+        run: make checks
       - name: build
         run: make build
       - name: test
@@ -42,3 +45,7 @@ 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: coverage report
+        run: |
+          make coverage-report
+          cat .tmp/coverage.md >> "$GITHUB_STEP_SUMMARY"

.gitignore

@@ -2,7 +2,9 @@
 .env
 !/act/runner/testdata/secrets/.env
 .runner
+.runner.lock
 coverage.txt
+.tmp/
 /config.yaml
 
 # Jetbrains
@@ -12,3 +14,4 @@ coverage.txt
 __debug_bin
 # gorelease binary folder
 /dist
+.DS_Store

.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
@@ -102,6 +99,9 @@ linters:
       - linters:
           - forbidigo
         path: cmd
+      - linters:
+          - forcetypeassert
+        path: _test\.go
 issues:
   max-issues-per-linter: 0
   max-same-issues: 0

.goreleaser.yaml

@@ -83,18 +83,37 @@ 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.
+#
+# This publisher fires 109 times for 73 distinct keys because
+# goreleaser's release pipe already registers `release.extra_files`
+# as UploadableFile artifacts, and `internal/exec`'s filterArtifacts
+# appends this block's own extra_files with no de-duplication. It
+# can't be globbed away, since gobwas/glob (via goreleaser/fileglob)
+# has no substring-exclusion matcher. It's harmless: PUT is
+# idempotent, and the `./**.xz` glob below is kept deliberately so
+# this publisher declares its own complete file set rather than
+# implicitly depending on the `release:` block's globs.
+publishers:
+  - name: cloudflare-r2
+    checksum: true
     extra_files:
       - glob: ./**.xz
       - glob: ./**.sha256
+    cmd: sh scripts/upload-r2.sh {{ abs .ArtifactPath }} gitea-runner/{{ .Version }}/{{ .ArtifactName }}
+    env:
+      - R2_ENDPOINT={{ index .Env "R2_ENDPOINT" }}
+      - R2_BUCKET={{ index .Env "R2_BUCKET" }}
+      - R2_ACCESS_KEY_ID={{ index .Env "R2_ACCESS_KEY_ID" }}
+      - 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

DEVELOPMENT.md

@@ -0,0 +1,32 @@
+# 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.

Dockerfile

@@ -17,14 +17,14 @@ RUN make clean && make build
 ### DIND VARIANT
 #
 #
-FROM docker:29.5.2-dind AS dind
+FROM docker:29.7.1-dind AS dind
 
 ARG VERSION=dev
 
 LABEL org.opencontainers.image.source="https://gitea.com/gitea/runner"
 LABEL org.opencontainers.image.version="${VERSION}"
 
-RUN apk add --no-cache s6 bash git tzdata
+RUN apk add --no-cache s6 bash git tzdata nftables
 
 COPY --from=builder /opt/src/runner/gitea-runner /usr/local/bin/gitea-runner
 COPY scripts/run.sh /usr/local/bin/run.sh
@@ -37,7 +37,7 @@ ENTRYPOINT ["s6-svscan","/etc/s6"]
 ### DIND-ROOTLESS VARIANT
 #
 #
-FROM docker:29.5.2-dind-rootless AS dind-rootless
+FROM docker:29.7.1-dind-rootless AS dind-rootless
 
 ARG VERSION=dev
 
@@ -45,7 +45,7 @@ LABEL org.opencontainers.image.source="https://gitea.com/gitea/runner"
 LABEL org.opencontainers.image.version="${VERSION}"
 
 USER root
-RUN apk add --no-cache s6 bash git tzdata
+RUN apk add --no-cache s6 bash git tzdata nftables
 
 COPY --from=builder /opt/src/runner/gitea-runner /usr/local/bin/gitea-runner
 COPY scripts/run.sh /usr/local/bin/run.sh
@@ -63,7 +63,7 @@ ENTRYPOINT ["s6-svscan","/etc/s6"]
 ### BASIC VARIANT
 #
 #
-FROM alpine:3.23 AS basic
+FROM alpine:3.24 AS basic
 
 ARG VERSION=dev
 

Makefile

@@ -4,9 +4,9 @@ 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
-GXZ_PACKAGE ?= github.com/ulikunitz/xz/cmd/[email protected]0
+GXZ_PACKAGE ?= github.com/ulikunitz/xz/cmd/[email protected]6 # renovate: datasource=go
 
 LINUX_ARCHS ?= linux/amd64,linux/arm64
 DARWIN_ARCHS ?= darwin-12/amd64,darwin-12/arm64
@@ -18,8 +18,10 @@ 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]
+GOLANGCI_LINT_PACKAGE ?= github.com/golangci/golangci-lint/v2/cmd/[email protected] # renovate: datasource=go
-GOVULNCHECK_PACKAGE ?= golang.org/x/vuln/cmd/govulncheck@v1.3.0
+GOVULNCHECK_PACKAGE ?= golang.org/x/vuln/cmd/govulncheck@v1.6.0 # renovate: datasource=go
+
+GOTEST_FLAGS ?= -race -timeout 20m -parallel 8
 
 STATIC ?=
 EXTLDFLAGS ?=
@@ -38,12 +40,15 @@ endif
 ifeq ($(OS), Windows_NT)
 	GOFLAGS := -v -buildmode=exe
 	EXECUTABLE ?= $(EXECUTABLE).exe
+	GO_ENV_WINDOWS := set GOOS=windows&&
 else ifeq ($(OS), Windows)
 	GOFLAGS := -v -buildmode=exe
 	EXECUTABLE ?= $(EXECUTABLE).exe
+	GO_ENV_WINDOWS := set GOOS=windows&&
 else
 	GOFLAGS := -v
 	EXECUTABLE ?= $(EXECUTABLE)
+	GO_ENV_WINDOWS := GOOS=windows
 endif
 
 STORED_VERSION_FILE := VERSION
@@ -92,7 +97,7 @@ go-check:
 
 .PHONY: fmt-check
 fmt-check: fmt
-	@diff=$$(git diff --color=always); \
+	@diff=$$(git diff --color=always -- '*.go'); \
 	if [ -n "$$diff" ]; then \
 		echo "Please run 'make fmt' and commit the result:"; \
 		printf "%s" "$${diff}"; \
@@ -107,13 +112,21 @@ 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 everything
+lint: lint-go lint-go-windows ## lint everything
 
 .PHONY: lint-go
 lint-go: ## lint go files
 	$(GO) run $(GOLANGCI_LINT_PACKAGE) run
 
+.PHONY: lint-go-windows
+lint-go-windows: ## lint Windows go files
+	$(GO) install $(GOLANGCI_LINT_PACKAGE)
+	$(GO_ENV_WINDOWS) golangci-lint run
+
 .PHONY: lint-go-fix
 lint-go-fix: ## lint go files and fix issues
 	$(GO) run $(GOLANGCI_LINT_PACKAGE) run --fix
@@ -123,12 +136,17 @@ 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
 
 .PHONY: tidy
 tidy: ## run go mod tidy
+	$(eval GO_TOOLCHAIN := $(shell grep -Eo '^toolchain\s+go[0-9.]+' go.mod | cut -d' ' -f2))
 	$(GO) mod tidy
+	@# workaround https://github.com/golang/go/issues/75331: restore toolchain if tidy dropped it
+	@if [ -n "$(GO_TOOLCHAIN)" ] && ! grep -qE '^toolchain\s' go.mod; then \
+		$(GO) mod edit -toolchain=$(GO_TOOLCHAIN); \
+	fi
 
 .PHONY: tidy-check
 tidy-check: tidy
@@ -140,8 +158,14 @@ tidy-check: tidy
 	fi
 
 .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
+	@mkdir -p .tmp
+	@node ./tools/coverage-report.ts -i coverage.txt -o .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)
@@ -210,7 +234,7 @@ docker: ## build the docker image
 .PHONY: clean
 clean: ## delete binary and coverage files
 	$(GO) clean -x -i ./...
-	rm -rf coverage.txt $(EXECUTABLE) $(DIST)
+	rm -rf coverage.txt .tmp $(EXECUTABLE) $(DIST)
 
 .PHONY: version
 version: ## print the version

README.md

@@ -85,46 +85,273 @@ docker run -e GITEA_INSTANCE_URL=https://your_gitea.com -e GITEA_RUNNER_REGISTRA
 
 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
+```
+
+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.
+
+#### `latest-dind-rootless` — rootless Docker-in-Docker
+
+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).
+
+> **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.
+
+> **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). A small set of **deprecated** environment variables can still override parts of that default config, but **only when no `-c` path was given**; they are ignored if you use a config file:
+`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.
 
-| Variable | Effect |
+#### 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. 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
+
+Labels decide **which jobs a runner accepts** and **how it runs them**. A job's `runs-on` is matched against the runner's label names; the first match wins and selects the execution environment for that job.
+
+A label is written as:
+
+```text
+<name>[:<schema>[:<args>]]
+```
+
+| Part | Meaning |
 | --- | --- |
-| `GITEA_DEBUG` | If true, sets log level to `debug` |
+| `name` | The name a workflow refers to in `runs-on`, e.g. `ubuntu-latest`. |
-| `GITEA_TRACE` | If true, sets log level to `trace` |
+| `schema` | Either `docker` or `host`. Defaults to `host` when omitted. |
-| `GITEA_RUNNER_CAPACITY` | Concurrent jobs (integer) |
+| `args` | Only used by the `docker` schema: the image to run the job in. |
-| `GITEA_RUNNER_FILE` | Registration state file path (default `.runner`) |
-| `GITEA_RUNNER_ENVIRON` | Extra job env vars as comma-separated `KEY:VALUE` pairs |
-| `GITEA_RUNNER_ENV_FILE` | Path to an env file merged into job env (same idea as `runner.env_file` in YAML) |
 
-Prefer a YAML file for all settings.
+Two schemas are supported:
+
+- **`docker://<image>`** — the job runs inside a container created from `<image>`:
+
+  ```text
+  ubuntu-latest:docker://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
+  macos:host
+  ```
+
+So with the labels
+
+```text
+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.
+
+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).
+
+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.
 
 #### Registration vs config labels
 
 If `runner.labels` is set in the YAML file, those labels are used during `register` and the `--labels` CLI flag is ignored.
 
-#### External cache (`actions/cache`)
+The `daemon` command also accepts `--labels` (which defaults to the `GITEA_RUNNER_LABELS` environment variable), so the labels of an already registered runner can be changed without deleting its registration file. The most explicit source wins:
 
-If `cache.external_server` is set, you must set `cache.external_secret` to the same value on this runner and on the standalone cache server. Run the server with `gitea-runner cache-server` using a config that defines `cache.external_secret` (and matching `cache.dir` / host / port as needed). Flags `--dir`, `--host`, and `--port` on `cache-server` override the file.
+```
+--labels / GITEA_RUNNER_LABELS   >   runner.labels in the config file   >   labels in the .runner file
+```
+
+Whenever the resulting labels differ from the ones in the registration file, they are written back to it and re-declared to the Gitea instance on startup.
+
+> **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.
+
+#### 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 cache server
+- `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.
+
+To change a value for one job, set it in a step's `env:` or in the job's `container.env`. Setting it at workflow or job level 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.
+
+#### 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.
+
+**Eviction**
+
+An entry nothing has read or written for `retention` is removed, and a repository past `repo_size_limit` loses its least recently accessed entries until it fits; `size_limit` caps the whole cache the same way. Age alone never retires an entry still in use, and whatever these allow, the cache keeps free space above `health_check.min_free_disk_space_mb` when health checks are enabled.
+
+These apply where the cache server runs, so on a shared server they belong in *its* config, not the runners'. See `retention`, `repo_size_limit`, `size_limit` and `sweep_interval` in [config.example.yaml](internal/pkg/config/config.example.yaml) for units and defaults.
+
+**Cache service v2**
+
+`actions/cache@v4.2` and later can use the *cache service v2* API. The runner serves it from the same store as v1, on by default, and it works with `external_server`. Turn it off with:
+
+```yaml
+cache:
+  v2: false
+```
+
+Those actions refuse any host they do not take for GitHub. Rather than misreport the server URL, the runner edits that check out of the action's own bundle and keeps the untouched copy beside it; a bundle it does not recognise is left alone and keeps to v1. The same edit lets the stock `actions/upload-artifact` and `actions/download-artifact` work from `v4.4.0` on, without the `gitea-upload-artifact` fork.
+
+**Shared cache across multiple runners**
+
+Run one dedicated `gitea-runner cache-server` that all runners point at.
+
+1. Create a config file for the cache server host:
+
+   ```yaml
+   cache:
+     dir: /data/actcache
+     port: 8088
+     external_secret: "replace-with-a-strong-random-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-config.yaml cache-server
+   ```
+
+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_file: /path/to/secret # secret can also be passed via a file
+   ```
+
+Jobs reach the cache server at `external_server`, so when a reverse proxy fronts the server, point `external_server` at the proxy. The cache server itself needs no extra configuration.
+
+Alternatively, mount the same NFS/CIFS share on every runner and point `cache.dir` at it — simpler, but with weaker isolation between repositories.
+
+**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.
+
+Flags `--dir`, `--host`, and `--port` on `cache-server` override the corresponding `cache.*` YAML keys; all other settings, including `external_secret`, require the config file.
 
 #### Official Docker image
 
@@ -132,11 +359,32 @@ 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
+- 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`)
+
+Optional host script that runs **after** each task's built-in cleanup (post-steps, container teardown, bind-workdir removal). Use it for extra machine housekeeping — Docker pruning, disk cleanup, and similar.
+
+**While the script runs, the runner stops task heartbeats and stays offline from Gitea's perspective until the script exits (or hits `runner.post_task_script_timeout`, default `5m`).** A script that blocks without exiting keeps the runner from taking new work for up to that timeout. Script output goes to the runner log, not the job log; a non-zero exit is warned but does not change the job result.
+
+On Windows, use `.exe`, `.bat`, or `.cmd` paths; **PowerShell (`.ps1`) is not supported yet** as the configured path — wrap commands in a `.cmd` file instead.
+
+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.
 
 ### Example Deployments
 

act/artifactcache/handler_v2.go

@@ -0,0 +1,274 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"cmp"
+	"encoding/json"
+	"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 is the only error code that is not the client's fault.
+	twirpInternal = "internal"
+)
+
+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_request", 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
+	}
+
+	now := time.Now().Unix()
+	cache := &Cache{
+		Repo:      cred.Repo,
+		Key:       req.Key,
+		Version:   req.Version,
+		Size:      -1, // the size is only known at finalize time
+		CreatedAt: now,
+		UsedAt:    now,
+	}
+	if err := 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_request", 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 {
+		h.twirpError(w, r, twirpInternal, err)
+		return
+	}
+	if cache == nil {
+		h.twirpNotOK(w, r)
+		return
+	}
+	db.Close() // commitCache needs the store closed
+
+	cache.Size, _ = cmp.Or(req.SizeBytes, req.SizeBytesCamel).Int64()
+	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_request", 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
+	if code == twirpInternal {
+		status = http.StatusInternalServerError
+	}
+	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      json.Number `json:"size_bytes"`
+		SizeBytesCamel json.Number `json:"sizeBytes"`
+	}
+
+	v2DownloadRequest struct {
+		Key              string   `json:"key"`
+		Version          string   `json:"version"`
+		RestoreKeys      []string `json:"restore_keys"`
+		RestoreKeysCamel []string `json:"restoreKeys"`
+	}
+)
+
+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.NewDecoder(io.LimitReader(r.Body, 1<<20)).Decode(&req)
+	return req, err
+}

act/artifactcache/handler_v2_test.go

@@ -0,0 +1,243 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"bytes"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"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.NewDecoder(resp.Body).Decode(&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()))
+
+	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.NewDecoder(resp.Body).Decode(&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"])
+	})
+}

act/artifactcache/results.go

@@ -0,0 +1,59 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"crypto/tls"
+	"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.
+//
+// The instance to forward to travels with the job registration rather than with configuration, so
+// a cache server shared between runners serves each of their instances.
+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) {
+	cred, ok := h.lookupCredential(bearerToken(r))
+	if !ok || cred.Results == "" || !strings.HasPrefix(r.URL.Path, artifactServicePath) {
+		http.NotFound(w, r)
+		return
+	}
+	target, err := url.Parse(strings.TrimSuffix(cred.Results, "/"))
+	if err != nil {
+		h.logger.Errorf("artifact service forward to %q: %v", cred.Results, err)
+		w.WriteHeader(http.StatusBadGateway)
+		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, _ *http.Request, err error) {
+			h.logger.Warnf("artifact service forward to %s: %v", target, err)
+			w.WriteHeader(http.StatusBadGateway)
+		},
+	}
+	if cred.InsecureTLS {
+		proxy.Transport = insecureTransport
+	}
+	proxy.ServeHTTP(w, r)
+}
+
+// insecureTransport is shared, because a transport per request would pool no connections.
+var insecureTransport = &http.Transport{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 +1,56 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package artifactcache
+
+import (
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// The artifact half is forwarded under the Host Gitea knows itself by, so the URLs it hands back
+// still point at Gitea, and nothing else is proxied.
+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()
+
+	handler, err := StartHandler(Options{Dir: t.TempDir(), OutboundIP: "127.0.0.1"})
+	require.NoError(t, err)
+	defer handler.Close()
+	const token = "forward-token"
+
+	client := &http.Client{Transport: &bearerTransport{token: token}}
+	post := func(path string) int {
+		req, err := http.NewRequestWithContext(t.Context(), http.MethodPost, handler.ExternalURL()+path, nil)
+		require.NoError(t, err)
+		resp, err := client.Do(req)
+		require.NoError(t, err)
+		resp.Body.Close()
+		return resp.StatusCode
+	}
+
+	assert.Equal(t, http.StatusNotFound, post(artifactServicePath+"CreateArtifact"),
+		"an unregistered token is forwarded nowhere")
+
+	defer handler.RegisterJob(token, JobCredential{Repo: "owner/repo", Results: gitea.URL})()
+
+	assert.Equal(t, http.StatusOK, post(artifactServicePath+"CreateArtifact"))
+	assert.Equal(t, strings.TrimPrefix(gitea.URL, "http://"), 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, artifactServicePath+"CreateArtifact", gotPath)
+
+	gotPath = ""
+	assert.Equal(t, http.StatusNotFound, post("/twirp/github.actions.results.api.v1.OtherService/Do"))
+	assert.Equal(t, http.StatusNotFound, post("/api/v1/repos/owner/repo"))
+	assert.Empty(t, gotPath, "only the artifact service is forwarded")
+}

act/artifactcache/storage.go

@@ -5,12 +5,15 @@
 package artifactcache
 
 import (
+	"crypto/sha256"
+	"encoding/hex"
 	"fmt"
 	"io"
 	"net/http"
 	"os"
 	"path/filepath"
 	"strconv"
+	"strings"
 )
 
 type Storage struct {
@@ -37,7 +40,10 @@ 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 +57,26 @@ 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 {
+	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 nil
+}
+
 func (s *Storage) Commit(id uint64, size int64) (int64, error) {
 	defer func() {
 		_ = os.RemoveAll(s.tempDir(id))
@@ -65,6 +91,31 @@ 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 +135,6 @@ 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 +143,13 @@ 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 +164,17 @@ 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 +183,7 @@ 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_test.go

@@ -390,6 +390,43 @@ func TestMkdirFsImplSafeResolve(t *testing.T) {
 	}
 }
 
+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)
 

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/context_helpers_test.go

@@ -0,0 +1,73 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package common
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	"github.com/sirupsen/logrus"
+)
+
+func TestDryrunContext(t *testing.T) {
+	ctx := context.Background()
+	if Dryrun(ctx) {
+		t.Fatal("plain context should not be dryrun")
+	}
+	if !Dryrun(WithDryrun(ctx, true)) {
+		t.Fatal("WithDryrun(true) should set dryrun")
+	}
+	if Dryrun(WithDryrun(ctx, false)) {
+		t.Fatal("WithDryrun(false) should clear dryrun")
+	}
+}
+
+func TestJobErrorContainer(t *testing.T) {
+	ctx := context.Background()
+	err := errors.New("job failed")
+
+	SetJobError(ctx, err)
+	if got := JobError(ctx); got != nil {
+		t.Fatalf("JobError without container = %v, want nil", got)
+	}
+
+	ctx = WithJobErrorContainer(ctx)
+	SetJobError(ctx, err)
+	if got := JobError(ctx); !errors.Is(got, err) {
+		t.Fatalf("JobError = %v, want %v", got, err)
+	}
+}
+
+func TestLoggerAndHookContext(t *testing.T) {
+	ctx := context.Background()
+	if Logger(ctx) != logrus.StandardLogger() {
+		t.Fatal("plain context should use standard logger")
+	}
+	if LoggerHook(ctx) != nil {
+		t.Fatal("plain context should not have a logger hook")
+	}
+
+	logger := logrus.New()
+	ctx = WithLogger(ctx, logger)
+	if Logger(ctx) != logger {
+		t.Fatal("WithLogger should set logger")
+	}
+
+	hook := testHook{}
+	ctx = WithLoggerHook(ctx, hook)
+	if LoggerHook(ctx) != hook {
+		t.Fatal("WithLoggerHook should set hook")
+	}
+}
+
+type testHook struct{}
+
+func (testHook) Levels() []logrus.Level {
+	return logrus.AllLevels
+}
+
+func (testHook) Fire(*logrus.Entry) error {
+	return nil
+}

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

@@ -7,9 +7,11 @@ package common
 import (
 	"context"
 	"errors"
+	"reflect"
+	"strings"
+	"sync"
 	"sync/atomic"
 	"testing"
-	"time"
 
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -80,44 +82,45 @@ func TestNewConditionalExecutor(t *testing.T) {
 	assert.Equal(1, falseCount)
 }
 
-func TestNewParallelExecutor(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)
-	var count, activeCount, maxCount atomic.Int32
+		running := active.Add(1)
-	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) {
@@ -170,3 +173,60 @@ func TestNewParallelExecutorCanceled(t *testing.T) {
 	assert.Equal(int32(3), count.Load())
 	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
+	record := func(name string) Executor {
+		return func(ctx context.Context) error {
+			calls = append(calls, name)
+			return nil
+		}
+	}
+
+	require.NoError(t, record("if-true").If(func(context.Context) bool { return true })(ctx))
+	require.NoError(t, record("if-false").If(func(context.Context) bool { return false })(ctx))
+	require.NoError(t, record("if-not").IfNot(func(context.Context) bool { return false })(ctx))
+	require.NoError(t, record("if-bool").IfBool(true)(ctx))
+	require.NoError(t, record("main").Finally(record("finally"))(ctx))
+
+	want := []string{"if-true", "if-not", "if-bool", "main", "finally"}
+	if !reflect.DeepEqual(calls, want) {
+		t.Fatalf("calls = %v, want %v", calls, want)
+	}
+}
+
+func TestExecutorFinallyReturnsFinallyErrorWithOriginal(t *testing.T) {
+	mainErr := errors.New("main failed")
+	finalErr := errors.New("cleanup failed")
+
+	err := NewErrorExecutor(mainErr).Finally(NewErrorExecutor(finalErr))(context.Background())
+	require.Error(t, err)
+	if !strings.Contains(err.Error(), "cleanup failed") || !strings.Contains(err.Error(), "main failed") {
+		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.go

@@ -16,6 +16,7 @@ import (
 	"sync"
 
 	"gitea.com/gitea/runner/act/common"
+	"gitea.com/gitea/runner/internal/pkg/lock"
 
 	"github.com/go-git/go-git/v5"
 	"github.com/go-git/go-git/v5/config"
@@ -32,7 +33,7 @@ var (
 	githubHTTPRegex     = regexp.MustCompile(`^https?://.*github.com.*/(.+)/(.+?)(?:.git)?$`)
 	githubSSHRegex      = regexp.MustCompile(`github.com[:/](.+)/(.+?)(?:.git)?$`)
 
-	cloneLocks sync.Map // key: clone target directory; value: *sync.Mutex
+	cloneLocks lock.Keyed[string] // key: clone target directory
 
 	ErrShortRef = errors.New("short SHA references are not supported")
 	ErrNoRepo   = errors.New("unable to find git repo")
@@ -43,10 +44,7 @@ var (
 // Callers reading files inside dir (e.g. tarring a checked-out action into a job container) must hold this lock too,
 // otherwise a concurrent NewGitCloneExecutor on the same dir can mutate the worktree mid-read.
 func AcquireCloneLock(dir string) func() {
-	v, _ := cloneLocks.LoadOrStore(dir, &sync.Mutex{})
+	return cloneLocks.Lock(dir)
-	mu := v.(*sync.Mutex)
-	mu.Lock()
-	return mu.Unlock
 }
 
 type Error struct {
@@ -257,6 +255,14 @@ type NewGitCloneExecutorInput struct {
 	Token       string
 	OfflineMode bool
 
+	// Depth limits the clone/fetch to the given number of commits from the tip of the requested ref.
+	// 0 for full clone.
+	Depth int
+
+	// Quiet drops the informational clone line to debug level, for callers that log their own
+	// download summary (the setup section's action report).
+	Quiet bool
+
 	// For Gitea
 	InsecureSkipTLS bool
 }
@@ -265,10 +271,25 @@ type NewGitCloneExecutorInput struct {
 func CloneIfRequired(ctx context.Context, refName plumbing.ReferenceName, input NewGitCloneExecutorInput, logger log.FieldLogger) (*git.Repository, bool, error) {
 	r, err := git.PlainOpen(input.Dir)
 	if err == nil {
+		// Verify the cached clone still points to the resolved URL before reusing it.
+		remote, err := r.Remote("origin")
+		if err == nil && len(remote.Config().URLs) > 0 && remote.Config().URLs[0] == input.URL {
 			// Reuse existing clone
 			return r, true, nil
 		}
 
+		if err != nil {
+			logger.Debugf("Removing cached clone at %s because origin cannot be read: %v", input.Dir, err)
+		} else if len(remote.Config().URLs) == 0 {
+			logger.Debugf("Removing cached clone at %s because origin has no URL", input.Dir)
+		} else {
+			logger.Debugf("Removing cached clone at %s because origin URL changed from %s to %s", input.Dir, remote.Config().URLs[0], input.URL)
+		}
+		if err := os.RemoveAll(input.Dir); err != nil {
+			return nil, false, fmt.Errorf("remove cached clone %s: %w", input.Dir, err)
+		}
+	}
+
 	var progressWriter io.Writer
 	if isatty.IsTerminal(os.Stdout.Fd()) || isatty.IsCygwinTerminal(os.Stdout.Fd()) {
 		if entry, ok := logger.(*log.Entry); ok {
@@ -294,7 +315,7 @@ func CloneIfRequired(ctx context.Context, refName plumbing.ReferenceName, input
 		}
 	}
 
-	r, err = git.PlainCloneContext(ctx, input.Dir, false, &cloneOptions)
+	r, err = cloneAtDepth(ctx, input, cloneOptions, logger)
 	if err != nil {
 		logger.Errorf("Unable to clone %v %s: %v", input.URL, refName, err)
 		return nil, false, err
@@ -324,11 +345,25 @@ func gitOptions(token string) (fetchOptions git.FetchOptions, pullOptions git.Pu
 	return fetchOptions, pullOptions
 }
 
+// staleRefreshErr reports why a failed refresh must abort: the resolve and
+// checkout that follow are local and succeed on a cancelled context, which
+// would hand back the cached revision as if it were fresh.
+func staleRefreshErr(ctx context.Context, err error) error {
+	if err == nil || errors.Is(err, git.NoErrAlreadyUpToDate) {
+		return nil
+	}
+	return ctx.Err()
+}
+
 // NewGitCloneExecutor creates an executor to clone git repos
 func NewGitCloneExecutor(input NewGitCloneExecutorInput) common.Executor {
 	return func(ctx context.Context) error {
 		logger := common.Logger(ctx)
+		if input.Quiet {
+			logger.Debugf("git clone '%s' # ref=%s", input.URL, input.Ref)
+		} else {
 			logger.Infof("git clone '%s' # ref=%s", input.URL, input.Ref)
+		}
 		logger.Debugf("  cloning %s to %s", input.URL, input.Dir)
 
 		defer AcquireCloneLock(input.Dir)()
@@ -349,8 +384,18 @@ func NewGitCloneExecutor(input NewGitCloneExecutorInput) common.Executor {
 			pullOptions.InsecureSkipTLS = true
 		}
 
+		// Action clones only ever need the tip commit, so keep a shallow cache cheap on update at depth 1 regardless of its original depth
+		// Turning action_shallow_clone off does not convert an existing shallow cache; evict it for a full clone.
+		shallow := isShallow(r)
+		if shallow {
+			fetchOptions.Depth = 1
+			if spec, ok := shallowFetchRefSpec(r, input.Ref); ok {
+				fetchOptions.RefSpecs = []config.RefSpec{spec}
+			}
+		}
+
 		if !isOfflineMode {
-			err = r.Fetch(&fetchOptions)
+			err = r.FetchContext(ctx, &fetchOptions)
 			if err != nil && !errors.Is(err, git.NoErrAlreadyUpToDate) {
 				return err
 			}
@@ -416,11 +461,16 @@ func NewGitCloneExecutor(input NewGitCloneExecutorInput) common.Executor {
 
 		reusedMsg := ""
 
-		if !isOfflineMode {
+		switch {
-			if err = w.Pull(&pullOptions); err != nil && err != git.NoErrAlreadyUpToDate {
+		case !isOfflineMode && !shallow:
+			// In shallow mode the depth-limited fetch above already advanced the ref.
+			if err = w.PullContext(ctx, &pullOptions); err != nil && !errors.Is(err, git.NoErrAlreadyUpToDate) {
 				logger.Debugf("Unable to pull %s: %v", refName, err)
 			}
-		} else if reused {
+			if err := staleRefreshErr(ctx, err); err != nil {
+				return err
+			}
+		case isOfflineMode && reused:
 			reusedMsg = " (reused in offline mode)"
 		}
 
@@ -453,3 +503,53 @@ func NewGitCloneExecutor(input NewGitCloneExecutorInput) common.Executor {
 		return nil
 	}
 }
+
+// cloneAtDepth clones input.URL into input.Dir using opts.
+// With input.Depth > 0 it first tries a shallow, single-branch clone of input.Ref, falling back when error.
+func cloneAtDepth(ctx context.Context, input NewGitCloneExecutorInput, opts git.CloneOptions, logger log.FieldLogger) (*git.Repository, error) {
+	if input.Depth > 0 {
+		for _, refName := range []plumbing.ReferenceName{
+			plumbing.NewBranchReferenceName(input.Ref),
+			plumbing.NewTagReferenceName(input.Ref),
+		} {
+			shallowOpts := opts
+			shallowOpts.Depth = input.Depth
+			shallowOpts.SingleBranch = true
+			shallowOpts.ReferenceName = refName
+			shallowOpts.Tags = git.NoTags
+
+			r, err := git.PlainCloneContext(ctx, input.Dir, false, &shallowOpts)
+			if err == nil {
+				return r, nil
+			}
+			logger.Debugf("Shallow clone of %s as %s failed: %v", input.URL, refName, err)
+			if rmErr := os.RemoveAll(input.Dir); rmErr != nil {
+				return nil, fmt.Errorf("remove partial clone %s: %w", input.Dir, rmErr)
+			}
+		}
+		logger.Debugf("Falling back to a full clone of %s for ref %q", input.URL, input.Ref)
+	}
+
+	return git.PlainCloneContext(ctx, input.Dir, false, &opts)
+}
+
+// isShallow reports whether the local repository was cloned with a limited depth.
+func isShallow(r *git.Repository) bool {
+	shallows, err := r.Storer.Shallow()
+	return err == nil && len(shallows) > 0
+}
+
+// shallowFetchRefSpec returns the single refspec that updates only input.Ref, keeping a shallow clone from re-downloading every branch's history.
+// ok is false when the ref is not present locally as a tag or remote-tracking branch, in which case the broad default refspec is used.
+func shallowFetchRefSpec(r *git.Repository, ref string) (config.RefSpec, bool) {
+	tagRef := plumbing.NewTagReferenceName(ref)
+	if _, err := r.Reference(tagRef, false); err == nil {
+		return config.RefSpec(fmt.Sprintf("+%s:%s", tagRef, tagRef)), true
+	}
+	remoteRef := plumbing.NewRemoteReferenceName("origin", ref)
+	if _, err := r.Reference(remoteRef, false); err == nil {
+		branchRef := plumbing.NewBranchReferenceName(ref)
+		return config.RefSpec(fmt.Sprintf("+%s:%s", branchRef, remoteRef)), true
+	}
+	return "", false
+}

act/common/git/git_test.go

@@ -6,16 +6,26 @@ package git
 
 import (
 	"context"
+	"errors"
 	"fmt"
+	"net/http"
+	"net/http/httptest"
 	"os"
 	"os/exec"
 	"path/filepath"
+	"strconv"
 	"strings"
 	"sync"
 	"syscall"
 	"testing"
 	"time"
 
+	"gitea.com/gitea/runner/act/common"
+
+	gogit "github.com/go-git/go-git/v5"
+	gogitconfig "github.com/go-git/go-git/v5/config"
+	log "github.com/sirupsen/logrus"
+	logrustest "github.com/sirupsen/logrus/hooks/test"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
@@ -49,6 +59,13 @@ func TestFindGitSlug(t *testing.T) {
 	}
 }
 
+func TestErrorWrapsCommitAndCause(t *testing.T) {
+	err := &Error{err: ErrShortRef, commit: "abc123"}
+	require.Equal(t, ErrShortRef.Error(), err.Error())
+	require.ErrorIs(t, err, ErrShortRef)
+	require.Equal(t, "abc123", err.Commit())
+}
+
 func cleanGitHooks(dir string) error {
 	hooksDir := filepath.Join(dir, ".git", "hooks")
 	files, err := os.ReadDir(hooksDir)
@@ -95,6 +112,22 @@ func TestFindGitRemoteURL(t *testing.T) {
 	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, "remote", "add", "ghe", "[email protected]:team/project.git"))
+
+	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) {
 	basedir := t.TempDir()
 
@@ -235,6 +268,51 @@ func TestGitCloneExecutor(t *testing.T) {
 	}
 }
 
+func TestGitCloneExecutorReclonesWhenOriginURLChanges(t *testing.T) {
+	createRemote := func(message string) string {
+		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", message))
+		require.NoError(t, gitCmd("-C", workDir, "push", "-u", "origin", "main"))
+
+		return remoteDir
+	}
+
+	oldRemoteDir := createRemote("old-action")
+	newRemoteDir := createRemote("new-action")
+	cacheDir := t.TempDir()
+
+	require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+		URL: oldRemoteDir,
+		Ref: "main",
+		Dir: cacheDir,
+	})(t.Context()))
+
+	markerPath := filepath.Join(cacheDir, "stale-marker")
+	require.NoError(t, os.WriteFile(markerPath, []byte("stale"), 0o644))
+
+	require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+		URL: newRemoteDir,
+		Ref: "main",
+		Dir: cacheDir,
+	})(t.Context()))
+
+	originURL, err := findGitRemoteURL(t.Context(), cacheDir, "origin")
+	require.NoError(t, err)
+	assert.Equal(t, newRemoteDir, originURL)
+
+	out, err := exec.Command("git", "-C", cacheDir, "log", "--oneline", "-1", "--format=%s").Output()
+	require.NoError(t, err)
+	assert.Equal(t, "new-action", strings.TrimSpace(string(out)))
+
+	_, err = os.Stat(markerPath)
+	require.True(t, os.IsNotExist(err), "stale cached directory should be removed before recloning")
+}
+
 func TestGitCloneExecutorNonFastForwardRef(t *testing.T) {
 	// Simulate the scenario where a remote ref (e.g. a GitHub PR head ref) changes
 	// non-fast-forward between two fetches. Before the fix, the fetch used Force=false,
@@ -335,6 +413,134 @@ func TestGitCloneExecutorOfflineMode(t *testing.T) {
 	})
 }
 
+func TestGitCloneExecutorQuietDemotesCloneLine(t *testing.T) {
+	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", "initial"))
+	require.NoError(t, gitCmd("-C", workDir, "push", "-u", "origin", "main"))
+
+	// Quiet callers report the download themselves, so the clone line must not reach the job log.
+	for name, quiet := range map[string]bool{"quiet": true, "not quiet": false} {
+		t.Run(name, func(t *testing.T) {
+			logger, hook := logrustest.NewNullLogger()
+			logger.SetLevel(log.InfoLevel)
+			ctx := common.WithLogger(context.Background(), logger.WithField("job", "j1"))
+
+			require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+				URL:   remoteDir,
+				Ref:   "main",
+				Dir:   t.TempDir(),
+				Quiet: quiet,
+			})(ctx))
+
+			var cloneLines int
+			for _, entry := range hook.AllEntries() {
+				if strings.HasPrefix(entry.Message, "git clone ") {
+					cloneLines++
+				}
+			}
+			if quiet {
+				assert.Zero(t, cloneLines)
+			} else {
+				assert.Equal(t, 1, cloneLines)
+			}
+		})
+	}
+}
+
+func TestGitCloneExecutorShallow(t *testing.T) {
+	// 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))
+	workDir := t.TempDir()
+	require.NoError(t, gitCmd("clone", remoteDir, workDir))
+	require.NoError(t, gitCmd("-C", workDir, "checkout", "-b", "main"))
+	for _, m := range []string{"c1", "c2", "c3"} {
+		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
+	require.NoError(t, gitCmd("-C", workDir, "push", "-u", "origin", "main"))
+	require.NoError(t, gitCmd("-C", workDir, "push", "origin", "v1"))
+
+	shallowMarker := func(dir string) string { return filepath.Join(dir, ".git", "shallow") }
+
+	t.Run("branch is cloned shallowly", func(t *testing.T) {
+		dir := t.TempDir()
+		require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+			URL: remoteDir, Ref: "main", Dir: dir, Depth: 1,
+		})(t.Context()))
+		assert.FileExists(t, shallowMarker(dir), "clone should be shallow")
+		assert.Equal(t, 1, gitRevCount(t, dir), "only the tip commit should be present")
+		assert.Equal(t, "c3", gitHeadSubject(t, dir))
+	})
+
+	t.Run("tag is cloned shallowly", func(t *testing.T) {
+		dir := t.TempDir()
+		require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+			URL: remoteDir, Ref: "v1", Dir: dir, Depth: 1,
+		})(t.Context()))
+		assert.FileExists(t, shallowMarker(dir), "clone should be shallow")
+		assert.Equal(t, 1, gitRevCount(t, dir))
+		assert.Equal(t, "c3", gitHeadSubject(t, dir))
+	})
+
+	t.Run("SHA falls back to a full clone", func(t *testing.T) {
+		dir := t.TempDir()
+		require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+			URL: remoteDir, Ref: sha, 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.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) {
+		dir := t.TempDir()
+		require.NoError(t, NewGitCloneExecutor(NewGitCloneExecutorInput{
+			URL: remoteDir, Ref: "main", Dir: dir, Depth: 1,
+		})(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, NewGitCloneExecutor(NewGitCloneExecutorInput{
+			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))
+	})
+}
+
+func gitRevParse(t *testing.T, dir, rev string) string {
+	t.Helper()
+	out, err := exec.Command("git", "-C", dir, "rev-parse", rev).Output()
+	require.NoError(t, err)
+	return strings.TrimSpace(string(out))
+}
+
+func gitRevCount(t *testing.T, dir string) int {
+	t.Helper()
+	out, err := exec.Command("git", "-C", dir, "rev-list", "--count", "HEAD").Output()
+	require.NoError(t, err)
+	n, err := strconv.Atoi(strings.TrimSpace(string(out)))
+	require.NoError(t, err)
+	return n
+}
+
+func gitHeadSubject(t *testing.T, dir string) string {
+	t.Helper()
+	out, err := exec.Command("git", "-C", dir, "log", "-1", "--format=%s").Output()
+	require.NoError(t, err)
+	return strings.TrimSpace(string(out))
+}
+
 func gitCmd(args ...string) error {
 	cmd := exec.Command("git", args...)
 	cmd.Stdout = os.Stdout
@@ -409,12 +615,56 @@ 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 TestNewGitCloneExecutorFetchHonoursContext(t *testing.T) {
+	block := make(chan struct{})
+	reached := make(chan struct{})
+	var once sync.Once
+	server := httptest.NewServer(http.HandlerFunc(func(_ http.ResponseWriter, _ *http.Request) {
+		once.Do(func() { close(reached) })
+		<-block
+	}))
+	t.Cleanup(func() {
+		close(block)
+		server.Close()
+	})
+
+	dir := filepath.Join(t.TempDir(), "cached-action")
+	repo, err := gogit.PlainInit(dir, false)
+	require.NoError(t, err)
+	_, err = repo.CreateRemote(&gogitconfig.RemoteConfig{Name: "origin", URLs: []string{server.URL}})
+	require.NoError(t, err)
+
+	ctx, cancel := context.WithCancel(t.Context())
+	defer cancel()
+
+	done := make(chan error, 1)
+	go func() {
+		done <- NewGitCloneExecutor(NewGitCloneExecutorInput{URL: server.URL, Ref: "main", Dir: dir})(ctx)
+	}()
 
-		v1, _ := cloneLocks.LoadOrStore(dir, &sync.Mutex{})
+	select {
-		v2, _ := cloneLocks.LoadOrStore(dir, &sync.Mutex{})
+	case <-reached:
-		require.Same(t, v1, v2)
+	case <-time.After(10 * time.Second):
-	})
+		t.Fatal("the executor never reached the remote")
+	}
+	cancel()
+
+	select {
+	case err := <-done:
+		require.Error(t, err)
+	case <-time.After(10 * time.Second):
+		t.Fatal("fetch ignored context cancellation")
+	}
+}
+
+func TestStaleRefreshErr(t *testing.T) {
+	ctx, cancel := context.WithCancel(t.Context())
+	require.NoError(t, staleRefreshErr(ctx, errors.New("remote hung up")))
+
+	cancel()
+	require.ErrorIs(t, staleRefreshErr(ctx, errors.New("remote hung up")), context.Canceled)
+	require.NoError(t, staleRefreshErr(ctx, gogit.NoErrAlreadyUpToDate))
 }

act/common/job_error.go

@@ -24,7 +24,9 @@ func JobError(ctx context.Context) error {
 }
 
 func SetJobError(ctx context.Context, err error) {
-	ctx.Value(jobErrorContextKeyVal).(map[string]error)["error"] = err
+	if container, ok := ctx.Value(jobErrorContextKeyVal).(map[string]error); ok {
+		container["error"] = err
+	}
 }
 
 // WithJobErrorContainer adds a value to the context as a container for an error

act/common/line_writer.go

@@ -12,6 +12,13 @@ import (
 // LineHandler is a callback function for handling a line
 type LineHandler func(line string) bool
 
+// Flusher is implemented by writers that buffer a trailing, not-yet-terminated
+// line. Callers should flush once the underlying stream has reached EOF so the
+// final line (when it is not newline-terminated) is not lost.
+type Flusher interface {
+	Flush()
+}
+
 type lineWriter struct {
 	buffer   bytes.Buffer
 	handlers []LineHandler
@@ -24,6 +31,14 @@ func NewLineWriter(handlers ...LineHandler) io.Writer {
 	return w
 }
 
+// FlushWriter flushes w if it implements Flusher. It is a no-op otherwise, so
+// callers can flush an io.Writer without knowing its concrete type.
+func FlushWriter(w io.Writer) {
+	if f, ok := w.(Flusher); ok {
+		f.Flush()
+	}
+}
+
 func (lw *lineWriter) Write(p []byte) (n int, err error) {
 	pBuf := bytes.NewBuffer(p)
 	written := 0
@@ -44,6 +59,17 @@ func (lw *lineWriter) Write(p []byte) (n int, err error) {
 	return written, nil
 }
 
+// Flush emits any buffered, not-yet-newline-terminated content as a final line.
+// It is safe to call multiple times; subsequent calls with an empty buffer are
+// no-ops.
+func (lw *lineWriter) Flush() {
+	if lw.buffer.Len() == 0 {
+		return
+	}
+	lw.handleLine(lw.buffer.String())
+	lw.buffer.Reset()
+}
+
 func (lw *lineWriter) handleLine(line string) {
 	for _, h := range lw.handlers {
 		ok := h(line)

act/common/line_writer_test.go

@@ -5,6 +5,7 @@
 package common
 
 import (
+	"io"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -39,3 +40,33 @@ func TestLineWriter(t *testing.T) {
 	assert.Equal(" and another\n", lines[2])
 	assert.Equal("last line\n", lines[3])
 }
+
+func TestLineWriterFlush(t *testing.T) {
+	lines := make([]string, 0)
+	lineHandler := func(s string) bool {
+		lines = append(lines, s)
+		return true
+	}
+
+	lineWriter := NewLineWriter(lineHandler)
+
+	assert := assert.New(t)
+	_, err := lineWriter.Write([]byte("complete line\npartial line without newline"))
+	assert.NoError(err) //nolint:testifylint // pre-existing pattern from nektos/act
+
+	// Only the newline-terminated line is emitted before flushing.
+	assert.Equal([]string{"complete line\n"}, lines)
+
+	// Flushing emits the buffered, not-yet-terminated trailing line.
+	FlushWriter(lineWriter)
+	assert.Equal([]string{"complete line\n", "partial line without newline"}, lines)
+
+	// Flushing again is a no-op: nothing is buffered.
+	FlushWriter(lineWriter)
+	assert.Len(lines, 2)
+}
+
+func TestFlushWriterIgnoresNonFlusher(t *testing.T) {
+	// FlushWriter must be a safe no-op for writers that do not buffer lines.
+	assert.NotPanics(t, func() { FlushWriter(io.Discard) })
+}

act/common/outbound_ip.go

@@ -19,7 +19,9 @@ 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.

act/container/container_types.go

@@ -6,12 +6,14 @@ package container
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"io"
 
 	"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.
@@ -57,6 +59,32 @@ 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)
+)
+
+// 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")
+
+// 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
+}
+
 // Container for managing docker run containers
 type Container interface {
 	Create(capAdd, capDrop []string) common.Executor
@@ -65,6 +93,8 @@ type Container interface {
 	CopyTarStream(ctx context.Context, destPath string, tarStream io.Reader) error
 	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
@@ -82,6 +112,14 @@ 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 +4,10 @@
 
 //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/dfc4efb1e2ab8c06d70d2a1366ad448d2f917e90/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
+// * 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.
@@ -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"
@@ -380,7 +383,7 @@ 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 +518,13 @@ 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)
@@ -1164,16 +1167,17 @@ 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

@@ -0,0 +1,86 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+//go:build !(WITHOUT_DOCKER || !(linux || darwin || windows || netbsd))
+
+package container
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"slices"
+
+	"github.com/kballard/go-shellquote"
+	"github.com/spf13/pflag"
+)
+
+const (
+	pullPolicyAlways  = "always"
+	pullPolicyMissing = "missing"
+	pullPolicyNever   = "never"
+)
+
+var pullPolicies = []string{pullPolicyAlways, pullPolicyMissing, pullPolicyNever}
+
+// createFlags are the flags docker/cli registers on the `create` and `run` commands
+// instead of in addFlags, so they are not part of containerOptions.
+type createFlags struct {
+	platform     string
+	pull         string
+	name         string
+	useAPISocket bool
+}
+
+func registerCreateFlags(flags *pflag.FlagSet) *createFlags {
+	cf := new(createFlags)
+	flags.StringVar(&cf.platform, "platform", "", "Set platform if server is multi-platform capable")
+	flags.StringVar(&cf.pull, "pull", pullPolicyMissing, `Pull image before creating ("always", "missing", "never")`)
+	flags.StringVar(&cf.name, "name", "", "Assign a name to the container")
+	flags.BoolVar(&cf.useAPISocket, "use-api-socket", false, "Bind mount Docker API socket and required auth")
+	// Accepted without effect: pull progress is only logged at debug level, and docker
+	// no longer implements content trust.
+	flags.BoolP("quiet", "q", false, "Suppress the pull output")
+	flags.Bool("disable-content-trust", true, "Skip image verification (deprecated)")
+	return cf
+}
+
+// parseContainerOptions parses a container options string. The flags are returned even
+// on error, holding whatever was read before the failure.
+func parseContainerOptions(options string) (*pflag.FlagSet, *containerOptions, *createFlags, error) {
+	flags := pflag.NewFlagSet("container_flags", pflag.ContinueOnError)
+	flags.SetOutput(io.Discard)
+	copts := addFlags(flags)
+	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)
+	}
+
+	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, nil
+}
+
+// createFlagsFromOptions reads the create-level flags that have to be known before the
+// container is created. Malformed options keep the defaults here and are reported by
+// mergeContainerConfigs at create time.
+func createFlagsFromOptions(options string) *createFlags {
+	_, _, cf, _ := parseContainerOptions(options)
+	return cf
+}
+
+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)
+	}
+
+	if cf.useAPISocket {
+		return errors.New("--use-api-socket is not supported, use the runner's container.docker_host setting to expose a docker socket")
+	}
+
+	return nil
+}

act/container/docker_create_flags_test.go

@@ -0,0 +1,62 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package container
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestCreateFlagsFromOptions(t *testing.T) {
+	for _, tc := range []struct {
+		options  string
+		platform string
+		pull     string
+	}{
+		{"", "", pullPolicyMissing},
+		{"-v /a:/b --platform=linux/arm64 --pull always", "linux/arm64", pullPolicyAlways},
+		{"--platform linux/arm/v7 --pull never", "linux/arm/v7", pullPolicyNever},
+		{`--platform "linux/amd64`, "", pullPolicyMissing}, // malformed, defaults kept
+	} {
+		t.Run(tc.options, func(t *testing.T) {
+			cf := createFlagsFromOptions(tc.options)
+			assert.Equal(t, tc.platform, cf.platform)
+			assert.Equal(t, tc.pull, cf.pull)
+		})
+	}
+}
+
+func TestCreateFlagsValidate(t *testing.T) {
+	for _, tc := range []struct {
+		options string
+		wantErr string
+	}{
+		{"--quiet --disable-content-trust --name mine", ""},
+		{"--pull sometimes", `invalid --pull option "sometimes"`},
+		{"--use-api-socket", "--use-api-socket is not supported"},
+	} {
+		t.Run(tc.options, func(t *testing.T) {
+			err := createFlagsFromOptions(tc.options).validate()
+			if tc.wantErr == "" {
+				require.NoError(t, err)
+				return
+			}
+			require.ErrorContains(t, err, tc.wantErr)
+		})
+	}
+}
+
+func TestNewContainerAppliesCreateFlags(t *testing.T) {
+	input := &NewContainerInput{Platform: "linux/amd64", Options: "--platform linux/arm64 --pull never"}
+	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"}
+	NewContainer(kept)
+	assert.Equal(t, "linux/amd64", kept.Platform)
+}

act/container/docker_logger.go

@@ -57,7 +57,7 @@ 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 != "" {

act/container/docker_network.go

@@ -8,13 +8,70 @@ package container
 
 import (
 	"context"
+	"errors"
+	"fmt"
+	"strings"
+	"time"
 
 	"gitea.com/gitea/runner/act/common"
 
 	"github.com/moby/moby/client"
 )
 
-func NewDockerNetworkCreateExecutor(name string) common.Executor {
+const (
+	networkCreateAttempts   = 3
+	networkCreateRetryDelay = time.Second
+
+	// marks the networks a runner creates for its jobs, so it can tell its own leftovers from
+	// those of another runner sharing the daemon
+	runnerUUIDLabel = "com.gitea.runner.uuid"
+)
+
+// 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 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, createdBefore)
+}
+
+func removeOrphanNetworks(ctx context.Context, cli client.APIClient, runnerUUID 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
+	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) != 0 || result.Network.Created.After(createdBefore) {
+			continue
+		}
+		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)
 		if err != nil {
@@ -36,16 +93,45 @@ func NewDockerNetworkCreateExecutor(name string) common.Executor {
 			}
 		}
 
-		_, 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,
-		if err != nil {
+				EnableIPv6: opts.EnableIPv6,
-			return err
+				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 NewDockerNetworkRemoveExecutor(name string) common.Executor {
@@ -64,6 +150,7 @@ 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{})
@@ -71,16 +158,17 @@ 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...)
 	}
 }

act/container/docker_network_test.go

@@ -0,0 +1,50 @@
+// Copyright 2026 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package container
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+
+	cerrdefs "github.com/containerd/errdefs"
+	"github.com/moby/moby/api/types/network"
+	mobyclient "github.com/moby/moby/client"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+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")))
+}
+
+// Of this runner's networks, only the ones nothing 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{
+		{Network: network.Network{ID: "orphan"}},
+		{Network: network.Network{ID: "busy"}},
+		{Network: network.Network{ID: "starting"}},
+	}}, 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)}}}, nil)
+	client.On("NetworkRemove", ctx, "orphan", mobyclient.NetworkRemoveOptions{}).
+		Return(mobyclient.NetworkRemoveResult{}, nil)
+
+	require.NoError(t, removeOrphanNetworks(ctx, client, "runner-1", cutoff))
+	client.AssertExpectations(t)
+}

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)
@@ -61,21 +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") {
-				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)
-
-				_ = logDockerResponse(logger, reader, err != nil)
-			}
 				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 = ""
+			err = pullOnce(imagePullOptions)
+		}
+		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) {
@@ -120,6 +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 {

act/container/docker_pull_test.go

@@ -6,11 +6,15 @@ package container
 
 import (
 	"context"
+	"io"
+	"strings"
 	"testing"
 
 	"github.com/docker/cli/cli/config"
 	log "github.com/sirupsen/logrus"
+	"github.com/sirupsen/logrus/hooks/test"
 	assert "github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )
 
 func init() {
@@ -65,3 +69,21 @@ func TestGetImagePullOptions(t *testing.T) {
 	assert.NoError(t, err, "Failed to create ImagePullOptions") //nolint:testifylint // pre-existing issue from nektos/act
 	assert.Equal(t, "eyJ1c2VybmFtZSI6InVzZXJuYW1lIiwicGFzc3dvcmQiOiJwYXNzd29yZFxuIiwic2VydmVyYWRkcmVzcyI6Imh0dHBzOi8vaW5kZXguZG9ja2VyLmlvL3YxLyJ9", options.RegistryAuth, "RegistryAuth should be taken from local docker config")
 }
+
+// A digest-pinned image is immutable, so its local copy is always current.
+func TestIsPinnedImage(t *testing.T) {
+	assert.True(t, isPinnedImage("alpine@sha256:28bd5fe8b56d1bd048e5babf5b10710ebe0bae67db86916198a6eec434943f8b"))
+	assert.False(t, isPinnedImage("alpine:latest"))
+}
+
+// The pull path reports a failure the daemon sent mid-stream, so it must carry the reason
+// whichever of the two shapes the daemon used.
+func TestLogDockerResponseError(t *testing.T) {
+	logger, _ := test.NewNullLogger()
+	streamErr := func(line string) error {
+		return logDockerResponse(logger, io.NopCloser(strings.NewReader(line)), false)
+	}
+	require.EqualError(t, streamErr(`{"error":"toomanyrequests: rate limit exceeded"}`), "toomanyrequests: rate limit exceeded")
+	require.EqualError(t, streamErr(`{"errorDetail":{"message":"unexpected EOF"}}`), "unexpected EOF")
+	require.NoError(t, streamErr(`{"status":"Downloading"}`))
+}

act/container/docker_run_test.go

@@ -5,9 +5,11 @@
 package container
 
 import (
+	"archive/tar"
 	"bufio"
 	"bytes"
 	"context"
+	"encoding/binary"
 	"errors"
 	"io"
 	"net"
@@ -20,7 +22,10 @@ import (
 	"gitea.com/gitea/runner/act/common"
 
 	cerrdefs "github.com/containerd/errdefs"
+	"github.com/moby/moby/api/pkg/stdcopy"
 	"github.com/moby/moby/api/types/container"
+	"github.com/moby/moby/api/types/mount"
+	"github.com/moby/moby/api/types/network"
 	mobyclient "github.com/moby/moby/client"
 	"github.com/sirupsen/logrus/hooks/test"
 	"github.com/stretchr/testify/assert"
@@ -74,6 +79,11 @@ type mockDockerClient struct {
 	mock.Mock
 }
 
+func (m *mockDockerClient) ServerVersion(ctx context.Context, opts mobyclient.ServerVersionOptions) (mobyclient.ServerVersionResult, error) {
+	args := m.Called(ctx, opts)
+	return args.Get(0).(mobyclient.ServerVersionResult), args.Error(1)
+}
+
 func (m *mockDockerClient) ExecCreate(ctx context.Context, id string, opts mobyclient.ExecCreateOptions) (mobyclient.ExecCreateResult, error) {
 	args := m.Called(ctx, id, opts)
 	return args.Get(0).(mobyclient.ExecCreateResult), args.Error(1)
@@ -89,6 +99,11 @@ func (m *mockDockerClient) ExecInspect(ctx context.Context, execID string, opts
 	return args.Get(0).(mobyclient.ExecInspectResult), args.Error(1)
 }
 
+func (m *mockDockerClient) ContainerAttach(ctx context.Context, containerID string, opts mobyclient.ContainerAttachOptions) (mobyclient.ContainerAttachResult, error) {
+	args := m.Called(ctx, containerID, opts)
+	return args.Get(0).(mobyclient.ContainerAttachResult), args.Error(1)
+}
+
 func (m *mockDockerClient) ContainerWait(ctx context.Context, containerID string, opts mobyclient.ContainerWaitOptions) mobyclient.ContainerWaitResult {
 	args := m.Called(ctx, containerID, opts)
 	return args.Get(0).(mobyclient.ContainerWaitResult)
@@ -109,6 +124,31 @@ func (m *mockDockerClient) ContainerList(ctx context.Context, opts mobyclient.Co
 	return args.Get(0).(mobyclient.ContainerListResult), args.Error(1)
 }
 
+func (m *mockDockerClient) ContainerRemove(ctx context.Context, id string, opts mobyclient.ContainerRemoveOptions) (mobyclient.ContainerRemoveResult, error) {
+	args := m.Called(ctx, id, opts)
+	return args.Get(0).(mobyclient.ContainerRemoveResult), args.Error(1)
+}
+
+func (m *mockDockerClient) ContainerKill(ctx context.Context, id string, opts mobyclient.ContainerKillOptions) (mobyclient.ContainerKillResult, error) {
+	args := m.Called(ctx, id, opts)
+	return args.Get(0).(mobyclient.ContainerKillResult), args.Error(1)
+}
+
+func (m *mockDockerClient) NetworkList(ctx context.Context, opts mobyclient.NetworkListOptions) (mobyclient.NetworkListResult, error) {
+	args := m.Called(ctx, opts)
+	return args.Get(0).(mobyclient.NetworkListResult), args.Error(1)
+}
+
+func (m *mockDockerClient) NetworkInspect(ctx context.Context, id string, opts mobyclient.NetworkInspectOptions) (mobyclient.NetworkInspectResult, error) {
+	args := m.Called(ctx, id, opts)
+	return args.Get(0).(mobyclient.NetworkInspectResult), args.Error(1)
+}
+
+func (m *mockDockerClient) NetworkRemove(ctx context.Context, id string, opts mobyclient.NetworkRemoveOptions) (mobyclient.NetworkRemoveResult, error) {
+	args := m.Called(ctx, id, opts)
+	return args.Get(0).(mobyclient.NetworkRemoveResult), args.Error(1)
+}
+
 type endlessReader struct {
 	io.Reader
 }
@@ -206,6 +246,71 @@ func TestDockerExecFailure(t *testing.T) {
 	client.AssertExpectations(t)
 }
 
+// stdcopyFrame wraps payload in a single Docker multiplexed-stream frame, the
+// format StdCopy expects: an 8-byte header (stream type + 4-byte big-endian
+// length) followed by the payload.
+func stdcopyFrame(stream stdcopy.StdType, payload string) []byte {
+	b := make([]byte, 8+len(payload))
+	b[0] = byte(stream)
+	binary.BigEndian.PutUint32(b[4:8], uint32(len(payload)))
+	copy(b[8:], payload)
+	return b
+}
+
+// TestDockerAttachFlushesTrailingLine verifies that wait() blocks until the
+// attach() streaming goroutine has drained and flushed the container's output,
+// so a final line without a trailing newline is not lost.
+func TestDockerAttachFlushesTrailingLine(t *testing.T) {
+	ctx := context.Background()
+
+	framed := bytes.NewBuffer(stdcopyFrame(stdcopy.Stdout, "line one\nlast line without newline"))
+
+	var lines []string
+	logWriter := common.NewLineWriter(func(s string) bool {
+		lines = append(lines, s)
+		return true
+	})
+
+	client := &mockDockerClient{}
+	client.On("ContainerAttach", ctx, "123", mock.AnythingOfType("client.ContainerAttachOptions")).
+		Return(mobyclient.ContainerAttachResult{
+			HijackedResponse: mobyclient.HijackedResponse{
+				Conn:   &mockConn{},
+				Reader: bufio.NewReader(framed),
+			},
+		}, nil)
+
+	statusCh := make(chan container.WaitResponse, 1)
+	statusCh <- container.WaitResponse{StatusCode: 0}
+	errCh := make(chan error, 1)
+	client.On("ContainerWait", ctx, "123", mobyclient.ContainerWaitOptions{Condition: container.WaitConditionNotRunning}).
+		Return(mobyclient.ContainerWaitResult{
+			Result: (<-chan container.WaitResponse)(statusCh),
+			Error:  (<-chan error)(errCh),
+		})
+
+	cr := &containerReference{
+		id:  "123",
+		cli: client,
+		input: &NewContainerInput{
+			Image:  "image",
+			Stdout: logWriter,
+			Stderr: logWriter,
+		},
+	}
+
+	require.NoError(t, cr.attach()(ctx))
+	require.NoError(t, cr.wait()(ctx))
+
+	// wait() must have blocked until the goroutine drained AND flushed; the
+	// trailing, non-newline-terminated line must therefore be present. Reading
+	// lines here is race-free because wait() synchronizes on attachDone, which
+	// the goroutine closes after the final append.
+	assert.Equal(t, []string{"line one\n", "last line without newline"}, lines)
+
+	client.AssertExpectations(t)
+}
+
 func TestDockerWaitFailure(t *testing.T) {
 	ctx := context.Background()
 
@@ -260,6 +365,44 @@ func TestDockerCopyTarStream(t *testing.T) {
 	client.AssertExpectations(t)
 }
 
+// Docker 29.5+ rejects absolute names in the mkdir tarball with
+// "path escapes from parent", since it is extracted relative to "/".
+func TestDockerCopyTarStreamMkdirEntryIsRelative(t *testing.T) {
+	ctx := context.Background()
+
+	var mkdirNames []string
+	client := &mockDockerClient{}
+	client.On("CopyToContainer", ctx, "123", mock.MatchedBy(func(opts mobyclient.CopyToContainerOptions) bool {
+		if opts.DestinationPath != "/" || opts.Content == nil {
+			return false
+		}
+		tr := tar.NewReader(opts.Content)
+		for {
+			hdr, err := tr.Next()
+			if err != nil {
+				break
+			}
+			mkdirNames = append(mkdirNames, hdr.Name)
+		}
+		return true
+	})).Return(mobyclient.CopyToContainerResult{}, nil)
+	client.On("CopyToContainer", ctx, "123", mock.MatchedBy(func(opts mobyclient.CopyToContainerOptions) bool {
+		return opts.DestinationPath == "/var/run/act" && opts.Content != nil
+	})).Return(mobyclient.CopyToContainerResult{}, nil)
+	cr := &containerReference{
+		id:  "123",
+		cli: client,
+		input: &NewContainerInput{
+			Image: "image",
+		},
+	}
+
+	require.NoError(t, cr.CopyTarStream(ctx, "/var/run/act", &bytes.Buffer{}))
+	assert.Equal(t, []string{"var/run/act"}, mkdirNames)
+
+	client.AssertExpectations(t)
+}
+
 func TestDockerCopyTarStreamErrorInCopyFiles(t *testing.T) {
 	ctx := context.Background()
 
@@ -309,6 +452,64 @@ func TestDockerCopyTarStreamErrorInMkdir(t *testing.T) {
 	client.AssertExpectations(t)
 }
 
+// A remove that raced the daemon's AutoRemove teardown is not a failure and must not
+// be logged as one.
+func TestRemoveIgnoresAutoRemoveRace(t *testing.T) {
+	removeOpts := mobyclient.ContainerRemoveOptions{RemoveVolumes: true, Force: true}
+	killOpts := mobyclient.ContainerKillOptions{Signal: "SIGKILL"}
+	for _, tc := range []struct {
+		name        string
+		err         error
+		wantWait    bool
+		wantFailure bool
+	}{
+		{name: "removal in progress", err: cerrdefs.ErrConflict.WithMessage("removal of container abc is already in progress"), wantWait: true},
+		{name: "already removed", err: cerrdefs.ErrNotFound.WithMessage("No such container: abc")},
+		{name: "removed cleanly", err: nil},
+		{name: "real failure", err: errors.New("driver failed to remove root filesystem"), wantFailure: true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			logger, hook := test.NewNullLogger()
+			ctx := common.WithLogger(context.Background(), logger)
+			client := &mockDockerClient{}
+			client.On("ContainerKill", ctx, "abc", killOpts).Return(mobyclient.ContainerKillResult{}, nil)
+			client.On("ContainerRemove", ctx, "abc", removeOpts).Return(mobyclient.ContainerRemoveResult{}, tc.err)
+			if tc.wantWait {
+				removed := make(chan container.WaitResponse, 1)
+				removed <- container.WaitResponse{}
+				client.On("ContainerWait", mock.Anything, "abc", mobyclient.ContainerWaitOptions{Condition: container.WaitConditionRemoved}).
+					Return(mobyclient.ContainerWaitResult{Result: removed})
+			}
+			cr := &containerReference{id: "abc", cli: client}
+
+			require.NoError(t, cr.remove()(ctx))
+			// a failure keeps the id, so a later Remove() can retry it
+			if tc.wantFailure {
+				assert.Equal(t, "abc", cr.id)
+				assert.Len(t, hook.AllEntries(), 1)
+			} else {
+				assert.Empty(t, cr.id)
+				assert.Empty(t, hook.AllEntries())
+			}
+			client.AssertExpectations(t)
+		})
+	}
+}
+
+// A container whose id was never learned, because find() could not reach the daemon or
+// create() lost its reply, must still be removed rather than leaking with its network. It
+// was never started here, so it is not worth a kill of its own.
+func TestRemoveWithoutIDUsesName(t *testing.T) {
+	ctx := context.Background()
+	client := &mockDockerClient{}
+	client.On("ContainerRemove", ctx, "job-1", mobyclient.ContainerRemoveOptions{RemoveVolumes: true, Force: true}).
+		Return(mobyclient.ContainerRemoveResult{}, nil)
+	cr := &containerReference{cli: client, input: &NewContainerInput{Name: "job-1"}}
+
+	require.NoError(t, cr.remove()(ctx))
+	client.AssertExpectations(t)
+}
+
 // find() must drop a stale cached id so later Copy/Exec don't hit the
 // daemon with a torn-down container.
 func TestFindRevalidatesStaleID(t *testing.T) {
@@ -376,7 +577,7 @@ func TestRejectsMissingContainer(t *testing.T) {
 	cr := &containerReference{cli: client, input: &NewContainerInput{Name: "job-1"}}
 	check := func(op string, err error) {
 		t.Helper()
-		require.Error(t, err, op)
+		require.ErrorIs(t, err, ErrContainerNotFound, op)
 		assert.Contains(t, err.Error(), `container "job-1" does not exist`, op)
 	}
 	check("copyContent", cr.copyContent("/var/run/act", &FileEntry{Name: "x", Mode: 0o644})(ctx))
@@ -385,6 +586,15 @@ func TestRejectsMissingContainer(t *testing.T) {
 	check("exec", cr.exec([]string{"echo"}, nil, "", "")(ctx))
 	_, err := cr.GetContainerArchive(ctx, "/var/run/act/x")
 	check("GetContainerArchive", err)
+	_, err = cr.Inspect(ctx)
+	check("Inspect", err)
+
+	// a known id the daemon has since dropped
+	client.On("ContainerInspect", ctx, "gone", mobyclient.ContainerInspectOptions{}).
+		Return(mobyclient.ContainerInspectResult{}, cerrdefs.ErrNotFound)
+	removed := &containerReference{id: "gone", cli: client, input: &NewContainerInput{Name: "job-1"}}
+	_, err = removed.Inspect(ctx)
+	check("Inspect after removal", err)
 }
 
 // End-to-end: a stale cr.id is cleared, repopulated from name lookup,
@@ -433,7 +643,7 @@ func TestDockerCopyToSymlinkPath(t *testing.T) {
 	// CopyTarStream first creates the destination directory by extracting a tar at "/",
 	// which makes the daemon mkdir var, then var/run (the symlink), then act — the exact
 	// step that fails on the broken daemon.
-	err := rc.CopyTarStream(ctx, "/var/run/act", &bytes.Buffer{})
+	err := rc.CopyTarStream(ctx, "/var/run/act/actions/", &bytes.Buffer{})
 	require.NoError(t, err)
 }
 
@@ -513,6 +723,110 @@ func TestCheckVolumes(t *testing.T) {
 	}
 }
 
+func TestSanitizeOptionsHostConfig(t *testing.T) {
+	logger, _ := test.NewNullLogger()
+
+	dangerous := func() *container.HostConfig {
+		return &container.HostConfig{
+			PidMode:      "host",
+			IpcMode:      "host",
+			UTSMode:      "host",
+			CgroupnsMode: "host",
+			UsernsMode:   "host",
+			CapAdd:       []string{"ALL"},
+			SecurityOpt:  []string{"seccomp=unconfined", "apparmor=unconfined"},
+			VolumesFrom:  []string{"other"},
+			Runtime:      "runc",
+			Resources: container.Resources{
+				CgroupParent:      "/custom",
+				Devices:           []container.DeviceMapping{{PathOnHost: "/dev/sda", PathInContainer: "/dev/sda", CgroupPermissions: "rwm"}},
+				DeviceCgroupRules: []string{"a *:* rwm"},
+			},
+			Sysctls: map[string]string{"net.ipv4.ip_forward": "1"},
+		}
+	}
+
+	hostConfig := dangerous()
+	sanitizeOptionsHostConfig(logger, hostConfig)
+
+	assert.Empty(t, string(hostConfig.PidMode))
+	assert.Empty(t, string(hostConfig.IpcMode))
+	assert.Empty(t, string(hostConfig.UTSMode))
+	assert.Empty(t, string(hostConfig.CgroupnsMode))
+	assert.Empty(t, string(hostConfig.UsernsMode))
+	assert.Empty(t, hostConfig.CapAdd)
+	assert.Empty(t, hostConfig.SecurityOpt)
+	assert.Empty(t, hostConfig.Devices)
+	assert.Empty(t, hostConfig.DeviceCgroupRules)
+	assert.Empty(t, hostConfig.VolumesFrom)
+	assert.Empty(t, hostConfig.Runtime)
+	assert.Empty(t, hostConfig.CgroupParent)
+	assert.Empty(t, hostConfig.Sysctls)
+}
+
+func TestMergeContainerConfigsStripsDangerousOptionsWhenUnprivileged(t *testing.T) {
+	// OS-independent options only: --device parsing requires a linux/windows
+	// server OS, which is not guaranteed for the test host.
+	const dangerousOptions = "--pid=host --ipc=host --uts=host --cgroupns=host " +
+		"--userns=host --cap-add=ALL --security-opt seccomp=unconfined " +
+		"--security-opt apparmor=unconfined --volumes-from other " +
+		"--runtime runc --cgroup-parent /custom --sysctl net.ipv4.ip_forward=1"
+
+	t.Run("unprivileged strips host-escape options", func(t *testing.T) {
+		logger, _ := test.NewNullLogger()
+		ctx := common.WithLogger(context.Background(), logger)
+		cr := &containerReference{
+			input: &NewContainerInput{
+				Options:     dangerousOptions,
+				NetworkMode: "bridge",
+				UsernsMode:  "private",
+			},
+		}
+
+		_, hostConfig, err := cr.mergeContainerConfigs(ctx, &container.Config{}, &container.HostConfig{
+			Privileged:  false,
+			UsernsMode:  container.UsernsMode("private"),
+			NetworkMode: container.NetworkMode("bridge"),
+		})
+		require.NoError(t, err)
+
+		assert.False(t, hostConfig.Privileged)
+		assert.Empty(t, string(hostConfig.PidMode))
+		assert.Empty(t, string(hostConfig.IpcMode))
+		assert.Empty(t, string(hostConfig.UTSMode))
+		assert.Empty(t, string(hostConfig.CgroupnsMode))
+		// UsernsMode must keep the runner-controlled value, not the one from options.
+		assert.Equal(t, "private", string(hostConfig.UsernsMode))
+		assert.Empty(t, hostConfig.CapAdd)
+		assert.Empty(t, hostConfig.SecurityOpt)
+		assert.Empty(t, hostConfig.VolumesFrom)
+		assert.Empty(t, hostConfig.Runtime)
+		assert.Empty(t, hostConfig.CgroupParent)
+		assert.Empty(t, hostConfig.Sysctls)
+	})
+
+	t.Run("privileged preserves options", func(t *testing.T) {
+		logger, _ := test.NewNullLogger()
+		ctx := common.WithLogger(context.Background(), logger)
+		cr := &containerReference{
+			input: &NewContainerInput{
+				Options:     "--pid=host --cap-add=ALL --security-opt seccomp=unconfined",
+				NetworkMode: "bridge",
+			},
+		}
+
+		_, hostConfig, err := cr.mergeContainerConfigs(ctx, &container.Config{}, &container.HostConfig{
+			Privileged:  true,
+			NetworkMode: container.NetworkMode("bridge"),
+		})
+		require.NoError(t, err)
+
+		assert.Equal(t, "host", string(hostConfig.PidMode))
+		assert.Equal(t, []string{"ALL"}, hostConfig.CapAdd)
+		assert.Equal(t, []string{"seccomp=unconfined"}, hostConfig.SecurityOpt)
+	})
+}
+
 func TestCheckVolumesRejectsEscapingHostPaths(t *testing.T) {
 	logger, _ := test.NewNullLogger()
 	ctx := common.WithLogger(context.Background(), logger)
@@ -549,3 +863,87 @@ func TestCheckVolumesRejectsEscapingHostPaths(t *testing.T) {
 	})
 	assert.Empty(t, hostConf.Binds)
 }
+
+func TestContainerInfoFromInspect(t *testing.T) {
+	t.Run("reports no healthcheck when the image declares none", func(t *testing.T) {
+		info := containerInfoFromInspect(container.InspectResponse{
+			ID:    "abc123",
+			State: &container.State{Status: "running", Running: true},
+		})
+
+		assert.Equal(t, "abc123", info.ID)
+		assert.Equal(t, "running", info.State)
+		assert.Equal(t, HealthNone, info.Health)
+		assert.Empty(t, info.Ports)
+	})
+
+	t.Run("reports the health status and the last probe output", func(t *testing.T) {
+		info := containerInfoFromInspect(container.InspectResponse{
+			State: &container.State{
+				Status: "running",
+				Health: &container.Health{
+					Status: container.Unhealthy,
+					Log: []*container.HealthcheckResult{
+						{Output: "first\n"},
+						{Output: "connection refused\n"},
+					},
+				},
+			},
+		})
+
+		assert.Equal(t, HealthUnhealthy, info.Health)
+		assert.Equal(t, "connection refused", info.HealthOutput)
+	})
+
+	t.Run("reports the published ports", func(t *testing.T) {
+		info := containerInfoFromInspect(container.InspectResponse{
+			State: &container.State{Status: "running"},
+			NetworkSettings: &container.NetworkSettings{
+				Ports: network.PortMap{
+					network.MustParsePort("5432/tcp"): []network.PortBinding{{HostPort: "49153"}},
+					network.MustParsePort("6379/tcp"): nil,
+				},
+			},
+		})
+
+		assert.Equal(t, map[string]string{"5432": "49153"}, info.Ports)
+	})
+
+	t.Run("tolerates a container without state", func(t *testing.T) {
+		info := containerInfoFromInspect(container.InspectResponse{ID: "abc123"})
+
+		assert.Equal(t, "abc123", info.ID)
+		assert.Equal(t, HealthNone, info.Health)
+	})
+}
+
+func TestMergeContainerConfigsVolumesReplaceRunnerMounts(t *testing.T) {
+	logger, _ := test.NewNullLogger()
+	ctx := common.WithLogger(context.Background(), logger)
+	cr := &containerReference{
+		input: &NewContainerInput{
+			NetworkMode: "bridge",
+			Options:     "--volume /host/tools:/opt/hostedtoolcache",
+		},
+	}
+
+	_, hostConf, err := cr.mergeContainerConfigs(ctx, &container.Config{}, &container.HostConfig{
+		Binds:  []string{"/var/run/docker.sock:/var/run/docker.sock"},
+		Mounts: []mount.Mount{{Type: mount.TypeVolume, Source: "act-toolcache", Target: "/opt/hostedtoolcache"}},
+	})
+	require.NoError(t, err)
+	assert.Equal(t, []string{"/var/run/docker.sock:/var/run/docker.sock", "/host/tools:/opt/hostedtoolcache"}, hostConf.Binds)
+	assert.Empty(t, hostConf.Mounts)
+}
+
+// A dead daemon must fail the job, not panic through logrus and not silently
+// drop the requested platform.
+func TestSupportsContainerImagePlatformDaemonError(t *testing.T) {
+	cli := &mockDockerClient{}
+	cli.On("ServerVersion", mock.Anything, mock.Anything).
+		Return(mobyclient.ServerVersionResult{}, errors.New("cannot connect to the Docker daemon"))
+
+	supported, err := supportsContainerImagePlatform(t.Context(), cli)
+	require.ErrorContains(t, err, "cannot connect to the Docker daemon")
+	assert.False(t, supported)
+}

act/container/docker_stub.go

@@ -8,12 +8,13 @@ package container
 
 import (
 	"context"
+	"errors"
 	"runtime"
+	"time"
 
 	"gitea.com/gitea/runner/act/common"
 
 	"github.com/moby/moby/api/types/system"
-	"github.com/pkg/errors"
 )
 
 // ImageExistsLocally returns a boolean indicating if an image with the
@@ -61,7 +62,7 @@ func NewDockerVolumeRemoveExecutor(volume string, force bool) common.Executor {
 	}
 }
 
-func NewDockerNetworkCreateExecutor(name string) common.Executor {
+func NewDockerNetworkCreateExecutor(name string, opts NewDockerNetworkCreateExecutorInput) common.Executor {
 	return func(ctx context.Context) error {
 		return nil
 	}
@@ -72,3 +73,7 @@ func NewDockerNetworkRemoveExecutor(name string) common.Executor {
 		return nil
 	}
 }
+
+func RemoveOrphanNetworks(ctx context.Context, runnerUUID string, createdBefore time.Time) error {
+	return nil
+}

act/container/host_environment.go

@@ -17,12 +17,14 @@ import (
 	"path/filepath"
 	"runtime"
 	"strings"
+	"sync"
 	"sync/atomic"
 	"time"
 
 	"gitea.com/gitea/runner/act/common"
 	"gitea.com/gitea/runner/act/filecollector"
 	"gitea.com/gitea/runner/act/lookpath"
+	"gitea.com/gitea/runner/internal/pkg/process"
 
 	"github.com/go-git/go-billy/v5/helper/polyfill"
 	"github.com/go-git/go-billy/v5/osfs"
@@ -42,6 +44,25 @@ type HostEnvironment struct {
 	CleanUp      func()
 	StdOut       io.Writer
 	AllocatePTY  bool // allocate a pseudo-TTY for each step's process
+
+	// procGroup owns every process the job's steps start. Atomic: Remove may read
+	// it while a step is still starting.
+	procGroupOnce sync.Once
+	procGroup     atomic.Pointer[process.Group]
+}
+
+// processGroup returns the job-scoped process group, creating it on first use.
+// Returns nil if the job object could not be created; Group is nil-safe.
+func (e *HostEnvironment) processGroup(ctx context.Context) *process.Group {
+	e.procGroupOnce.Do(func() {
+		group, err := process.NewGroup()
+		if err != nil {
+			common.Logger(ctx).Warnf("could not create the job's process group; processes a step leaves behind can only be reclaimed by the workspace scan: %v", err)
+			return
+		}
+		e.procGroup.Store(group)
+	})
+	return e.procGroup.Load()
 }
 
 func (e *HostEnvironment) Create(_, _ []string) common.Executor {
@@ -133,6 +154,14 @@ func (e *HostEnvironment) CopyDir(destPath, srcPath string, useGitIgnore bool) c
 	}
 }
 
+func (e *HostEnvironment) DumpLogs(_ context.Context) error {
+	return nil
+}
+
+func (e *HostEnvironment) Inspect(_ context.Context) (*Info, error) {
+	return &Info{Health: HealthNone, Ports: map[string]string{}}, nil
+}
+
 func (e *HostEnvironment) GetContainerArchive(ctx context.Context, srcPath string) (io.ReadCloser, error) {
 	buf := &bytes.Buffer{}
 	tw := tar.NewWriter(buf)
@@ -261,7 +290,7 @@ func setupPty(cmd *exec.Cmd, cmdline string) (*os.File, *os.File, error) {
 	cmd.Stdin = tty
 	cmd.Stdout = tty
 	cmd.Stderr = tty
-	cmd.SysProcAttr = getSysProcAttr(cmdline, true)
+	cmd.SysProcAttr = process.SysProcAttr(cmdline, true)
 	return ppty, tty, nil
 }
 
@@ -309,6 +338,10 @@ func (e *HostEnvironment) exec(ctx context.Context, command []string, cmdline st
 	} else {
 		wd = e.Path
 	}
+	// Flush any buffered, not-yet-newline-terminated trailing line, as the docker backend
+	// does in waitForCommand, so the final line of a command's output is not lost.
+	defer common.FlushWriter(e.StdOut)
+
 	f, err := lookupPathHost(command[0], env, e.StdOut)
 	if err != nil {
 		return err
@@ -321,30 +354,11 @@ func (e *HostEnvironment) exec(ctx context.Context, command []string, cmdline st
 	cmd.Env = envList
 	cmd.Stderr = e.StdOut
 	cmd.Dir = wd
-	cmd.SysProcAttr = getSysProcAttr(cmdline, false)
+	cmd.SysProcAttr = process.SysProcAttr(cmdline, false)
 
-	// On Windows a step often launches a process tree (a shell that starts a
+	// Kills the step's whole tree on cancellation and bounds the post-exit I/O
-	// child which spawns further GUI or background processes). The default
+	// wait, so an orphan holding cmd's stdout pipe cannot hang cmd.Wait().
-	// context cancellation only kills the direct child, leaving the rest of the
+	treeKill := process.NewTreeKill(cmd)
-	// tree running; and because the orphans inherit cmd's stdout/stderr pipe,
-	// cmd.Wait() would block forever, hanging the runner. Kill the whole tree
-	// via a Job Object on cancellation, and bound the wait so a leftover pipe
-	// writer can never hang Wait indefinitely.
-	var killer atomic.Pointer[processKiller]
-	if runtime.GOOS == "windows" {
-		cmd.Cancel = func() error {
-			if k := killer.Load(); k != nil {
-				return k.Kill()
-			}
-			if cmd.Process != nil {
-				return cmd.Process.Kill()
-			}
-			return nil
-		}
-		// Once the step process has exited, give its I/O pipes at most this long
-		// to drain before Wait force-closes them and returns (Go's WaitDelay).
-		cmd.WaitDelay = 10 * time.Second
-	}
 
 	var ppty *os.File
 	var tty *os.File
@@ -375,18 +389,16 @@ func (e *HostEnvironment) exec(ctx context.Context, command []string, cmdline st
 	if err := cmd.Start(); err != nil {
 		return err
 	}
-	if runtime.GOOS == "windows" {
+	// Assign before the step's Killer so the step's job nests inside the group's;
-		// Assign the started process to a Job Object so cmd.Cancel can kill the
+	// cancellation still scopes to this step's tree.
-		// whole descendant tree. Children spawned afterwards are auto-included.
+	if err := e.processGroup(ctx).Assign(cmd.Process); err != nil {
-		// On failure (e.g. nested-job restrictions) we fall back to the default
+		common.Logger(ctx).Warnf("could not assign the step's process to the job's process group; a process it leaves behind may outlive the job: %v", err)
-		// single-process kill; WaitDelay + end-of-job cleanup still apply.
+	}
-		if k, kerr := newProcessKiller(cmd.Process); kerr != nil {
+	if k, kerr := treeKill.Capture(cmd.Process); kerr != nil {
 		common.Logger(ctx).Warnf("process tree kill setup failed, falling back to single-process kill: %v", kerr)
 	} else {
-			killer.Store(k)
 		defer k.Close()
 	}
-	}
 	err = cmd.Wait()
 	if err != nil {
 		var exitErr *exec.ExitError
@@ -429,6 +441,23 @@ func (e *HostEnvironment) UpdateFromEnv(srcPath string, env *map[string]string)
 	return parseEnvFile(e, srcPath, env)
 }
 
+// removeAll is a var so tests can substitute a blocking stub.
+var removeAll = os.RemoveAll
+
+// removeAllWithContext returns once the delete finishes or ctx is cancelled. On
+// cancellation the goroutine leaks: a delete inside a syscall cannot be
+// interrupted (see runWithTimeout).
+func removeAllWithContext(ctx context.Context, path string) error {
+	done := make(chan error, 1)
+	go func() { done <- removeAll(path) }()
+	select {
+	case err := <-done:
+		return err
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+}
+
 func removePathWithRetry(ctx context.Context, path string) error {
 	if path == "" {
 		return nil
@@ -448,25 +477,23 @@ func removePathWithRetry(ctx context.Context, path string) error {
 			case <-time.After(delay):
 			}
 		}
-		lastErr = os.RemoveAll(path)
+		lastErr = removeAllWithContext(ctx, path)
 		if lastErr == nil {
 			return nil
 		}
+		if errors.Is(lastErr, context.DeadlineExceeded) {
+			return lastErr
+		}
 	}
 	return lastErr
 }
 
-// buildWindowsWorkspaceKillScript builds a PowerShell command that `taskkill
+// buildWindowsWorkspaceKillScript builds a PowerShell command that taskkills
-// /T /F`s every process tree whose ExecutablePath or CommandLine references one
+// every process tree whose ExecutablePath or CommandLine references one of the
-// of the given absolute workspace dirs, releasing file handles for cleanup.
+// given workspace dirs, releasing file handles for cleanup. Win32_Process
-//
+// exposes both fields (Get-Process doesn't, wmic is deprecated); matching is on
-// Win32_Process is used because it exposes both ExecutablePath and CommandLine
+// the dir+separator prefix via ordinal String methods, so a name-prefix sibling
-// (Get-Process doesn't, wmic is deprecated). Both match the dir+separator
+// (job1 vs job10) is spared and path metacharacters stay literal.
-// prefix, so a sibling dir sharing a name prefix (job1 vs job10) is spared.
-// Ordinal String methods, not -like, so path metacharacters ([ ] ? *) stay
-// literal.
-//
-// Pure function so the quote-escaping can be unit-tested without PowerShell.
 func buildWindowsWorkspaceKillScript(dirs []string) string {
 	quoted := make([]string, len(dirs))
 	for i, d := range dirs {
@@ -502,9 +529,8 @@ func (e *HostEnvironment) terminateRunningProcesses(ctx context.Context) {
 
 	logger := common.Logger(ctx)
 
-	// Workspace dirs we own. Any process running from or referencing one is a
+	// Dirs we own; a process referencing one is a leftover. ToolCache is shared
-	// leftover job process. ToolCache is shared across jobs; Workdir only when
+	// across jobs, and Workdir may be a caller-owned checkout.
-	// we own it (else it's a caller-provided checkout, e.g. act local mode).
 	owned := []string{e.Path, e.TmpDir}
 	if e.CleanWorkdir {
 		owned = append(owned, e.Workdir)
@@ -531,25 +557,67 @@ func (e *HostEnvironment) terminateRunningProcesses(ctx context.Context) {
 	if err != nil {
 		logger.Debugf("workspace process-tree kill via PowerShell failed: %v output=%s", err, strings.TrimSpace(string(out)))
 	}
+
+	// Win32_Process exposes no working directory, so the scan above misses a
+	// process that merely runs in a workspace dir while pinning a handle on it.
+	if killed, err := process.KillProcessesWithCWDUnder(killCtx, dirs); err != nil {
+		logger.Debugf("workspace process kill by working directory reported errors: %v", err)
+	} else if killed > 0 {
+		logger.Debugf("terminated %d leftover process(es) by workspace working directory", killed)
+	}
+}
+
+// hostCleanupTimeout bounds each teardown phase so one stalled delete cannot
+// wedge the runner slot. A var so tests can shrink it.
+var hostCleanupTimeout = 30 * time.Second
+
+// runWithTimeout returns context.DeadlineExceeded once timeout elapses, leaking
+// the goroutine: a delete blocked in a syscall (AV filter driver, dead network
+// mount) cannot be interrupted, and leaking scratch state beats losing the
+// runner's capacity slot forever. The idle stale-dir sweep reclaims it later.
+func runWithTimeout(fn func(), timeout time.Duration) error {
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		fn()
+	}()
+	timer := time.NewTimer(timeout)
+	defer timer.Stop()
+	select {
+	case <-done:
+		return nil
+	case <-timer.C:
+		return context.DeadlineExceeded
+	}
 }
 
 func (e *HostEnvironment) Remove() common.Executor {
 	return func(ctx context.Context) error {
-		// Ensure any lingering child processes are ended before attempting
+		logger := common.Logger(ctx)
-		// to remove the workspace (Windows file locks otherwise prevent cleanup).
+
+		// End lingering processes before removing the workspace; on Windows their
+		// file locks block cleanup. Closing the group is deterministic, the scan a net.
+		if err := e.procGroup.Load().Close(); err != nil {
+			logger.Debugf("closing the job's process group failed: %v", err)
+		}
 		e.terminateRunningProcesses(ctx)
 
-		// Only removes per-job misc state. Must not remove the cache/toolcache root.
+		// Removes per-job misc state only, never the toolcache root. Bounded because
+		// CleanUp is a caller-supplied, typically unbounded os.RemoveAll.
 		if e.CleanUp != nil {
-			e.CleanUp()
+			logger.Debugf("running host environment cleanup callback")
+			if err := runWithTimeout(e.CleanUp, hostCleanupTimeout); err != nil {
+				logger.Warnf("host environment cleanup did not finish within %s; continuing job completion, scratch state may be leaked and is reclaimed by the idle stale-dir sweep", hostCleanupTimeout)
+			} else {
+				logger.Debugf("host environment cleanup callback finished")
+			}
 		}
 
 		// Detach: a cancelled ctx would skip removePathWithRetry's retries,
 		// which absorb Windows file-handle release lag after the kill above.
-		rmCtx, rmCancel := context.WithTimeout(context.Background(), 30*time.Second)
+		rmCtx, rmCancel := context.WithTimeout(context.Background(), hostCleanupTimeout)
 		defer rmCancel()
 
-		logger := common.Logger(ctx)
 		var errs []error
 		if err := removePathWithRetry(rmCtx, e.Path); err != nil {
 			logger.Warnf("failed to remove host misc state %s: %v", e.Path, err)
@@ -561,8 +629,14 @@ func (e *HostEnvironment) Remove() common.Executor {
 				errs = append(errs, err)
 			}
 		}
+		for _, err := range errs {
+			if !errors.Is(err, context.DeadlineExceeded) {
 				return errors.Join(errs...)
 			}
+		}
+		// Teardown timed out; warned above. Do not fail job completion over it.
+		return nil
+	}
 }
 
 func (e *HostEnvironment) ToContainerPath(path string) string {

act/container/host_environment_test.go

@@ -15,6 +15,7 @@ import (
 	"runtime"
 	"strings"
 	"testing"
+	"time"
 
 	"gitea.com/gitea/runner/act/common"
 
@@ -188,6 +189,118 @@ func TestHostEnvironmentRemoveCleansWorkdirWhenOwned(t *testing.T) {
 	assert.ErrorIs(t, err, os.ErrNotExist)
 }
 
+func TestRemoveAllWithContextDoesNotHangOnStuckDelete(t *testing.T) {
+	release := make(chan struct{})
+	stubDone := make(chan struct{})
+
+	orig := removeAll
+	removeAll = func(string) error {
+		defer close(stubDone)
+		<-release
+		return nil
+	}
+	// removeAllWithContext intentionally leaks the delete goroutine on timeout,
+	// and that goroutine still references removeAll. Unblock it and wait for it
+	// to return before restoring the var, so the restore can't race the read.
+	t.Cleanup(func() {
+		close(release)
+		<-stubDone
+		removeAll = orig
+	})
+
+	ctx, cancel := context.WithTimeout(context.Background(), 100*time.Millisecond)
+	defer cancel()
+
+	err := removeAllWithContext(ctx, t.TempDir())
+	require.ErrorIs(t, err, context.DeadlineExceeded)
+}
+
+// TestHostEnvironmentRemoveDoesNotHangOnStuckCleanUp guards against a stalled
+// CleanUp callback (e.g. an os.RemoveAll blocked by an AV/EDR filter driver or
+// an unresponsive mount) wedging the runner slot forever at "Cleaning up
+// container". Remove must time out the callback and complete job teardown.
+func TestHostEnvironmentRemoveDoesNotHangOnStuckCleanUp(t *testing.T) {
+	// Keep the suite fast: shrink the per-phase teardown timeout for this test.
+	orig := hostCleanupTimeout
+	hostCleanupTimeout = 100 * time.Millisecond
+	t.Cleanup(func() { hostCleanupTimeout = orig })
+
+	logger := logrus.New()
+	ctx := common.WithLogger(context.Background(), logrus.NewEntry(logger))
+	base := t.TempDir()
+	path := filepath.Join(base, "misc", "hostexecutor")
+	require.NoError(t, os.MkdirAll(path, 0o700))
+
+	release := make(chan struct{})
+	t.Cleanup(func() { close(release) }) // unblock the leaked goroutine at test end
+
+	e := &HostEnvironment{
+		Path: path,
+		CleanUp: func() {
+			<-release // simulate a delete syscall stuck indefinitely
+		},
+		StdOut: os.Stdout,
+	}
+
+	done := make(chan error, 1)
+	go func() { done <- e.Remove()(ctx) }()
+
+	select {
+	case err := <-done:
+		require.NoError(t, err)
+	case <-time.After(10 * time.Second):
+		t.Fatal("Remove() hung on a stuck CleanUp callback")
+	}
+}
+
+// TestHostEnvironmentRemoveDoesNotHangOnStuckPathRemoval guards against a
+// stalled os.RemoveAll on the misc/workspace paths (same AV/EDR wedge as
+// #1023) wedging job completion after the CleanUp callback has already timed
+// out or finished.
+func TestHostEnvironmentRemoveDoesNotHangOnStuckPathRemoval(t *testing.T) {
+	origTimeout := hostCleanupTimeout
+	hostCleanupTimeout = 100 * time.Millisecond
+	t.Cleanup(func() { hostCleanupTimeout = origTimeout })
+
+	release := make(chan struct{})
+	stubDone := make(chan struct{})
+
+	origRemoveAll := removeAll
+	removeAll = func(string) error {
+		defer close(stubDone)
+		<-release
+		return nil
+	}
+	// The stuck delete goroutine outlives the timed-out Remove and still reads
+	// removeAll; unblock it and wait before restoring to avoid a restore/read race.
+	t.Cleanup(func() {
+		close(release)
+		<-stubDone
+		removeAll = origRemoveAll
+	})
+
+	logger := logrus.New()
+	ctx := common.WithLogger(context.Background(), logrus.NewEntry(logger))
+	base := t.TempDir()
+	path := filepath.Join(base, "misc", "hostexecutor")
+	require.NoError(t, os.MkdirAll(path, 0o700))
+
+	e := &HostEnvironment{
+		Path:   path,
+		StdOut: os.Stdout,
+	}
+
+	done := make(chan error, 1)
+	go func() { done <- e.Remove()(ctx) }()
+
+	select {
+	case err := <-done:
+		require.NoError(t, err)
+	case <-time.After(10 * time.Second):
+		t.Fatal("Remove() hung on a stuck path removal")
+	}
+}
+
 func TestBuildWindowsWorkspaceKillScript(t *testing.T) {
 	t.Run("single dir", func(t *testing.T) {
 		s := buildWindowsWorkspaceKillScript([]string{`C:\workspace\job1`})

act/container/linux_container_environment_extensions.go

@@ -66,12 +66,15 @@ func (*LinuxContainerEnvironmentExtensions) JoinPathVariable(paths ...string) st
 	return strings.Join(paths, ":")
 }
 
+// DefaultToolCache is where the runner mounts the tool cache inside job containers.
+const DefaultToolCache = "/opt/hostedtoolcache"
+
 func (*LinuxContainerEnvironmentExtensions) GetRunnerContext(ctx context.Context) map[string]any {
 	return map[string]any{
 		"os":         "Linux",
 		"arch":       RunnerArch(ctx),
 		"temp":       "/tmp",
-		"tool_cache": "/opt/hostedtoolcache",
+		"tool_cache": DefaultToolCache,
 	}
 }
 

act/container/parse_env_file.go

@@ -13,6 +13,9 @@ import (
 	"strings"
 
 	"gitea.com/gitea/runner/act/common"
+
+	"golang.org/x/text/encoding/unicode"
+	"golang.org/x/text/transform"
 )
 
 func parseEnvFile(e Container, srcPath string, env *map[string]string) common.Executor {
@@ -28,11 +31,19 @@ func parseEnvFile(e Container, srcPath string, env *map[string]string) common.Ex
 		if err != nil && err != io.EOF {
 			return err
 		}
-		s := bufio.NewScanner(reader)
+		// Decode by BOM: Windows PowerShell 5.1 redirection writes UTF-16, and some
+		// tools emit a UTF-8 BOM. Without a BOM the file is read as UTF-8, as before.
+		decoded := transform.NewReader(reader, unicode.BOMOverride(unicode.UTF8.NewDecoder()))
+
+		s := bufio.NewScanner(decoded)
 		// Default 64 KiB max token size is too small for realistic env-file lines; allow up to 16 MiB.
 		s.Buffer(make([]byte, 0, 64*1024), 16*1024*1024)
 		for s.Scan() {
 			line := s.Text()
+			// GitHub's runner ignores blank lines
+			if strings.TrimSpace(line) == "" {
+				continue
+			}
 			singleLineEnv := strings.Index(line, "=")
 			multiLineEnv := strings.Index(line, "<<")
 			if singleLineEnv != -1 && (multiLineEnv == -1 || singleLineEnv < multiLineEnv) {