fix: keep the shared action cache as extracted, re-fetch damaged trees (#1352)

The toolkit patch rewrote bundles in place in the shared action cache. Since https://gitea.com/gitea/runner/pulls/1316 trees are never refreshed, so a bundle left truncated failed every later job using it.

Patched bundles now only overwrite the job container's copy, so the runner no longer writes to an extracted tree.

Extraction records each file's size beside the tree, and a tree that no longer matches is fetched again, offline mode included, as a missing one always was. This also catches damage from outside the runner, such as storage losing a write, for a stat per file, a fraction of the copy each stage already makes. Trees from older runners have no record and download once more, except in offline mode, which keeps using them as before.

Generated with Claude Code.

Reviewed-on: https://gitea.com/gitea/runner/pulls/1352
Reviewed-by: Lunny Xiao <[email protected]>
Reviewed-by: bircni <[email protected]>
Co-authored-by: Zettat123 <[email protected]>

act/common/git/fetch_action_test.go

@@ -15,6 +15,7 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
+	"runtime"
 	"strings"
 	"sync"
 	"sync/atomic"
@@ -206,10 +207,17 @@ func TestFetchAction(t *testing.T) {
 		_, offlineSha, err := FetchAction(t.Context(), input)
 		require.NoError(t, err)
 		assert.Equal(t, movedSha, offlineSha)
+		require.NoError(t, os.Truncate(filepath.Join(moved, "action.yml"), 4)) // fetched again, offline or not
+		_, _, err = FetchAction(t.Context(), input)
+		require.NoError(t, err)
+		require.NoError(t, os.Remove(moved+treeSizes)) // from an older runner, used as it is offline
+		reused, _, err := FetchAction(t.Context(), input)
+		require.NoError(t, err)
+		assert.Equal(t, moved, reused)
 
 		requests, headers := served()
-		assert.Equal(t, 3, requests)
-		require.Len(t, headers, 3)
+		assert.Equal(t, 4, requests)
+		require.Len(t, headers, 4)
 		assert.Equal(t, "Basic "+base64.StdEncoding.EncodeToString([]byte("token:secret")), headers[0].Get("Authorization"))
 		assert.Empty(t, headers[0].Get("If-None-Match"))
 		assert.Equal(t, fmt.Sprintf(`W/"%s-tar.gz-true"`, sha), headers[1].Get("If-None-Match"))
@@ -233,6 +241,68 @@ func TestFetchAction(t *testing.T) {
 		assert.NoDirExists(t, filepath.Join(input.Dir, "git"))
 	})
 
+	t.Run("downloads again over a tree whose files changed size or that has no record, not under a job copying it", func(t *testing.T) {
+		sha := gitRevParse(t, remote, "main")
+		for _, mode := range []string{"prefixed", "none"} {
+			serve(mode, 0)
+			input := newInput(sha)
+			tree, _, err := FetchAction(t.Context(), input)
+			require.NoError(t, err, mode)
+			yml := filepath.Join(tree, "action.yml")
+			require.NoError(t, os.WriteFile(filepath.Join(tree, "trampoline.js"), nil, 0o644)) // the runner's own, no damage
+			_, _, err = FetchAction(t.Context(), input)
+			require.NoError(t, err, mode)
+			for _, damage := range []func() error{
+				func() error { return os.Truncate(yml, 4) },
+				func() error { return os.Remove(tree + treeSizes) },
+			} {
+				require.NoError(t, damage(), mode)
+				_, _, err := FetchAction(t.Context(), input)
+				require.NoError(t, err, mode)
+				body, err := os.ReadFile(yml)
+				require.NoError(t, err, mode)
+				assert.Equal(t, "name: action\n", string(body), mode)
+			}
+			_, headers := served()
+			assert.Len(t, headers, 3, mode)
+		}
+
+		serve("prefixed", 0)
+		input := newInput(sha)
+		tree, _, err := FetchAction(t.Context(), input)
+		require.NoError(t, err)
+		yml := filepath.Join(tree, "action.yml")
+		require.NoError(t, os.Truncate(yml, 4))
+		unlock, err := AcquireCloneLock(t.Context(), tree)
+		require.NoError(t, err)
+		release := sync.OnceFunc(unlock)
+		defer release()
+		waiting := make(chan struct{})
+		lockTree = func(ctx context.Context, dir string) (func(), error) {
+			close(waiting)
+			return AcquireCloneLock(ctx, dir)
+		}
+		t.Cleanup(func() { lockTree = AcquireCloneLock })
+		done := make(chan error, 1)
+		go func() {
+			_, _, err := FetchAction(t.Context(), input)
+			done <- err
+		}()
+		select {
+		case err := <-done:
+			t.Fatalf("FetchAction replaced the tree without waiting for the job holding it: %v", err)
+		case <-waiting:
+		}
+		info, err := os.Stat(yml)
+		require.NoError(t, err)
+		assert.Equal(t, int64(4), info.Size(), "the tree a job holds is left alone")
+		release()
+		require.NoError(t, <-done)
+		body, err := os.ReadFile(yml)
+		require.NoError(t, err)
+		assert.Equal(t, "name: action\n", string(body))
+	})
+
 	t.Run("retries a failing download and drops the token on a redirect to another origin", func(t *testing.T) {
 		serve("redirect", 2)
 		input := newInput("main")
@@ -306,10 +376,30 @@ func TestExtractTreeConfinesEntries(t *testing.T) {
 			require.NoError(t, writer.Close())
 
 			dir := t.TempDir()
-			_, err := extractTree(&archive, dir, false)
+			_, err := extractTree(t.Context(), &archive, dir, false)
 			require.Error(t, err)
 			assert.NoFileExists(t, filepath.Join(dir, "escape"))
 			assert.NoFileExists(t, filepath.Join(outside, "escape"))
 		})
 	}
 }
+
+func TestTreeIntactFindsNamesThatAreNotUTF8(t *testing.T) {
+	if runtime.GOOS != "linux" {
+		t.Skip("macOS and Windows store no name that is not UTF-8")
+	}
+	sha := strings.Repeat("a", 40)
+	var archive bytes.Buffer
+	writer := tar.NewWriter(&archive)
+	require.NoError(t, writer.WriteHeader(&tar.Header{Typeflag: tar.TypeXGlobalHeader, PAXRecords: map[string]string{"comment": sha}}))
+	require.NoError(t, writer.WriteHeader(&tar.Header{Name: "caf\xe9.js", Typeflag: tar.TypeReg, Mode: 0o644, Size: 1, Format: tar.FormatGNU}))
+	_, err := writer.Write([]byte("x"))
+	require.NoError(t, err)
+	require.NoError(t, writer.Close())
+
+	dir := t.TempDir()
+	_, err = extractTree(t.Context(), &archive, dir, false)
+	require.NoError(t, err)
+	require.FileExists(t, filepath.Join(dir, sha, "caf\xe9.js"))
+	assert.True(t, treeIntact(dir, sha))
+}