fix: clean up job network and container when container start fails (#986)

The teardown that removes a job's per-job network and container runs as a `Finally` on the step pipeline in `newJobExecutor`, which only executes after a successful start. When the start itself fails (e.g. a `docker cp` error from a buggy daemon), that `Finally` is skipped, so the network and container leak until Docker's address pool is exhausted and later jobs can no longer create networks.

This tears them down in `startContainer` when the start returns an error, reusing the existing `cleanUpJobContainer` teardown.

Exposed by the daemon regression in https://gitea.com/gitea/runner/issues/981, where every failed `docker cp` leaked a per-job network.

---
This PR was written with the help of Claude Opus 4.7

Reviewed-on: https://gitea.com/gitea/runner/pulls/986
Reviewed-by: Nicolas <bircni@icloud.com>
Co-authored-by: silverwind <me@silverwind.io>
Co-committed-by: silverwind <me@silverwind.io>

act/runner/run_context.go

@@ -601,10 +601,34 @@ func (rc *RunContext) interpolateOutputs() common.Executor {
 
 func (rc *RunContext) startContainer() common.Executor {
 	return func(ctx context.Context) error {
+		var err error
 		if rc.IsHostEnv(ctx) {
-			return rc.startHostEnvironment()(ctx)
+			err = rc.startHostEnvironment()(ctx)
+		} else {
+			err = rc.startJobContainer()(ctx)
 		}
-		return rc.startJobContainer()(ctx)
+		if err != nil {
+			// The job executor's teardown only runs after a successful start, so a failed
+			// start would otherwise leak the per-job network and container.
+			rc.cleanupFailedStart(ctx)
+		}
+		return err
+	}
+}
+
+func (rc *RunContext) cleanupFailedStart(ctx context.Context) {
+	if rc.cleanUpJobContainer == nil {
+		return
+	}
+	cleanCtx := ctx
+	if ctx.Err() != nil {
+		// the start likely failed because ctx was cancelled, detach so teardown still runs
+		var cancel context.CancelFunc
+		cleanCtx, cancel = context.WithTimeout(common.WithLogger(context.Background(), common.Logger(ctx)), time.Minute)
+		defer cancel()
+	}
+	if err := rc.cleanUpJobContainer(cleanCtx); err != nil {
+		common.Logger(ctx).Errorf("Error while cleaning up after failed container start for job %s: %v", rc.JobName, err)
 	}
 }
 

act/runner/run_context_test.go

@@ -19,6 +19,7 @@ import (
 
 	log "github.com/sirupsen/logrus"
 	assert "github.com/stretchr/testify/assert"
+	require "github.com/stretchr/testify/require"
 	yaml "go.yaml.in/yaml/v4"
 )
 
@@ -659,3 +660,53 @@ func TestPrintStartJobContainerGroupGolden(t *testing.T) {
 	}, "\n")
 	assert.Equal(t, want, buf.String())
 }
+
+func TestRunContext_cleanupFailedStart(t *testing.T) {
+	type ctxKey string
+	const sentinel = ctxKey("sentinel")
+
+	// the fresh context is cancelled via defer on return, so capture state inside the stub
+	type capture struct {
+		calls    int
+		err      error
+		sentinel any
+	}
+	newRC := func(c *capture) *RunContext {
+		return &RunContext{
+			JobName: "job",
+			cleanUpJobContainer: func(ctx context.Context) error {
+				c.calls++
+				c.err = ctx.Err()
+				c.sentinel = ctx.Value(sentinel)
+				return nil
+			},
+		}
+	}
+
+	t.Run("runs teardown on the live context", func(t *testing.T) {
+		var c capture
+		ctx := context.WithValue(context.Background(), sentinel, "v")
+
+		newRC(&c).cleanupFailedStart(ctx)
+
+		assert.Equal(t, 1, c.calls)
+		require.NoError(t, c.err)
+		assert.Equal(t, "v", c.sentinel)
+	})
+
+	t.Run("falls back to a fresh context when the input is done", func(t *testing.T) {
+		var c capture
+		ctx, cancel := context.WithCancel(context.WithValue(context.Background(), sentinel, "v"))
+		cancel()
+
+		newRC(&c).cleanupFailedStart(ctx)
+
+		assert.Equal(t, 1, c.calls)
+		require.NoError(t, c.err)
+		assert.Nil(t, c.sentinel)
+	})
+
+	t.Run("no-op when there is nothing to clean up", func(t *testing.T) {
+		assert.NotPanics(t, func() { (&RunContext{}).cleanupFailedStart(context.Background()) })
+	})
+}