fix(deploy): retry the post-commit deploy fence write

Author: raisedadead

internal/handler/deployfence.go

@@ -17,7 +17,10 @@ func (h *Handlers) fenceFinalizedDeploy(ctx context.Context, site sitekey.Slug,
 				"later upload with the same token can overwrite what is live")
 		return
 	}
-	if err := h.DeployFence.MarkDeployFinalized(ctx, site, deployID, mode, h.DeployJWTTTL); err != nil {
+	err := retryIdempotentCommit(ctx, func(ctx context.Context) error {
+		return h.DeployFence.MarkDeployFinalized(ctx, site, deployID, mode, h.DeployJWTTTL)
+	})
+	if err != nil {
 		slog.ErrorContext(ctx, "deploy.fence.failed", "site", site, "deployId", deployID, "mode", mode, "err", err,
 			"detail", "neither the upload fence nor the mode fence was written, so the permit can still "+
 				"overwrite this deploy or repoint its alias until it expires")

internal/handler/deployfence_retry_test.go

@@ -0,0 +1,45 @@
+package handler
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+
+	"github.com/freeCodeCamp/artemis/internal/sitekey"
+)
+
+type flakyDeployFence struct {
+	failures int
+	attempts int
+	marked   bool
+}
+
+func (f *flakyDeployFence) MarkDeployFinalized(context.Context, sitekey.Slug, string, string, time.Duration) error {
+	f.attempts++
+	if f.attempts <= f.failures {
+		return errors.New("dial tcp 10.43.0.1:6379: connect: connection refused")
+	}
+	f.marked = true
+	return nil
+}
+
+func (f *flakyDeployFence) IsDeployFinalized(context.Context, sitekey.Slug, string) (bool, error) {
+	return f.marked, nil
+}
+
+func (f *flakyDeployFence) IsDeployModeFinalized(context.Context, sitekey.Slug, string, string) (bool, error) {
+	return f.marked, nil
+}
+
+func TestFenceFinalizedDeployRetriesATransientWriteFailure(t *testing.T) {
+	fence := &flakyDeployFence{failures: indexCommitAttempts - 1}
+	h := &Handlers{DeployFence: fence, DeployJWTTTL: time.Minute}
+
+	h.fenceFinalizedDeploy(context.Background(), "www", "20260420-141522-abc1234", "preview")
+
+	require.True(t, fence.marked, "the fence must be written after a transient failure")
+	require.Equal(t, indexCommitAttempts, fence.attempts)
+}