@remit/search-index-worker 0.0.14 → 0.0.16

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remit/search-index-worker",
-  "version": "0.0.14",
+  "version": "0.0.16",
   "type": "module",
   "main": "dist/index.js",
   "exports": {

package/src/consumer.ts

@@ -50,10 +50,9 @@ export interface SqsConsumerConfig {
  * via `processBatch` (a batch of one) — the same per-message logic the AWS
  * Lambda handler runs, reused so the two deployment shapes (Lambda event
  * source mapping vs. a long-running container/pm2 process) share one
- * indexing implementation. Used by the Postgres-parity stack, where the
- * search-index queue has no Lambda event source; `remit-pg-index-worker`
- * only relays committed outbox events onto this queue (the producer side —
- * see its `worker.ts`), it does not consume them.
+ * indexing implementation. Used by the self-host stack, where the search-index
+ * queue has no Lambda event source: the outbox drain relays committed events
+ * onto this queue (the producer side) and this consumer takes them off it.
  *
  * A message is deleted only when `processBatch` reports no failure for it;
  * a failure leaves it on the queue so its visibility timeout lapses and SQS

package/src/data-ports.ts

@@ -17,11 +17,10 @@ export interface SearchIndexDataPorts {
 	 * one — this hook is `undefined` there, and the handler uses the message's
 	 * own `accountId` unchanged (see `prepareUpsert` in handler.ts).
 	 *
-	 * The Postgres outbox trigger fires from a plain `message_id`, so the pg
-	 * relay (`remit-pg-index-worker`) has no accountId to attach; this hook
-	 * derives it from the message's mailbox at consume time instead. Returns
-	 * null when the mailbox can't be resolved (the message is skipped, not
-	 * retried — see handler.ts).
+	 * The outbox carries a plain `message_id`, so the relational drain has no
+	 * accountId to attach; this hook derives it from the message's mailbox at
+	 * consume time instead. Returns null when the mailbox can't be resolved (the
+	 * message is skipped, not retried — see handler.ts).
 	 */
 	resolveAccountId?: (messageId: string) => Promise<string | null>;
 }

package/src/handler.ts

@@ -86,9 +86,8 @@ const deleteMessage = async (
 // upsert per message means a record S3 Vectors rejects (e.g. a metadata
 // ValidationException) dead-letters on its own — its siblings in the
 // batch still index instead of retrying forever behind a poison record.
-// Exported so the long-running Postgres consumer (`consumer.ts`) and the
-// bulk reindex script (`reindex.ts`) can process one message at a time
-// outside the Lambda batch shape, reusing this exact logic.
+// Exported so the long-running consumer (`consumer.ts`) can process one
+// message at a time outside the Lambda batch shape, reusing this exact logic.
 export const upsertMessage = async (
 	message: Extract<ParsedQueueMessage, { kind: "upsert" }>,
 	services: Services,

package/src/services.ts

@@ -73,8 +73,6 @@ export const getServices = async (): Promise<Services> => {
 	return cached;
 };
 
-export const createServices = (overrides: Services): Services => overrides;
-
 /** Reset the singleton — test use only. */
 export const _resetForTest = (): void => {
 	cached = undefined;

package/src/reindex.ts

@@ -1,63 +0,0 @@
-import type { Logger } from "@remit/logger-lambda";
-import pMap from "p-map";
-import type { Pool } from "pg";
-import { type IndexOutcome, upsertMessage } from "./handler.js";
-import type { Services } from "./services.js";
-
-export interface ReindexResult {
-	total: number;
-	indexed: number;
-	skipped: number;
-}
-
-const NOT_APPLICABLE_ACCOUNT_ID = "reindex-all";
-
-/**
- * Re-embed every body-synced Postgres message. Concurrent (pMap) and
- * force-upserting so a model change or a repair repopulates the whole store;
- * keys-only scan keeps the hot path off `describe()`. Postgres-only — the
- * `accountId` on the synthetic upsert message is a placeholder: `services`
- * must carry `resolveAccountId` (true whenever `DATA_BACKEND=postgres`; see
- * `data-ports.ts`), which derives the real one from each message's mailbox.
- */
-export const reindexAll = async (
-	pool: Pool,
-	services: Services,
-	log: Logger,
-	options?: { concurrency?: number },
-): Promise<ReindexResult> => {
-	const rows = await pool.query<{ message_id: string }>(
-		"SELECT message_id FROM message WHERE body_storage_key IS NOT NULL",
-	);
-	const messageIds = rows.rows.map((row) => row.message_id);
-
-	let indexed = 0;
-	let skipped = 0;
-	await pMap(
-		messageIds,
-		async (messageId) => {
-			let outcome: IndexOutcome | undefined;
-			const taskServices: Services = {
-				...services,
-				onIndexOutcome: (o) => {
-					outcome = o;
-				},
-			};
-			await upsertMessage(
-				{
-					kind: "upsert",
-					accountId: NOT_APPLICABLE_ACCOUNT_ID,
-					messageId,
-					force: true,
-				},
-				taskServices,
-				log,
-			);
-			if (outcome?.status === "indexed") indexed += 1;
-			else skipped += 1;
-		},
-		{ concurrency: options?.concurrency ?? 8 },
-	);
-
-	return { total: messageIds.length, indexed, skipped };
-};

package/src/run-reindex.ts

@@ -1,22 +0,0 @@
-import { createLogger } from "@remit/logger-lambda";
-import pg from "pg";
-import { reindexAll } from "./reindex.js";
-import { getServices } from "./services.js";
-
-const log = createLogger();
-
-const main = async (): Promise<void> => {
-	const connectionString = process.env.PG_CONNECTION_URL;
-	if (!connectionString) throw new Error("PG_CONNECTION_URL is required");
-
-	const pool = new pg.Pool({ connectionString });
-	const services = await getServices();
-
-	const result = await reindexAll(pool, services, log);
-	log.info("reindex complete", { ...result });
-
-	await pool.end();
-	process.exit(0);
-};
-
-await main();