@powersync/service-module-mongodb-storage 0.23.2 → 0.24.0

package/test/src/replication_uploads.test.ts

@@ -0,0 +1,393 @@
+import { mongoTestStorageFactoryGenerator } from '@module/utils/test-utils.js';
+import { mongo } from '@powersync/lib-service-mongodb';
+import { logger } from '@powersync/lib-services-framework';
+import { storage, updateSyncRulesFromYaml } from '@powersync/service-core';
+import { getTestStorage, test_utils } from '@powersync/service-core-tests';
+import * as bson from 'bson';
+import { describe, expect, test, vi } from 'vitest';
+import { MongoWriteBatch } from '../../src/storage/implementation/MongoWriteBatch.js';
+import { MongoBucketBatchV3 } from '../../src/storage/implementation/v3/MongoBucketBatchV3.js';
+import { MongoSyncBucketStorageV3 } from '../../src/storage/implementation/v3/MongoSyncBucketStorageV3.js';
+import {
+  OBJECT_STORAGE_PUBLICATION_SAFETY_MARGIN_MS,
+  ObjectStorageLifecycle
+} from '../../src/storage/implementation/v3/object-storage/ObjectStorageLifecycle.js';
+import { ObjectStorageUsage } from '../../src/storage/implementation/v3/object-storage/ObjectStorageUsage.js';
+import { PersistedBatchV3 } from '../../src/storage/implementation/v3/PersistedBatchV3.js';
+import { env } from './env.js';
+import { MemoryObjectStorage } from './helpers/MemoryObjectStorage.js';
+import { PublicationCommandMonitor } from './helpers/PublicationCommandMonitor.js';
+
+const CONFIG = `
+config:
+  edition: 3
+streams:
+  items:
+    query: SELECT * FROM items
+`;
+
+async function open(version: number, options: Partial<storage.CreateWriterOptions> = {}) {
+  const objectStorage = new MemoryObjectStorage();
+  const factoryGen = mongoTestStorageFactoryGenerator({
+    url: env.MONGO_TEST_URL,
+    isCI: env.CI,
+    monitorCommands: true,
+    objectStorage,
+    inlineThresholdBytes: 0
+  });
+  const factory = await factoryGen.factory();
+  const stream = await factory.updateSyncRules(updateSyncRulesFromYaml(CONFIG, { storageVersion: version }));
+  const bucketStorage = (await getTestStorage(factory, stream)) as MongoSyncBucketStorageV3;
+  const writer = (await bucketStorage.createWriter({ ...test_utils.BATCH_OPTIONS, ...options })) as MongoBucketBatchV3;
+  const table = await test_utils.resolveTestTable(writer, 'items', ['id'], factoryGen, 1);
+  await writer.markAllSnapshotDone('1/1');
+  const db = bucketStorage.db;
+  const markers = db.pendingObjectStorageDeletes(stream.replicationStreamId);
+  // The generic test factory clears stream data but retains old outbox
+  // collections; reset this test's outbox as well when stream IDs are reused.
+  await markers.deleteMany({});
+  const lifecycle = new ObjectStorageLifecycle(db, stream.replicationStreamId, objectStorage);
+  const data = db.bucketData(
+    stream.replicationStreamId,
+    stream.syncConfigContent[0].mapping.allBucketDefinitionIds()[0]
+  );
+  const records = db.sourceRecords(stream.replicationStreamId, table.id as bson.ObjectId);
+  await writer.save({
+    sourceTable: table,
+    tag: storage.SaveOperationTag.INSERT,
+    after: { id: 'A', description: 'offloaded' },
+    afterReplicaId: test_utils.rid('A')
+  });
+  return { factory, stream, bucketStorage, writer, table, db, objectStorage, markers, lifecycle, data, records };
+}
+
+describe.each([3, 4])('replication uploads v%s', (version) => {
+  test('registers once with majority acknowledgement before PUT and reuses uploads on transaction retry', async () => {
+    const context = await open(version);
+    await using factory = context.factory;
+    await using writer = context.writer;
+    const { markers, objectStorage, db } = context;
+    using monitor = new PublicationCommandMonitor(factory.db.client);
+    using registered = vi.spyOn(ObjectStorageLifecycle.prototype, 'prepareUploads');
+    const put = objectStorage.put.bind(objectStorage);
+    using uploading = vi.spyOn(objectStorage, 'put').mockImplementation(async (...args) => {
+      expect(monitor.activeReplicationAttempts).toHaveLength(0);
+      expect(await markers.findOne({ path: args[0] })).toMatchObject({ path: args[0], delete_after: expect.any(Date) });
+      await put(...args);
+    });
+    const registrations: { majority: boolean; transactional: boolean }[] = [];
+    const command = (event: mongo.CommandStartedEvent) => {
+      if (event.commandName === 'insert' && event.command.insert === markers.collectionName) {
+        registrations.push({
+          majority: event.command.writeConcern?.w === 'majority',
+          transactional: event.command.autocommit === false
+        });
+      }
+    };
+    db.client.on('commandStarted', command);
+    using _listener = { [Symbol.dispose]: () => db.client.off('commandStarted', command) };
+    const publish = PersistedBatchV3.prototype.publish;
+    using publishing = vi.spyOn(PersistedBatchV3.prototype, 'publish').mockImplementationOnce(async function (
+      this: PersistedBatchV3,
+      ...args
+    ) {
+      await publish.apply(this, args);
+      throw new mongo.MongoServerError({
+        code: 112,
+        message: 'publication conflict',
+        errorLabels: ['TransientTransactionError']
+      });
+    });
+    await writer.commit('1/2');
+    expect(publishing).toHaveBeenCalledTimes(2);
+    expect(uploading).toHaveBeenCalledTimes(1);
+    expect(registered).toHaveBeenCalledTimes(1);
+    expect(registrations).toEqual([{ majority: true, transactional: false }]);
+    expect(await markers.countDocuments()).toBe(0);
+    expect(await context.data.countDocuments()).toBe(1);
+    const [usage] = await ObjectStorageUsage.readAllDefinitionUsage(db);
+    expect(usage.active_bytes).toBe(BigInt([...objectStorage.store.values()][0].data.byteLength));
+    // The first commit also activates the config in its own fenced transaction.
+    expect(monitor.replicationAttempts.map((attempt) => attempt.outcome)).toEqual([
+      'aborted',
+      'committed',
+      'committed'
+    ]);
+  });
+
+  test('does not PUT after ambiguous marker registration', async () => {
+    const context = await open(version);
+    await using _factory = context.factory;
+    await using writer = context.writer;
+    const prepare = ObjectStorageLifecycle.prototype.prepareUploads;
+    const failure = new Error('lost marker registration response');
+    using _registration = vi
+      .spyOn(ObjectStorageLifecycle.prototype, 'prepareUploads')
+      .mockImplementationOnce(async function (this: ObjectStorageLifecycle, ...args) {
+        await prepare.apply(this, args);
+        throw failure;
+      });
+    using put = vi.spyOn(context.objectStorage, 'put');
+    await expect(writer.commit('1/2')).rejects.toThrow(failure);
+    expect(put).not.toHaveBeenCalled();
+    expect(await context.markers.countDocuments()).toBe(1);
+    expect(await context.records.countDocuments()).toBe(0);
+    expect((await context.bucketStorage.getCheckpoint()).checkpoint).toBe(0n);
+    // A later explicit retry reevaluates the retained input using a fresh path.
+    await writer.commit('1/2');
+    expect(put).toHaveBeenCalledTimes(1);
+    expect(await context.markers.countDocuments()).toBe(1);
+    expect(await context.records.countDocuments()).toBe(1);
+  });
+
+  test('lost PUT responses retain orphan markers, and cleanup retries deletion failures', async () => {
+    const context = await open(version);
+    await using _factory = context.factory;
+    await using writer = context.writer;
+    const put = context.objectStorage.put.bind(context.objectStorage);
+    using _put = vi.spyOn(context.objectStorage, 'put').mockImplementationOnce(async (...args) => {
+      await put(...args);
+      throw new Error('lost PUT response');
+    });
+    await expect(writer.commit('1/2')).rejects.toThrow('lost PUT response');
+    expect(context.objectStorage.store.size).toBe(1);
+    expect(await context.markers.countDocuments()).toBe(1);
+    expect(await context.data.countDocuments()).toBe(0);
+    expect(await ObjectStorageUsage.readAllDefinitionUsage(context.db)).toEqual([]);
+    await context.lifecycle.cleanup(logger);
+    expect(context.objectStorage.store.size).toBe(1);
+    await context.markers.updateMany({}, { $set: { delete_after: new Date(0) } });
+    using deleting = vi.spyOn(context.objectStorage, 'delete').mockRejectedValueOnce(new Error('DELETE failed'));
+    await context.lifecycle.cleanup(logger);
+    expect(await context.markers.countDocuments()).toBe(1);
+    expect(context.objectStorage.store.size).toBe(1);
+    await context.lifecycle.cleanup(logger);
+    expect(deleting).toHaveBeenCalledTimes(2);
+    expect(await context.markers.countDocuments()).toBe(0);
+    expect(context.objectStorage.store.size).toBe(0);
+  });
+
+  test('rejects an uploaded generation whose assigned ID does not advance the persisted head', async () => {
+    const context = await open(version);
+    await using _factory = context.factory;
+    await using writer = context.writer;
+    const put = context.objectStorage.put.bind(context.objectStorage);
+    using uploads = vi.spyOn(context.objectStorage, 'put').mockImplementationOnce(async (...args) => {
+      await put(...args);
+      await context.db.sync_rules.updateOne(
+        { _id: context.stream.replicationStreamId },
+        { $set: { last_persisted_op: 10n } },
+        { writeConcern: { w: 'majority' } }
+      );
+    });
+    await expect(writer.commit('1/2')).rejects.toThrow('does not advance persisted stream head');
+    expect(uploads).toHaveBeenCalledTimes(1);
+    const [abandoned] = uploads.mock.calls.map(([path]) => path);
+    expect(await context.markers.find({}).toArray()).toMatchObject([{ path: abandoned }]);
+    expect(await context.data.find({}).toArray()).toEqual([]);
+    expect(writer.last_flushed_op).toBeNull();
+    expect((await context.bucketStorage.getCheckpoint()).checkpoint).toBe(0n);
+    expect(await context.db.sync_rules.findOne({ _id: context.stream.replicationStreamId })).toMatchObject({
+      last_persisted_op: 10n
+    });
+    await context.markers.updateMany({}, { $set: { delete_after: new Date(0) } });
+    await context.lifecycle.cleanup(logger);
+    expect(context.objectStorage.store.size).toBe(0);
+    expect(await context.markers.countDocuments()).toBe(0);
+    expect(await ObjectStorageUsage.readAllDefinitionUsage(context.db)).toEqual([]);
+  });
+
+  test('stopped-stream termination clears abandoned replication objects and their markers', async () => {
+    const context = await open(version);
+    await using factory = context.factory;
+    await using writer = context.writer;
+    const put = context.objectStorage.put.bind(context.objectStorage);
+    using _put = vi.spyOn(context.objectStorage, 'put').mockImplementationOnce(async (...args) => {
+      await put(...args);
+      throw new Error('lost PUT response');
+    });
+    await expect(writer.flush()).rejects.toThrow('lost PUT response');
+    expect(context.objectStorage.store.size).toBe(1);
+    await factory.db.sync_rules.updateOne(
+      { _id: context.stream.replicationStreamId },
+      { $set: { state: storage.SyncRuleState.STOP, 'sync_configs.$[].state': storage.SyncRuleState.STOP } }
+    );
+    // Supported termination runs after the replication job has stopped/drained
+    // and released its lease. It removes the entire prefix, including orphans.
+    await test_utils.releaseTestStorageLease(factory, context.stream.replicationStreamId);
+    await context.bucketStorage.terminate();
+    expect(context.objectStorage.store.size).toBe(0);
+    expect(await context.markers.countDocuments()).toBe(0);
+    expect(await factory.db.sync_rules.findOne({ _id: context.stream.replicationStreamId })).toMatchObject({
+      state: storage.SyncRuleState.TERMINATED
+    });
+  });
+
+  test.each([false, true])(
+    'missing markers roll back references, usage and progress and stop the lease (fallback: %s)',
+    async (fallback) => {
+      const context = await open(version);
+      await using factory = context.factory;
+      await using writer = context.writer;
+      const createWriteBatch = factory.db.createWriteBatch.bind(factory.db);
+      using _bulk = vi
+        .spyOn(factory.db, 'createWriteBatch')
+        .mockImplementation((session, options) =>
+          fallback
+            ? new MongoWriteBatch(factory.db.client, false, session, options)
+            : createWriteBatch(session, options)
+        );
+      const put = context.objectStorage.put.bind(context.objectStorage);
+      using _put = vi.spyOn(context.objectStorage, 'put').mockImplementationOnce(async (...args) => {
+        await put(...args);
+        await context.markers.deleteMany({});
+      });
+      await expect(writer.commit('1/2')).rejects.toThrow('Missing object storage publication markers');
+      expect(await context.data.countDocuments()).toBe(0);
+      expect(await context.records.countDocuments()).toBe(0);
+      expect(await ObjectStorageUsage.readAllDefinitionUsage(context.db)).toEqual([]);
+      expect(writer.last_flushed_op).toBeNull();
+      expect((await context.bucketStorage.getCheckpoint()).checkpoint).toBe(0n);
+      await expect(writer.commit('1/2')).rejects.toThrow('restart replication from persisted progress');
+      expect(await context.data.countDocuments()).toBe(0);
+    }
+  );
+
+  test('rechecks expiry on a retry and leaves the generation for cleanup', async () => {
+    const context = await open(version);
+    await using _factory = context.factory;
+    await using writer = context.writer;
+    using registered = vi.spyOn(ObjectStorageLifecycle.prototype, 'prepareUploads');
+    using put = vi.spyOn(context.objectStorage, 'put');
+    const publish = PersistedBatchV3.prototype.publish;
+    let attempts = 0;
+    using _publish = vi.spyOn(PersistedBatchV3.prototype, 'publish').mockImplementation(async function (
+      this: PersistedBatchV3,
+      ...args
+    ) {
+      attempts++;
+      if (attempts === 2) {
+        const uploads = await registered.mock.results[0].value;
+        uploads[0].deleteAfter = new Date(Date.now() + OBJECT_STORAGE_PUBLICATION_SAFETY_MARGIN_MS);
+      }
+      await publish.apply(this, args);
+      throw new mongo.MongoServerError({
+        code: 112,
+        message: 'publication conflict',
+        errorLabels: ['TransientTransactionError']
+      });
+    });
+    await expect(writer.commit('1/2')).rejects.toThrow('publication lease expired');
+    expect(attempts).toBe(2);
+    expect(put).toHaveBeenCalledTimes(1);
+    expect(registered).toHaveBeenCalledTimes(1);
+    expect(await context.markers.countDocuments()).toBe(1);
+    expect(await context.data.countDocuments()).toBe(0);
+    expect(await ObjectStorageUsage.readAllDefinitionUsage(context.db)).toEqual([]);
+  });
+
+  test.each([false, true])(
+    'terminal commit uncertainty preserves the correct marker state (committed: %s)',
+    async (committed) => {
+      const context = await open(version);
+      await using _factory = context.factory;
+      await using writer = context.writer;
+      await using other = await context.bucketStorage.createWriter(test_utils.BATCH_OPTIONS);
+      const failure = new mongo.MongoServerError({
+        message: 'commit outcome unknown',
+        errorLabels: ['UnknownTransactionCommitResult']
+      });
+      const session = writer.session;
+      const transact = session.withTransaction.bind(session);
+      using _transaction = vi.spyOn(session, 'withTransaction').mockImplementationOnce(async (...args) => {
+        if (committed) {
+          await transact(...args);
+          throw failure;
+        }
+        return transact(async (...callbackArgs) => {
+          await args[0](...callbackArgs);
+          throw failure;
+        }, args[1]);
+      });
+      await expect(writer.commit('1/2')).rejects.toThrow(failure);
+      expect(await context.markers.countDocuments()).toBe(committed ? 0 : 1);
+      expect(await context.data.countDocuments()).toBe(committed ? 1 : 0);
+      expect(writer.last_flushed_op).toBeNull();
+      await expect(other.setResumeLsn('1/3')).rejects.toThrow('restart replication');
+      await context.markers.updateMany({}, { $set: { delete_after: new Date(0) } });
+      await context.lifecycle.cleanup(logger);
+      expect(context.objectStorage.store.size).toBe(committed ? 1 : 0);
+    }
+  );
+
+  test('truncate PUTs complete before its transaction and publish their markers atomically', async () => {
+    const context = await open(version);
+    await using factory = context.factory;
+    await using writer = context.writer;
+    await writer.commit('1/2');
+    using monitor = new PublicationCommandMonitor(factory.db.client);
+    const put = context.objectStorage.put.bind(context.objectStorage);
+    using uploading = vi.spyOn(context.objectStorage, 'put').mockImplementation(async (...args) => {
+      expect(monitor.activeReplicationAttempts).toHaveLength(0);
+      expect(await context.markers.countDocuments({ path: args[0] })).toBe(1);
+      await put(...args);
+    });
+    await writer.truncate([context.table]);
+    expect(uploading).toHaveBeenCalledTimes(1);
+    expect(await context.markers.countDocuments()).toBe(0);
+    expect(await context.records.countDocuments()).toBe(0);
+    expect(await context.data.countDocuments()).toBe(2);
+  });
+
+  test('cancels queued PUTs and drains active requests before returning', async () => {
+    const controller = new AbortController();
+    const context = await open(version, { signal: controller.signal });
+    await using _factory = context.factory;
+    await using writer = context.writer;
+    // Enough data for separate chunks, within a single bounded publication.
+    await writer.save({
+      sourceTable: context.table,
+      tag: storage.SaveOperationTag.INSERT,
+      after: { id: 'B', description: 'x'.repeat(1_100_000) },
+      afterReplicaId: test_utils.rid('B')
+    });
+    const active = Promise.withResolvers<void>();
+    const release = Promise.withResolvers<void>();
+    const queued = Promise.withResolvers<void>();
+    let remoteCompletions = 0;
+    using put = vi
+      .spyOn(context.objectStorage, 'put')
+      .mockImplementationOnce(async (path, data, metadata) => {
+        active.resolve();
+        await release.promise;
+        // An accepted request can complete after local cancellation. Its marker
+        // remains protected for an hour; cancellation must not delete it early.
+        context.objectStorage.store.set(path, { data, metadata });
+        remoteCompletions++;
+      })
+      .mockImplementationOnce(async (_path, _data, _metadata, options) => {
+        queued.resolve();
+        await active.promise;
+        if (!options?.signal?.aborted) {
+          await new Promise<void>((resolve) =>
+            options!.signal!.addEventListener('abort', () => resolve(), { once: true })
+          );
+        }
+        options?.signal?.throwIfAborted();
+      });
+    const failed = expect(writer.commit('1/2')).rejects.toThrow('cancel uploads');
+    await queued.promise;
+    controller.abort(new Error('cancel uploads'));
+    expect(remoteCompletions).toBe(0);
+    release.resolve();
+    await failed;
+    expect(remoteCompletions).toBe(1);
+    expect(put).toHaveBeenCalledTimes(2);
+    expect(await context.markers.countDocuments()).toBe(2);
+    await context.markers.updateMany({}, { $set: { delete_after: new Date(0) } });
+    await context.lifecycle.cleanup(logger);
+    expect(context.objectStorage.store.size).toBe(0);
+    expect(await context.markers.countDocuments()).toBe(0);
+    expect(await context.records.countDocuments()).toBe(0);
+  });
+});

package/test/src/source_table_config.test.ts

@@ -1,7 +1,7 @@
 import { mongoTestStorageFactoryGenerator } from '@module/utils/test-utils.js';
 import { updateSyncRulesFromConfig } from '@powersync/service-core';
 import { test_utils } from '@powersync/service-core-tests';
-import { PrecompiledSyncConfig, SqlSyncRules } from '@powersync/service-sync-rules';
+import { PrecompiledSyncConfig, parseSyncConfig } from '@powersync/service-sync-rules';
 import { describe, expect, test } from 'vitest';
 import { env } from './env.js';
 import { TEST_STORAGE_VERSIONS } from './util.js';
@@ -39,7 +39,7 @@ describe.for(TEST_STORAGE_VERSIONS)('source-table config with storage v%i', (sto
     let expected;
     {
       await using factory = await storageFactory().factory();
-      const parsed = SqlSyncRules.fromYaml(CONFIGURED, test_utils.PARSE_OPTIONS);
+      const parsed = parseSyncConfig(CONFIGURED, test_utils.PARSE_OPTIONS);
       expected = (parsed.config as PrecompiledSyncConfig).plan.sourceTableConfig;
       await factory.updateSyncRules(updateSyncRulesFromConfig(parsed, { storageVersion }));
     }
@@ -56,7 +56,7 @@ describe.for(TEST_STORAGE_VERSIONS)('source-table config with storage v%i', (sto
   test('continues loading unfiltered plans without source options', async () => {
     await using factory = await storageFactory().factory();
     await factory.updateSyncRules(
-      updateSyncRulesFromConfig(SqlSyncRules.fromYaml(STREAMS, test_utils.PARSE_OPTIONS), {
+      updateSyncRulesFromConfig(parseSyncConfig(STREAMS, test_utils.PARSE_OPTIONS), {
         storageVersion
       })
     );
@@ -69,7 +69,7 @@ describe.for(TEST_STORAGE_VERSIONS)('source-table config with storage v%i', (sto
     await using factory = await storageFactory().factory();
     const deploy = (yaml: string) =>
       factory.updateSyncRules(
-        updateSyncRulesFromConfig(SqlSyncRules.fromYaml(yaml, test_utils.PARSE_OPTIONS), {
+        updateSyncRulesFromConfig(parseSyncConfig(yaml, test_utils.PARSE_OPTIONS), {
           storageVersion
         })
       );

package/test/src/storage_object_storage_inline_threshold.test.ts

@@ -2,8 +2,8 @@ import { mongoTestStorageFactoryGenerator } from '@module/utils/test-utils.js';
 import { storage, updateSyncRulesFromYaml } from '@powersync/service-core';
 import { bucketRequest, getTestStorage, test_utils } from '@powersync/service-core-tests';
 import { describe, expect, test } from 'vitest';
-import { DEFAULT_INLINE_THRESHOLD_BYTES } from '../../src/storage/implementation/common/PersistedBatch.js';
 import { MongoSyncBucketStorage } from '../../src/storage/implementation/createMongoSyncBucketStorage.js';
+import { DEFAULT_INLINE_THRESHOLD_BYTES } from '../../src/storage/implementation/v3/PersistedBatchV3.js';
 import { VersionedPowerSyncMongoV3 } from '../../src/storage/implementation/v3/VersionedPowerSyncMongoV3.js';
 import { env } from './env.js';
 import { MemoryObjectStorage } from './helpers/MemoryObjectStorage.js';

package/test/src/storage_s3_compaction_lifecycle.test.ts

@@ -1,7 +1,7 @@
 import * as lib_mongo from '@powersync/lib-service-mongodb';
 import { storage, updateSyncRulesFromYaml } from '@powersync/service-core';
 import { bucketRequest, compactActive, getTestStorage, test_utils } from '@powersync/service-core-tests';
-import { describe, expect, test } from 'vitest';
+import { describe, expect, test, vi } from 'vitest';
 import { MongoSyncBucketStorage } from '../../src/storage/implementation/createMongoSyncBucketStorage.js';
 import { MongoSyncBucketStorageV3 } from '../../src/storage/implementation/v3/MongoSyncBucketStorageV3.js';
 import { VersionedPowerSyncMongoV3 } from '../../src/storage/implementation/v3/VersionedPowerSyncMongoV3.js';
@@ -53,6 +53,98 @@ async function expectUsageMatchesBucketData(bucketStorage: MongoSyncBucketStorag
 }
 
 describe('S3 compaction storage lifecycle', () => {
+  test.each(
+    [false, true].flatMap((compactChunksOnly) =>
+      [
+        ...[-1, 0, 1].map((thresholdOffset) => ({ thresholdOffset, initiallyInline: true })),
+        { thresholdOffset: -1, initiallyInline: false }
+      ].map((placement) => ({ compactChunksOnly, ...placement }))
+    )
+  )(
+    'unchanged chunk placement: V4, chunks only=$compactChunksOnly, inline=$initiallyInline, threshold offset=$thresholdOffset',
+    async ({ compactChunksOnly, thresholdOffset, initiallyInline }) => {
+      const { memoryStorage, factory: factoryGen } = memoryS3Factory({
+        inlineThresholdBytes: initiallyInline ? 10_000 : 0
+      });
+      await using factory = await factoryGen.factory();
+      const syncRules = await factory.updateSyncRules(
+        updateSyncRulesFromYaml(
+          `
+config:
+  edition: 3
+streams:
+  global:
+    query: SELECT id, description FROM items
+`,
+          { storageVersion: 4 }
+        )
+      );
+      const bucketStorage = (await getTestStorage(factory, syncRules)) as MongoSyncBucketStorageV3;
+      await using writer = await bucketStorage.createWriter(test_utils.BATCH_OPTIONS);
+      const sourceTable = await test_utils.resolveTestTable(writer, 'items', ['id'], factoryGen, 1);
+      await writer.markAllSnapshotDone('1/1');
+      await writer.save({
+        sourceTable,
+        tag: storage.SaveOperationTag.INSERT,
+        after: { id: 'A', description: 'unchanged' },
+        afterReplicaId: test_utils.rid('A')
+      });
+      await writer.commit('1/1');
+
+      const definitionId = syncRules.syncConfigContent[0].mapping.allBucketDefinitionIds()[0];
+      const db = bucketStorage.db;
+      const collection = db.bucketData(bucketStorage.replicationStreamId, definitionId);
+      const [before] = await collection.find({}).toArray();
+      expect(before.ops != null).toBe(initiallyInline);
+      expect(before.storage_ref != null).toBe(!initiallyInline);
+      const checkpoint = await bucketStorage.getCheckpoint();
+      const request = bucketRequest(syncRules.syncConfigContent[0], 'global|0[]', 0n);
+      const dataBefore = await test_utils.getBatchArray(bucketStorage.getBucketDataBatch(checkpoint, [request]));
+
+      // Simulate changing the threshold after this chunk was written.
+      const reconfiguredStorage = new MongoSyncBucketStorageV3(
+        factory,
+        bucketStorage.replicationStreamId,
+        bucketStorage.replicationStream,
+        bucketStorage.replicationStreamName,
+        undefined,
+        {
+          storageConfig: bucketStorage.storageConfig,
+          objectStorage: memoryStorage,
+          inlineThresholdBytes: before.size + thresholdOffset
+        }
+      );
+      const put = vi.spyOn(memoryStorage, 'put');
+      expect(
+        await reconfiguredStorage
+          .createMongoCompactor({
+            maxOpId: checkpoint.checkpoint,
+            ...(compactChunksOnly
+              ? { compactChunksOnly: true, forceChunkCompaction: true }
+              : { compactBuckets: [request.bucket] })
+          })
+          .compact()
+      ).toBe(1);
+
+      const documents = await collection.find({}).toArray();
+      expect(documents).toHaveLength(1);
+      const [after] = documents;
+      if (initiallyInline && thresholdOffset < 0) {
+        expect(after.storage_ref).toBeDefined();
+        expect(after.ops).toBeUndefined();
+        expect(put).toHaveBeenCalledTimes(1);
+        const { ops, ...metadata } = before;
+        expect(after).toMatchObject(metadata);
+      } else {
+        expect(after).toEqual(before);
+        expect(put).not.toHaveBeenCalled();
+      }
+      await expectUsageMatchesBucketData(reconfiguredStorage, definitionId);
+      const dataAfter = await test_utils.getBatchArray(reconfiguredStorage.getBucketDataBatch(checkpoint, [request]));
+      expect(test_utils.getBatchData(dataAfter)).toEqual(test_utils.getBatchData(dataBefore));
+    }
+  );
+
   test('a later compaction recovers from a transient object storage failure', async () => {
     const { memoryStorage, factory: factoryGen } = memoryS3Factory();
     await using factory = await factoryGen.factory();

package/test/src/storage_sync.test.ts

@@ -1,4 +1,5 @@
 import { MongoSyncBucketStorageV3 } from '@module/storage/implementation/v3/MongoSyncBucketStorageV3.js';
+import { ReplicationAbortedError } from '@powersync/lib-services-framework';
 import {
   deserializeParameterLookup,
   JwtPayload,
@@ -7,7 +8,7 @@ import {
   updateSyncRulesFromYaml
 } from '@powersync/service-core';
 import { bucketRequest, getTestStorage, register, test_utils } from '@powersync/service-core-tests';
-import { RequestParameters, ScopedParameterLookup, SqlSyncRules } from '@powersync/service-sync-rules';
+import { parseSyncConfig, RequestParameters, ScopedParameterLookup } from '@powersync/service-sync-rules';
 import * as bson from 'bson';
 import { describe, expect, test } from 'vitest';
 import { MongoBucketStorage } from '../../src/storage/MongoBucketStorage.js';
@@ -72,7 +73,7 @@ function objectIdGenerator(id: string) {
  * `ResolveTablesOptions.parsedSyncConfig`.
  */
 function parsedSyncConfigSetFor(yaml: string, storageVersion: number) {
-  const parsed = SqlSyncRules.fromYaml(yaml, test_utils.PARSE_OPTIONS);
+  const parsed = parseSyncConfig(yaml, test_utils.PARSE_OPTIONS);
   // These fixtures are legacy Sync Rules, which always report a deprecation warning.
   expect(parsed.errors.filter((error) => error.type !== 'warning')).toEqual([]);
   return new MongoParsedSyncConfigSet(1, getMongoStorageConfig(storageVersion), 'test_slot', [
@@ -939,6 +940,70 @@ event_definitions:
     expect(deserializeParameterLookup(parameterEntry!.lookup)).toEqual(['shape-check']);
   });
 
+  test.runIf(storageVersion >= 3)('fetches parameter sets beyond the aggregation stage limit', async () => {
+    await using factory = await storageConfig.factory();
+    const syncRules = await factory.updateSyncRules(
+      updateSyncRulesFromYaml(
+        /* yaml */ `
+          # Parameter lookup batching regression fixture.
+          config:
+            edition: 3
+          streams:
+            by_owner:
+              auto_subscribe: true
+              query: |
+                SELECT * FROM test
+                WHERE id IN (SELECT owner_id FROM test WHERE id = auth.parameter('test'))
+        `,
+        { storageVersion }
+      )
+    );
+    const bucketStorage = await getTestStorage(factory, syncRules);
+    await using writer = await bucketStorage.createWriter(test_utils.BATCH_OPTIONS);
+    const sourceTable = await test_utils.resolveTestTable(writer, 'test', ['id'], INITIALIZED_MONGO_STORAGE_FACTORY);
+
+    for (const id of ['row-first', 'row-middle', 'row-last']) {
+      await writer.save({
+        sourceTable,
+        tag: storage.SaveOperationTag.INSERT,
+        after: { id, description: id, owner_id: `user-${id}` },
+        afterReplicaId: test_utils.rid(id)
+      });
+    }
+    await writer.markAllSnapshotDone('1/1');
+    await writer.commit('1/1');
+
+    const parsed = syncRules.syncConfigContent[0].parsed(test_utils.PARSE_OPTIONS);
+    const [parameterSource] = [...parsed.syncConfigs[0].config.bucketParameterLookupSources];
+    const scope = parsed.hydrationState.getParameterIndexLookupScope(parameterSource);
+
+    // MongoDB rejects aggregation pipelines with more than 1000 stages, and each
+    // lookup after the first adds a $unionWith stage (#873).
+    const lookups: ScopedParameterLookup[] = [];
+    for (let i = 0; i < 996; i++) {
+      let id = `missing-${i}`;
+      if (i == 0) {
+        id = 'row-first';
+      } else if (i == 900) {
+        id = 'row-middle';
+      } else if (i == 995) {
+        id = 'row-last';
+      }
+      lookups.push(ScopedParameterLookup.direct(scope, [id]));
+    }
+
+    const checkpoint = await bucketStorage.getCheckpoint();
+    const parameterSets = await checkpoint.getParameterSets(lookups, 1000);
+
+    expect(parameterSets.length).toEqual(3);
+    await expect(checkpoint.getParameterSets(lookups, 2)).rejects.toThrow('Too many parameter results');
+    expect(await checkpoint.getParameterSets(lookups, 3)).toEqual(parameterSets);
+    const rowsByKey = new Map(parameterSets.map((set) => [set.lookup.indexKey[0], set.rows]));
+    expect(rowsByKey.get('row-first')).toEqual([{ '0': 'user-row-first' }]);
+    expect(rowsByKey.get('row-middle')).toEqual([{ '0': 'user-row-middle' }]);
+    expect(rowsByKey.get('row-last')).toEqual([{ '0': 'user-row-last' }]);
+  });
+
   test.runIf(storageVersion >= 3)('replaces an existing deploying sync config', async () => {
     await using factory = await storageConfig.factory();
 
@@ -1343,6 +1408,22 @@ streams:
       expect(updatedActiveStorage.replicationStream.replicationJobId).toContain(
         (await factory.getActiveSyncConfig())!.content.syncConfigId
       );
+
+      // The full-stream writer still holds the old config's definition mapping.
+      // Both metadata writes and the next checkpoint must exit so a replacement
+      // job can load only live mappings.
+      await expect(
+        secondWriter.resolveTables({
+          connection_id: 1,
+          source: sourceDescriptor('todos', { objectId: 'todos-relation-after-activation' }),
+          idGenerator: objectIdGenerator('6544e3899293153fa7b3835e')
+        })
+      ).rejects.toBeInstanceOf(ReplicationAbortedError);
+      await expect(secondWriter.commit('3/1')).rejects.toBeInstanceOf(ReplicationAbortedError);
+      const persistedStream = (await (factory as MongoBucketStorage).db.sync_rules.findOne({
+        _id: first.replicationStreamId
+      })) as ReplicationStreamDocumentV3;
+      expect(persistedStream.resume_lsn).toBe('2/1');
     }
   );