feat: enable retention period for warehouse uploads

jobsdb/jobsdb.go

@@ -1483,8 +1483,9 @@
 
 	_, err = stmt.Exec()
 	if err != nil {
-		pqError, ok := err.(*pq.Error)
-		if ok && pqError.Code == pq.ErrorCode("42P01") {
+		var pqError *pq.Error
+		ok := errors.As(err, &pqError)
+		if ok && pqError.Code == ("42P01") {
 			jd.logger.Infof("[%s] sql statement(%s) exec failed because table doesn't exist", jd.tablePrefix, sqlStatement)
 			_, err = tx.Exec(rollbackSql)
 			jd.assertError(err)
@@ -3178,7 +3179,7 @@
 	var job JobT
 	sqlStatement := fmt.Sprintf(`SELECT %[1]s.job_id, %[1]s.uuid, %[1]s.user_id, %[1]s.parameters, %[1]s.custom_val, %[1]s.event_payload, %[1]s.created_at, %[1]s.expire_at FROM %[1]s WHERE %[1]s.job_id = %[2]d`, dsList[len(dsList)-1].JobTable, maxID)
 	err := jd.dbHandle.QueryRow(sqlStatement).Scan(&job.JobID, &job.UUID, &job.UserID, &job.Parameters, &job.CustomVal, &job.EventPayload, &job.CreatedAt, &job.ExpireAt)
-	if err != nil && err != sql.ErrNoRows {
+	if err != nil && !errors.Is(err, sql.ErrNoRows) {
 		jd.assertError(err)
 	}
 	return &job

warehouse/app.go

@@ -11,15 +11,13 @@ import (
 	"sync/atomic"
 	"time"
 
-	"golang.org/x/sync/errgroup"
-
-	"github.com/rudderlabs/rudder-go-kit/sqlutil"
-
 	"github.com/cenkalti/backoff/v4"
+	"golang.org/x/sync/errgroup"
 
 	"github.com/rudderlabs/rudder-go-kit/config"
 	"github.com/rudderlabs/rudder-go-kit/filemanager"
 	"github.com/rudderlabs/rudder-go-kit/logger"
+	"github.com/rudderlabs/rudder-go-kit/sqlutil"
 	"github.com/rudderlabs/rudder-go-kit/stats"
 
 	"github.com/rudderlabs/rudder-server/admin"

warehouse/archive/archiver.go

@@ -9,10 +9,6 @@
 	"net/url"
 	"time"
 
-	sqlmw "github.com/rudderlabs/rudder-server/warehouse/integrations/middleware/sqlquerywrapper"
-
-	"github.com/rudderlabs/rudder-go-kit/stats"
-
 	"github.com/iancoleman/strcase"
 	"github.com/lib/pq"
 	"github.com/tidwall/gjson"
@@ -21,11 +17,13 @@
 	"github.com/rudderlabs/rudder-go-kit/config"
 	"github.com/rudderlabs/rudder-go-kit/filemanager"
 	"github.com/rudderlabs/rudder-go-kit/logger"
+	"github.com/rudderlabs/rudder-go-kit/stats"
 
 	"github.com/rudderlabs/rudder-server/services/archiver/tablearchiver"
 	"github.com/rudderlabs/rudder-server/utils/filemanagerutil"
 	"github.com/rudderlabs/rudder-server/utils/misc"
 	"github.com/rudderlabs/rudder-server/utils/timeutil"
+	sqlmw "github.com/rudderlabs/rudder-server/warehouse/integrations/middleware/sqlquerywrapper"
 	"github.com/rudderlabs/rudder-server/warehouse/internal/model"
 	"github.com/rudderlabs/rudder-server/warehouse/multitenant"
 	warehouseutils "github.com/rudderlabs/rudder-server/warehouse/utils"
@@ -61,7 +59,9 @@
 
 	config struct {
 		archiveUploadRelatedRecords config.ValueLoader[bool]
+		canDeleteUploads            config.ValueLoader[bool]
 		uploadsArchivalTimeInDays   config.ValueLoader[int]
+		uploadRetentionTimeInDays   config.ValueLoader[int]
 		archiverTickerTime          config.ValueLoader[time.Duration]
 		backupRowsBatchSize         config.ValueLoader[int]
 		maxLimit                    config.ValueLoader[int]
@@ -88,7 +88,9 @@
 	}
 
 	a.config.archiveUploadRelatedRecords = a.conf.GetReloadableBoolVar(true, "Warehouse.archiveUploadRelatedRecords")
+	a.config.canDeleteUploads = a.conf.GetReloadableBoolVar(false, "Warehouse.canDeleteUploads")
 	a.config.uploadsArchivalTimeInDays = a.conf.GetReloadableIntVar(5, 1, "Warehouse.uploadsArchivalTimeInDays")
+	a.config.uploadRetentionTimeInDays = a.conf.GetReloadableIntVar(90, 1, "Warehouse.uploadRetentionTimeInDays")
 	a.config.backupRowsBatchSize = a.conf.GetReloadableIntVar(100, 1, "Warehouse.Archiver.backupRowsBatchSize")
 	a.config.archiverTickerTime = a.conf.GetReloadableDurationVar(360, time.Minute, "Warehouse.archiverTickerTime", "Warehouse.archiverTickerTimeInMin") // default 6 hours
 	a.config.maxLimit = a.conf.GetReloadableIntVar(10000, 1, "Warehouse.Archiver.maxLimit")
@@ -510,3 +512,52 @@
 
 	return nil
 }
+
+func (a *Archiver) Delete(ctx context.Context) error {
+	a.log.Infon(`Started deleting for warehouse`)
+
+	maxLimit := a.config.maxLimit.Load()
+
+	for {
+		count, err := a.deleteUploads(ctx, maxLimit)
+		if err != nil {
+			return fmt.Errorf("deleting uploads: %w", err)
+		}
+		if count == 0 {
+			break
+		}
+	}
+	return nil
+}
+
+func (a *Archiver) deleteUploads(ctx context.Context, limit int) (int64, error) {
+	skipWorkspaceIDs := []string{""}
+	skipWorkspaceIDs = append(skipWorkspaceIDs, a.tenantManager.DegradedWorkspaces()...)
+
+	sqlStatement := fmt.Sprintf(`
+		WITH rows_to_delete AS (
+    		SELECT ctid
+    			FROM %[1]s
+    			WHERE created_at < NOW() - $1::interval
+      			AND status = $2
+      			AND NOT workspace_id = ANY ($3)
+    			LIMIT $4
+		)
+		DELETE FROM %[1]s
+		WHERE ctid IN (SELECT ctid FROM rows_to_delete);
+`,
+		pq.QuoteIdentifier(warehouseutils.WarehouseUploadsTable))
+
+	result, err := a.db.ExecContext(
+		ctx,
+		sqlStatement,
+		fmt.Sprintf("%d DAY", a.config.uploadRetentionTimeInDays.Load()),
+		model.ExportedData,
+		pq.Array(skipWorkspaceIDs),
+		limit,
+	)
+	if err != nil {
+		return 0, fmt.Errorf("error deleting uploads: %w", err)
+	}
+	return result.RowsAffected()
+}

warehouse/archive/archiver_test.go

@@ -155,7 +155,7 @@ func TestArchiver(t *testing.T) {
 			db := sqlmw.New(pgResource.DB)
 
 			archiver := archive.New(
-				config.New(),
+				c,
 				logger.NOP,
 				mockStats,
 				db,
@@ -214,6 +214,66 @@ func TestArchiver(t *testing.T) {
 	}
 }
 
+func TestArchiver_Delete(t *testing.T) {
+	var pgResource *postgres.Resource
+
+	pool, err := dockertest.NewPool("")
+	require.NoError(t, err)
+	pgResource, err = postgres.Setup(pool, t)
+	require.NoError(t, err)
+
+	t.Log("db:", pgResource.DBDsn)
+
+	err = (&migrator.Migrator{
+		Handle:          pgResource.DB,
+		MigrationsTable: "wh_schema_migrations",
+	}).Migrate("warehouse")
+	require.NoError(t, err)
+
+	sqlStatement, err := os.ReadFile("testdata/dump.sql")
+	require.NoError(t, err)
+
+	_, err = pgResource.DB.Exec(string(sqlStatement))
+	require.NoError(t, err)
+
+	ctrl := gomock.NewController(t)
+	mockStats := mock_stats.NewMockStats(ctrl)
+	mockStats.EXPECT().NewStat(gomock.Any(), gomock.Any()).Times(1)
+
+	status := model.ExportedData
+	workspaceID := "1"
+	_, err = pgResource.DB.Exec(`
+				UPDATE wh_uploads SET workspace_id = $1, status = $2
+			`, workspaceID, status)
+	require.NoError(t, err)
+
+	c := config.New()
+	c.Set("Warehouse.uploadRetentionTimeInDays", 0)
+	tenantManager := multitenant.New(c, backendConfig.DefaultBackendConfig)
+
+	db := sqlmw.New(pgResource.DB)
+
+	archiver := archive.New(
+		c,
+		logger.NOP,
+		mockStats,
+		db,
+		filemanager.New,
+		tenantManager,
+	)
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	err = archiver.Delete(ctx)
+	require.NoError(t, err)
+
+	var count int
+	err = pgResource.DB.QueryRow(fmt.Sprintf(`SELECT COUNT(*) FROM %q`, warehouseutils.WarehouseUploadsTable)).Scan(&count)
+	require.NoError(t, err)
+	require.Zero(t, count, "wh_uploads rows should be deleted")
+}
+
 func jsonTestData(t require.TestingT, file string, value any) {
 	f, err := os.Open(file)
 	require.NoError(t, err)

warehouse/archive/cron.go

@@ -3,6 +3,8 @@
 import (
 	"context"
 	"time"
+
+	obskit "github.com/rudderlabs/rudder-observability-kit/go/labels"
 )
 
 func CronArchiver(ctx context.Context, a *Archiver) {
@@ -15,7 +17,13 @@
 			if a.config.archiveUploadRelatedRecords.Load() {
 				err := a.Do(ctx)
 				if err != nil {
-					a.log.Errorf(`Error archiving uploads: %v`, err)
+					a.log.Errorn(`Error archiving uploads`, obskit.Error(err))
+				}
+			}
+			if a.config.canDeleteUploads.Load() {
+				err := a.Delete(ctx)
+				if err != nil {
+					a.log.Errorn(`Error deleting uploads`, obskit.Error(err))
 				}
 			}
 		}