blobfixture: add PreventGC flag

pkg/roachprod/blobfixture/metadata.go

@@ -43,6 +43,11 @@ type FixtureMetadata struct {
 
 	// FingerprintTime is the aost used by the fingerprint command.
 	FingerprintTime string `json:"fingerprint_time,omitempty"`
+
+	// LastFailureAt indicates that some test using this fixture has failed and we
+	// want to prevent GC of this fixture for investigation purposes. If this
+	// field is set, the fixture will not be GC'd until a week after this time.
+	LastFailureAt *time.Time `json:"last_failure_at,omitempty"`
 }
 
 func (f *FixtureMetadata) MarshalJson() ([]byte, error) {
@@ -113,6 +118,10 @@ func fixturesToGc(gcAt time.Time, allFixtures []FixtureMetadata) []fixtureToDele
 	// made ready more than 24 hours ago.
 	obsoleteThreshold := gcAt.Add(-24 * time.Hour)
 
+	// A fixture that has had a test failure within the past week is not eligible
+	// for GC.
+	failureGCThreshold := gcAt.Add(-7 * 24 * time.Hour)
+
 	toDelete := []fixtureToDelete{}
 
 	byKind := make(map[string][]FixtureMetadata)
@@ -144,10 +153,15 @@ func fixturesToGc(gcAt time.Time, allFixtures []FixtureMetadata) []fixtureToDele
 		// NOTE: starting at 1 because index 0 is the most recent fixture and is
 		// not eligible for garbage collection.
 		for i := 1; i < len(fixtures); i++ {
+			fixture := fixtures[i]
 			successor := fixtures[i-1]
 			if successor.ReadyAt.Before(obsoleteThreshold) {
+				if fixture.LastFailureAt != nil && !fixture.LastFailureAt.Before(failureGCThreshold) {
+					// Fixture has had a test failure within the past week, skip gc.
+					continue
+				}
 				toDelete = append(toDelete, fixtureToDelete{
-					metadata: fixtures[i],
+					metadata: fixture,
 					reason:   fmt.Sprintf("fixture '%s' is was mode obsolete by '%s' at '%s'", fixtures[i].DataPath, successor.DataPath, successor.ReadyAt),
 				})
 			}

pkg/roachprod/blobfixture/registry.go

@@ -167,6 +167,22 @@ func (r *Registry) GC(ctx context.Context, l *logger.Logger) error {
 	return nil
 }
 
+// MarkFailure marks the fixture at the given metadata path as having resulted
+// in a test failure. This prevents the fixture from being garbage collected,
+// providing time for investigation.
+func (r *Registry) MarkFailure(ctx context.Context, l *logger.Logger, metadataPath string) error {
+	setTime := r.clock()
+	err := r.updateMetadata(metadataPath, func(m *FixtureMetadata) error {
+		m.LastFailureAt = &setTime
+		return nil
+	})
+	if err != nil {
+		return err
+	}
+	l.Printf("fixture '%s' marked last failure at '%s'", metadataPath, setTime)
+	return nil
+}
+
 func (r *Registry) Close() {
 	_ = r.storage.Close()
 }
@@ -263,6 +279,45 @@ func (r *Registry) upsertMetadata(metadata FixtureMetadata) error {
 	return writer.Close()
 }
 
+func (r *Registry) updateMetadata(
+	metadataPath string, update func(metadata *FixtureMetadata) error,
+) error {
+	ctx := context.Background()
+	json, err := r.maybeReadFile(ctx, metadataPath)
+	if err != nil {
+		return errors.Wrap(err, "failed to read metadata for update")
+	}
+	if json == nil {
+		return errors.New("metadata does not exist for update")
+	}
+
+	metadata := &FixtureMetadata{}
+	if err := metadata.UnmarshalJson(json); err != nil {
+		return errors.Wrap(err, "failed to unmarshal metadata for update")
+	}
+
+	if err := update(metadata); err != nil {
+		return errors.Wrap(err, "failed to update metadata")
+	}
+
+	updatedJson, err := metadata.MarshalJson()
+	if err != nil {
+		return errors.Wrap(err, "failed to marshal updated metadata")
+	}
+
+	writer, err := r.storage.Writer(ctx, metadata.MetadataPath)
+	if err != nil {
+		return errors.Wrap(err, "failed to create writer for updated metadata")
+	}
+
+	if _, err := writer.Write(updatedJson); err != nil {
+		_ = writer.Close()
+		return errors.Wrap(err, "failed to write updated metadata")
+	}
+
+	return writer.Close()
+}
+
 func (r *Registry) deleteMetadata(metadata FixtureMetadata) error {
 	return errors.Wrap(r.storage.Delete(context.Background(), metadata.MetadataPath), "failed to delete metadata")
 }

pkg/roachprod/blobfixture/registry_test.go

@@ -58,6 +58,7 @@ func TestFixtureRegistry(t *testing.T) {
 		kind           string
 		createdAt      time.Time
 		readyAt        time.Time
+		lastFailureAt  time.Time
 		isLatestOfKind bool
 		survivesGC     bool
 	}
@@ -127,6 +128,35 @@ func TestFixtureRegistry(t *testing.T) {
 			survivesGC:     true,
 			isLatestOfKind: false,
 		},
+		{
+			// This fixture was marked with a failure long ago, so it will be deleted
+			// despite the flag.
+			kind:           "kind-marked-failure",
+			createdAt:      makeTime(-10),
+			readyAt:        makeTime(-9),
+			lastFailureAt:  makeTime(-8),
+			survivesGC:     false,
+			isLatestOfKind: false,
+		},
+		{
+			// This fixture was marked with a failure recently, so despite the fact
+			// that it has been obsolete for more than a day, it will not be
+			// deleted.
+			kind:           "kind-marked-failure",
+			createdAt:      makeTime(-6),
+			readyAt:        makeTime(-5),
+			lastFailureAt:  makeTime(-4),
+			survivesGC:     true,
+			isLatestOfKind: false,
+		},
+		{
+			// This is the most recent fixture of its kind, so it will not be deleted.
+			kind:           "kind-marked-failure",
+			createdAt:      makeTime(-1),
+			readyAt:        makeTime(-0.5),
+			survivesGC:     true,
+			isLatestOfKind: true,
+		},
 	}
 
 	type fixturesCreated struct {
@@ -169,6 +199,11 @@ func TestFixtureRegistry(t *testing.T) {
 			require.NoError(t, handle.SetReadyAt(ctx))
 		}
 
+		if !f.lastFailureAt.IsZero() {
+			now = f.lastFailureAt
+			require.NoError(t, registry.MarkFailure(ctx, l, metadata.MetadataPath))
+		}
+
 		created = append(created, fixturesCreated{
 			fixture:  f,
 			metadata: metadata,