Cleanup orphaned using ids, not timestamp sorting

Author: TheodoreSpeaks

apps/sim/background/cleanup-soft-deletes.ts

@@ -15,13 +15,12 @@ import {
 import { createLogger } from '@sim/logger'
 import { task } from '@trigger.dev/sdk'
 import { and, inArray, isNotNull, lt } from 'drizzle-orm'
-import type { PgColumn, PgTable } from 'drizzle-orm/pg-core'
 import { type CleanupJobPayload, resolveCleanupScope } from '@/lib/billing/cleanup-dispatcher'
 import {
   batchDeleteByWorkspaceAndTimestamp,
   DEFAULT_BATCH_SIZE,
   DEFAULT_MAX_BATCHES_PER_TABLE,
-  type TableCleanupResult,
+  deleteRowsById,
 } from '@/lib/cleanup/batch-delete'
 import { prepareChatCleanup } from '@/lib/cleanup/chat-cleanup'
 import type { StorageContext } from '@/lib/uploads'
@@ -112,25 +111,6 @@ async function cleanupWorkspaceFileStorage(
   return stats
 }
 
-async function deleteRowsById(
-  tableDef: PgTable,
-  idCol: PgColumn,
-  ids: string[],
-  tableName: string
-): Promise<TableCleanupResult> {
-  const result: TableCleanupResult = { table: tableName, deleted: 0, failed: 0 }
-  if (ids.length === 0) return result
-  try {
-    const deleted = await db.delete(tableDef).where(inArray(idCol, ids)).returning({ id: idCol })
-    result.deleted = deleted.length
-    logger.info(`[${tableName}] Deleted ${deleted.length} rows`)
-  } catch (error) {
-    result.failed++
-    logger.error(`[${tableName}] Delete failed:`, { error })
-  }
-  return result
-}
-
 /**
  * Tables cleaned by the generic workspace-scoped batched DELETE. Tables whose
  * hard-delete triggers external side effects (workflow → copilot chats cascade,

apps/sim/background/cleanup-tasks.ts

@@ -15,6 +15,7 @@ import {
   batchDeleteByWorkspaceAndTimestamp,
   DEFAULT_BATCH_SIZE,
   DEFAULT_MAX_BATCHES_PER_TABLE,
+  deleteRowsById,
   type TableCleanupResult,
 } from '@/lib/cleanup/batch-delete'
 import { prepareChatCleanup } from '@/lib/cleanup/chat-cleanup'
@@ -161,15 +162,14 @@ export async function runCleanupTasks(payload: CleanupJobPayload): Promise<void>
     tableName: `${label}/copilotRuns`,
   })
 
-  // Delete copilot chats (has workspaceId directly)
-  const chatsResult = await batchDeleteByWorkspaceAndTimestamp({
-    tableDef: copilotChats,
-    workspaceIdCol: copilotChats.workspaceId,
-    timestampCol: copilotChats.updatedAt,
-    workspaceIds,
-    retentionDate,
-    tableName: `${label}/copilotChats`,
-  })
+  // Delete copilot chats using the exact IDs collected above so the chat
+  // cleanup (S3 + copilot backend) and the DB delete can never disagree.
+  const chatsResult = await deleteRowsById(
+    copilotChats,
+    copilotChats.id,
+    doomedChatIds,
+    `${label}/copilotChats`
+  )
 
   // Delete mothership inbox tasks (has workspaceId directly)
   const inboxResult = await batchDeleteByWorkspaceAndTimestamp({

apps/sim/lib/cleanup/batch-delete.ts

@@ -92,3 +92,28 @@ export async function batchDeleteByWorkspaceAndTimestamp({
 
   return result
 }
+
+/**
+ * Delete rows by an explicit list of IDs. Use this when the IDs were selected
+ * upstream (e.g., to drive external cleanup like S3 deletes or a backend API
+ * call) so the DB delete cannot drift from the upstream selection. Paired with
+ * `batchDeleteByWorkspaceAndTimestamp` for tables with no external side effects.
+ */
+export async function deleteRowsById(
+  tableDef: PgTable,
+  idCol: PgColumn,
+  ids: string[],
+  tableName: string
+): Promise<TableCleanupResult> {
+  const result: TableCleanupResult = { table: tableName, deleted: 0, failed: 0 }
+  if (ids.length === 0) return result
+  try {
+    const deleted = await db.delete(tableDef).where(inArray(idCol, ids)).returning({ id: idCol })
+    result.deleted = deleted.length
+    logger.info(`[${tableName}] Deleted ${deleted.length} rows`)
+  } catch (error) {
+    result.failed++
+    logger.error(`[${tableName}] Delete failed:`, { error })
+  }
+  return result
+}