feat(indexworker): add max users threshold for rollout

Author: fadymak

hack/test.env

@@ -132,6 +132,7 @@ GOTRUE_SAML_ENABLED="true"
 GOTRUE_SAML_PRIVATE_KEY="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"
 GOTRUE_MAX_VERIFIED_FACTORS=10
 GOTRUE_SMS_TEST_OTP_VALID_UNTIL=""
+GOTRUE_INDEX_WORKER_MAX_USERS_THRESHOLD=1000000
 GOTRUE_SECURITY_DB_ENCRYPTION_ENCRYPT=true
 GOTRUE_SECURITY_DB_ENCRYPTION_ENCRYPTION_KEY_ID=abc
 GOTRUE_SECURITY_DB_ENCRYPTION_ENCRYPTION_KEY=pwFoiPyybQMqNmYVN0gUnpbfpGQV2sDv9vp0ZAxi_Y4

internal/conf/configuration.go

@@ -1359,5 +1359,6 @@ func (t *SmsProviderConfiguration) IsTwilioVerifyProvider() bool {
 
 // IndexWorkerConfiguration holds the configuration for database indexes.
 type IndexWorkerConfiguration struct {
-	EnsureUserSearchIndexesExist bool `json:"ensure_user_search_indexes_exist" split_words:"true" default:"false"`
+	EnsureUserSearchIndexesExist bool  `json:"ensure_user_search_indexes_exist" split_words:"true" default:"false"`
+	MaxUsersThreshold            int64 `json:"max_users_threshold" split_words:"true" default:"0"`
 }

internal/indexworker/indexworker.go

@@ -147,14 +147,24 @@ func CreateIndexes(ctx context.Context, config *conf.GlobalConfiguration, le *lo
 		}
 	}
 
-	userCount, err := getApproximateUserCount(db, config.DB.Namespace)
-	if err != nil {
-		le.WithError(err).Warn("Failed to get approximate user count, proceeding with index creation")
+	if config.IndexWorker.MaxUsersThreshold > 0 {
+		userCount, err := getApproximateUserCount(db, config.DB.Namespace)
+		if err != nil {
+			le.WithError(err).Warn("Failed to get approximate user count, proceeding with index creation")
+		}
+		if userCount > config.IndexWorker.MaxUsersThreshold {
+			le.WithFields(logrus.Fields{
+				"code":                "index_creation_skipped",
+				"user_count":          userCount,
+				"max_users_threshold": config.IndexWorker.MaxUsersThreshold,
+			}).Info("Skipping index creation because user count exceeds threshold")
+			return nil
+		}
+		le.WithFields(logrus.Fields{
+			"code":       "index_creation_starting",
+			"user_count": userCount,
+		}).Info("Starting index creation")
 	}
-	le.WithFields(logrus.Fields{
-		"code":       "index_creation_starting",
-		"user_count": userCount,
-	}).Info("Starting index creation")
 
 	// First, clean up any invalid indexes from previous interrupted attempts
 	dropInvalidIndexes(db, le, config.DB.Namespace, indexNames)

internal/indexworker/indexworker_test.go

@@ -15,23 +15,26 @@ import (
 	"github.com/stretchr/testify/require"
 	"github.com/stretchr/testify/suite"
 	"github.com/supabase/auth/internal/conf"
+	"github.com/supabase/auth/internal/models"
 	"github.com/supabase/auth/internal/storage"
 )
 
 type IndexWorkerTestSuite struct {
 	suite.Suite
-	config    *conf.GlobalConfiguration
-	db        *storage.Connection
-	popDB     *pop.Connection
-	namespace string
-	logger    *logrus.Entry
+	config            *conf.GlobalConfiguration
+	db                *storage.Connection
+	popDB             *pop.Connection
+	namespace         string
+	logger            *logrus.Entry
+	maxUsersThreshold int64
 }
 
 func (ts *IndexWorkerTestSuite) SetupSuite() {
 	// Load test configuration
 	config, err := conf.LoadGlobal("../../hack/test.env")
 	require.NoError(ts.T(), err)
 	ts.config = config
+	ts.maxUsersThreshold = config.IndexWorker.MaxUsersThreshold
 	ts.namespace = config.DB.Namespace
 	ts.logger = logrus.NewEntry(logrus.New())
 	ts.logger.Logger.SetLevel(logrus.DebugLevel)
@@ -69,8 +72,9 @@ func (ts *IndexWorkerTestSuite) TearDownSuite() {
 }
 
 func (ts *IndexWorkerTestSuite) SetupTest() {
-	// Clean up before each test
+	models.TruncateAll(ts.db)
 	ts.cleanupIndexes()
+	ts.config.IndexWorker.MaxUsersThreshold = ts.maxUsersThreshold
 }
 
 func (ts *IndexWorkerTestSuite) cleanupIndexes() {
@@ -282,6 +286,55 @@ func getIndexNames(indexes []struct {
 	return names
 }
 
+func (ts *IndexWorkerTestSuite) TestMaxUsersThresholdSkipsIndexCreation() {
+	ctx := context.Background()
+
+	// SetupTest already called TruncateAll, so the users table is empty.
+	// Insert test users so the approximate count exceeds the threshold.
+	for i := 0; i < 5; i++ {
+		insertQuery := fmt.Sprintf(
+			`INSERT INTO %q.users (instance_id, id, aud, role, email, encrypted_password, created_at, updated_at) VALUES ('00000000-0000-0000-0000-000000000000', gen_random_uuid(), 'authenticated', 'authenticated', 'threshold_test_%d@example.com', '', now(), now())`,
+			ts.namespace, i,
+		)
+		require.NoError(ts.T(), ts.db.RawQuery(insertQuery).Exec())
+	}
+	// Update pg_class.reltuples so getApproximateUserCount reflects the inserts
+	analyzeQuery := fmt.Sprintf(`ANALYZE %q.users`, ts.namespace)
+	require.NoError(ts.T(), ts.db.RawQuery(analyzeQuery).Exec())
+
+	ts.config.IndexWorker.MaxUsersThreshold = 1
+
+	indexes := getUsersIndexes(ts.namespace)
+	existingIndexes, err := getIndexStatuses(ts.popDB, ts.namespace, getIndexNames(indexes))
+	require.NoError(ts.T(), err)
+	assert.Empty(ts.T(), existingIndexes, "No indexes should exist before the test")
+
+	err = CreateIndexes(ctx, ts.config, ts.logger)
+	require.NoError(ts.T(), err)
+
+	existingIndexes, err = getIndexStatuses(ts.popDB, ts.namespace, getIndexNames(indexes))
+	require.NoError(ts.T(), err)
+	assert.Empty(ts.T(), existingIndexes, "No indexes should be created when user count exceeds threshold")
+}
+
+func (ts *IndexWorkerTestSuite) TestMaxUsersThresholdZeroCreatesIndexes() {
+	ctx := context.Background()
+
+	ts.config.IndexWorker.MaxUsersThreshold = 0
+
+	err := CreateIndexes(ctx, ts.config, ts.logger)
+	require.NoError(ts.T(), err)
+
+	indexes := getUsersIndexes(ts.namespace)
+	existingIndexes, err := getIndexStatuses(ts.popDB, ts.namespace, getIndexNames(indexes))
+	require.NoError(ts.T(), err)
+	assert.Equal(ts.T(), len(indexes), len(existingIndexes), "All indexes should be created when threshold is 0 (no limit)")
+	for _, idx := range existingIndexes {
+		assert.True(ts.T(), idx.IsValid, "Index %s should be valid", idx.IndexName)
+		assert.True(ts.T(), idx.IsReady, "Index %s should be ready", idx.IndexName)
+	}
+}
+
 // TestCreateIndexesWithInvalidIndexes tests that CreateIndexes can recover from invalid indexes
 // This test simulates a scenario where indexes become invalid (e.g., from interrupted CONCURRENT creation)
 // and verifies that CreateIndexes properly handles them by dropping and recreating.