Merged
Conversation
dejan-velimirovic-calendly
suggested changes
Feb 20, 2026
- Remove 30s waitUntil delay from per-batch task re-queue (was causing test timeouts since the original code had no such delay) - Add failedChunkData JSON field to batch collection so per-batch tasks can store chunk-level failure data independently - Aggregate failedChunkData from batch records in finalizeRunIfComplete() instead of relying on in-memory accumulation from the old single-task flow Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…rker architecture Splits prepare-bulk-embedding into coordinator + per-collection workers. Each worker processes one page of one collection, queuing a continuation job before processing to ensure crash safety. Default batchLimit is 1000 when not explicitly set. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The second test was creating a separate Payload instance sharing the same DB and job queues, causing two crons to compete for jobs. This led to double-execution and mock state inconsistency (expected 4 to be 2). Now both tests use the single beforeAll instance with cleanup between. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Every test file that creates a Payload instance now calls payload.destroy() in afterAll (or try/finally for in-test instances). This stops background cron jobs from accumulating across tests, which was causing heap exhaustion in CI. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add --max-old-space-size=8192 to test:int NODE_OPTIONS (cross-env was overriding the CI env var, so the heap limit never took effect) - Fix polling.spec.ts queueSpy assertions: coordinator/worker adds an extra queue call, so poll-or-complete-single-batch is now call 3 and 4 instead of 2 and 3 - Add extensive [vectorize-debug] console.log throughout task handlers (coordinator, worker, poll-single, finalize, streamAndBatchDocs) to diagnose any remaining CI hangs - Remove redundant NODE_OPTIONS from CI workflow (now in the script) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ncrementally Remove the backward-compatible fan-out task since the per-batch architecture hasn't been released yet. Refactor finalizeRunIfComplete to aggregate batch counts incrementally during pagination instead of collecting all batch objects into memory. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Bump version 0.5.4 → 0.5.5 - Add 0.5.5 entry to CHANGELOG.md (coordinator/worker, batchLimit, per-batch polling) - Document batchLimit in README CollectionVectorizeOption section - Remove all diagnostic console.log statements from bulkEmbedAll.ts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
techiejd
added a commit
that referenced
this pull request
Feb 26, 2026
* adds should embed (#38) * adds should embed * Ups version to get ready for release * splits the job into one per batch (#41) * splits the job into one per batch * fix: remove waitUntil delay and persist failedChunkData on batch records - Remove 30s waitUntil delay from per-batch task re-queue (was causing test timeouts since the original code had no such delay) - Add failedChunkData JSON field to batch collection so per-batch tasks can store chunk-level failure data independently - Aggregate failedChunkData from batch records in finalizeRunIfComplete() instead of relying on in-memory accumulation from the old single-task flow Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat: add batchLimit to CollectionVectorizeOption with coordinator/worker architecture Splits prepare-bulk-embedding into coordinator + per-collection workers. Each worker processes one page of one collection, queuing a continuation job before processing to ensure crash safety. Default batchLimit is 1000 when not explicitly set. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: rewrite batchLimit test 2 to reuse same Payload instance The second test was creating a separate Payload instance sharing the same DB and job queues, causing two crons to compete for jobs. This led to double-execution and mock state inconsistency (expected 4 to be 2). Now both tests use the single beforeAll instance with cleanup between. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: add payload.destroy() in afterAll to prevent OOM from leaked crons Every test file that creates a Payload instance now calls payload.destroy() in afterAll (or try/finally for in-test instances). This stops background cron jobs from accumulating across tests, which was causing heap exhaustion in CI. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Trying to not destroy our heap * Runs tests in parallel now that each test gets its own db * WIP * fix: fix OOM, polling test assertions, and add diagnostic logging - Add --max-old-space-size=8192 to test:int NODE_OPTIONS (cross-env was overriding the CI env var, so the heap limit never took effect) - Fix polling.spec.ts queueSpy assertions: coordinator/worker adds an extra queue call, so poll-or-complete-single-batch is now call 3 and 4 instead of 2 and 3 - Add extensive [vectorize-debug] console.log throughout task handlers (coordinator, worker, poll-single, finalize, streamAndBatchDocs) to diagnose any remaining CI hangs - Remove redundant NODE_OPTIONS from CI workflow (now in the script) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor: remove poll-or-complete-bulk-embedding task and aggregate incrementally Remove the backward-compatible fan-out task since the per-batch architecture hasn't been released yet. Refactor finalizeRunIfComplete to aggregate batch counts incrementally during pagination instead of collecting all batch objects into memory. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore: bump to 0.5.5, update changelog, remove debug logging - Bump version 0.5.4 → 0.5.5 - Add 0.5.5 entry to CHANGELOG.md (coordinator/worker, batchLimit, per-batch polling) - Document batchLimit in README CollectionVectorizeOption section - Remove all diagnostic console.log statements from bulkEmbedAll.ts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Adds upgrade note --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * chore: bump version to 0.6.0-beta.5 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: resolve 4 CI test failures from merge - chunkers.spec.ts: remove getPayload() call that crashes on dummy db, pass SanitizedConfig directly to chunkRichText - batchLimit.spec.ts: add missing dbAdapter (createMockAdapter) required by split_db_adapter architecture - extensionFieldsVectorSearch.spec.ts: pass adapter as second arg to createVectorSearchHandlers (new signature from split_db_adapter) - versionBump.spec.ts: destroy payload0 before creating payload1 to prevent cron worker race condition between two instances Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Cleans a nit double line * Undoes a weird test fix done by the bot * fix: harden versionBump test with sequential steps and queue isolation - Use test.step() to enforce sequential execution of each phase - Add separate realtimeQueueName per payload instance to prevent cron worker cross-talk on the default queue - Use dynamic Date.now() keys to avoid cached state interference - Increase waitForBulkJobs timeout to 30s for CI Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: prevent waitForBulkJobs from returning prematurely waitForBulkJobs could return early in the coordinator/worker fan-out pattern when there's a brief window with 0 pending jobs between job transitions. Now it also checks the bulk embeddings run status — only returns when both no pending jobs exist AND no runs are in queued/running state. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: remove test.step() — not available in Vitest test.step() is a Playwright API, not Vitest. Reverted to flat sequential code with phase comments for readability. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: rewrite versionBump test with single Payload instance Instead of creating two Payload instances (which caused cron cross-talk, timeout, and queue isolation issues on CI), use one instance and mutate the knowledgePools config version between bulk embed runs. Tests the same code path (versionMismatch in streamAndBatchDocs) without the multi-instance fragility. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
techiejd
added a commit
that referenced
this pull request
Mar 5, 2026
* WIP WIP WIP WIP Uses mock adapter WIP WIP WIP WIP WIP WIP * fix: ignore node_modules everywhere * Adds split_db_adapter to CI run * feat(cf-adapter): add Cloudflare Vectorize adapter (#28) * feat(cf-adapter): add Cloudflare Vectorize adapter * feat(cf-adapter): enhance Cloudflare Vectorize integration with config-based bindings and add tests * feat(cf-adapter): refactor Cloudflare Vectorize integration to use config-based bindings and update tests * chore: update pnpm-lock.yaml * Preparing for automated pubishes. This one beta will be done by hand but hopefully the rest will be done automatically * Bumps version since we added deleteEmbeddings. Also runs tsc so that we can be sure the whole project compiles * Adds the type check to ci * fixes type check * removes silly double checking on split_db_adapter for push * Adds root pnpm workspace * feat(cf-adapter): update query parameters and method for deleting embeddings in Cloudflare Vectorize integration (#31) * Bumps version to release * Better typings (#34) * Adds better id tracking for deletion and does only one search instead of many for querying (#35) * Deduplicate shared logic across plugin and adapter packages (#36) * Deduplicate shared logic across plugin and adapter packages Extract repeated production patterns (chunk validation, delete embeddings, task slug constants) into shared utilities exported from the root plugin. Consolidate test helpers via vitest path aliases so adapter tests import from the canonical root dev/ copies. Remove CF adapter dead test code (unused utils, constants, helpers). Fix chunkRichText join bug in CF adapter tests (was joining child nodes without spaces). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * cf adapter limitation acknowledgement and more DRY --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * Removes dead code (#37) * Bumps version for rollout * Merge main (#40) * adds should embed (#38) * adds should embed - merged * Merge main into split_db_adapter (beta.5) (#42) * adds should embed (#38) * adds should embed * Ups version to get ready for release * splits the job into one per batch (#41) * splits the job into one per batch * fix: remove waitUntil delay and persist failedChunkData on batch records - Remove 30s waitUntil delay from per-batch task re-queue (was causing test timeouts since the original code had no such delay) - Add failedChunkData JSON field to batch collection so per-batch tasks can store chunk-level failure data independently - Aggregate failedChunkData from batch records in finalizeRunIfComplete() instead of relying on in-memory accumulation from the old single-task flow Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat: add batchLimit to CollectionVectorizeOption with coordinator/worker architecture Splits prepare-bulk-embedding into coordinator + per-collection workers. Each worker processes one page of one collection, queuing a continuation job before processing to ensure crash safety. Default batchLimit is 1000 when not explicitly set. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: rewrite batchLimit test 2 to reuse same Payload instance The second test was creating a separate Payload instance sharing the same DB and job queues, causing two crons to compete for jobs. This led to double-execution and mock state inconsistency (expected 4 to be 2). Now both tests use the single beforeAll instance with cleanup between. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: add payload.destroy() in afterAll to prevent OOM from leaked crons Every test file that creates a Payload instance now calls payload.destroy() in afterAll (or try/finally for in-test instances). This stops background cron jobs from accumulating across tests, which was causing heap exhaustion in CI. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Trying to not destroy our heap * Runs tests in parallel now that each test gets its own db * WIP * fix: fix OOM, polling test assertions, and add diagnostic logging - Add --max-old-space-size=8192 to test:int NODE_OPTIONS (cross-env was overriding the CI env var, so the heap limit never took effect) - Fix polling.spec.ts queueSpy assertions: coordinator/worker adds an extra queue call, so poll-or-complete-single-batch is now call 3 and 4 instead of 2 and 3 - Add extensive [vectorize-debug] console.log throughout task handlers (coordinator, worker, poll-single, finalize, streamAndBatchDocs) to diagnose any remaining CI hangs - Remove redundant NODE_OPTIONS from CI workflow (now in the script) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor: remove poll-or-complete-bulk-embedding task and aggregate incrementally Remove the backward-compatible fan-out task since the per-batch architecture hasn't been released yet. Refactor finalizeRunIfComplete to aggregate batch counts incrementally during pagination instead of collecting all batch objects into memory. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore: bump to 0.5.5, update changelog, remove debug logging - Bump version 0.5.4 → 0.5.5 - Add 0.5.5 entry to CHANGELOG.md (coordinator/worker, batchLimit, per-batch polling) - Document batchLimit in README CollectionVectorizeOption section - Remove all diagnostic console.log statements from bulkEmbedAll.ts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Adds upgrade note --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * chore: bump version to 0.6.0-beta.5 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: resolve 4 CI test failures from merge - chunkers.spec.ts: remove getPayload() call that crashes on dummy db, pass SanitizedConfig directly to chunkRichText - batchLimit.spec.ts: add missing dbAdapter (createMockAdapter) required by split_db_adapter architecture - extensionFieldsVectorSearch.spec.ts: pass adapter as second arg to createVectorSearchHandlers (new signature from split_db_adapter) - versionBump.spec.ts: destroy payload0 before creating payload1 to prevent cron worker race condition between two instances Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Cleans a nit double line * Undoes a weird test fix done by the bot * fix: harden versionBump test with sequential steps and queue isolation - Use test.step() to enforce sequential execution of each phase - Add separate realtimeQueueName per payload instance to prevent cron worker cross-talk on the default queue - Use dynamic Date.now() keys to avoid cached state interference - Increase waitForBulkJobs timeout to 30s for CI Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: prevent waitForBulkJobs from returning prematurely waitForBulkJobs could return early in the coordinator/worker fan-out pattern when there's a brief window with 0 pending jobs between job transitions. Now it also checks the bulk embeddings run status — only returns when both no pending jobs exist AND no runs are in queued/running state. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: remove test.step() — not available in Vitest test.step() is a Playwright API, not Vitest. Reverted to flat sequential code with phase comments for readability. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: rewrite versionBump test with single Payload instance Instead of creating two Payload instances (which caused cron cross-talk, timeout, and queue isolation issues on CI), use one instance and mutate the knowledgePools config version between bulk embed runs. Tests the same code path (versionMismatch in streamAndBatchDocs) without the multi-instance fragility. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * chore: update beta version * Remove Cloudflare adapter to unblock main branch merge Split the CF adapter work out so the core DbAdapter API and pg adapter can be merged to main independently. The CF adapter will continue in a separate branch. * docs: call for help adding more database adapters * chore: bump version to 0.7.0 --------- Co-authored-by: dejan-velimirovic-calendly <dejan.velimirovic@calendly.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.