Free, Open Source, Self-Hosted WhatsApp API Gateway
TypeScript apibotgatewayself-hostedwhatsappwhatsapp-apiwhatsapp-automationwhatsapp-botwhatsapp-gatewaywhatsapp-gateway-apiwhatsapp-webwhatsapp-web-api
docs(migrations): correct the lock and timeout notes on index builds
Yudhi Armyndharis36e9d4b40557
diff --git a/src/database/migrations/1785123853000-AddMessagesCreatedAtIndex.ts b/src/database/migrations/1785123853000-AddMessagesCreatedAtIndex.tsindex 9643228fa..902c4595f 100644--- a/src/database/migrations/1785123853000-AddMessagesCreatedAtIndex.ts+++ b/src/database/migrations/1785123853000-AddMessagesCreatedAtIndex.ts@@ -12,13 +12,14 @@ import { MigrationInterface, QueryRunner } from 'typeorm';* `synchronize` already created the same-named index declared on the Message entity.** The `messages` table is the hottest data table (every inbound/outbound row), so on Postgres a- * CREATE INDEX over a large table at boot can exceed the runtime pool's statement_timeout and abort- * the migration. Lift it for THIS transaction (SET LOCAL auto-reverts at COMMIT; SQLite rejects it- * syntactically, hence the guard) — the same discipline AddWebhooksSessionIdIndex and- * AddMessagesWaMessageIdUnique already use. `CONCURRENTLY` would avoid the ACCESS EXCLUSIVE lock- * but cannot run inside a transaction, so it is deliberately NOT used here (the boot-time blocking- * window is acceptable for a self-hosted gateway; consistency with the sibling index migrations- * outweighs it).+ * CREATE INDEX over a large table can run long. Boot and CLI migrations run on a pool without the+ * runtime statement_timeout (pg-boot-migrations.ts); lifting it for THIS transaction only guards+ * against a role- or database-level default cancelling the build (SET LOCAL auto-reverts at COMMIT;+ * SQLite rejects it syntactically, hence the guard), the same discipline AddWebhooksSessionIdIndex+ * and AddMessagesWaMessageIdUnique already use. `CONCURRENTLY` would avoid the SHARE lock (writes to+ * messages wait for the build, reads do not) but cannot run inside a transaction, so it is+ * deliberately NOT used here (the boot-time blocking window is acceptable for a self-hosted gateway;+ * consistency with the sibling index migrations outweighs it).*/export class AddMessagesCreatedAtIndex1785123853000 implements MigrationInterface {name = 'AddMessagesCreatedAtIndex1785123853000';diff --git a/src/database/migrations/1786700000000-AddMessagesSessionChatCreatedAtIndex.ts b/src/database/migrations/1786700000000-AddMessagesSessionChatCreatedAtIndex.tsindex 23af2dd74..c59cad9f4 100644--- a/src/database/migrations/1786700000000-AddMessagesSessionChatCreatedAtIndex.ts+++ b/src/database/migrations/1786700000000-AddMessagesSessionChatCreatedAtIndex.ts@@ -19,8 +19,9 @@ export class AddMessagesSessionChatCreatedAtIndex1786700000000 implements Migratname = 'AddMessagesSessionChatCreatedAtIndex1786700000000';public async up(queryRunner: QueryRunner): Promise<void> {- // The data pool boots with a runtime statement_timeout (default 30s), which a build over a large- // messages table can outlast. Lift it for this transaction only, like AddMessageMediaPathIndex.+ // Boot and CLI migrations run on a pool without the runtime statement_timeout (pg-boot-migrations.ts);+ // this guards against a role- or database-level default cancelling a long build. Lifted for this+ // transaction only, like AddMessageMediaPathIndex.if (queryRunner.dataSource.options.type === 'postgres') {await queryRunner.query('SET LOCAL statement_timeout = 0');}