Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

Flaky: note_vector_flow_test 'forty notes' under parallel test load

Open
#282 1 comment 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
1-2 days
Newbie friendliness
30/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active
Tech stack
dart
Domain
testing

Research direction

Start with test/integration/note_vector_flow_test.dart, the test named "forty notes are each found first by their own text", and the drained() helper and fake embedder it uses. Reproduce by running the file repeatedly under --concurrency=4 and record the found count on any failure, which separates a drain race from ANN recall misses. Done means the flake is explained with evidence and either the assertion or the drain wait is changed so repeated parallel runs pass.

Written by the indexing model from the issue text.

Description

P3-Low

Summary

test/integration/note_vector_flow_test.dart › "forty notes are each found first by their own text" failed once during a full-suite run at --concurrency=4 (all other runs pass). Root cause not confirmed; filing so it is tracked rather than rediscovered.

What I verified

  • Passes in isolation and 3 of 3 repeated runs of the file at --concurrency=4, and in 3 full-suite runs of #88 work.
  • The failure was observed once, while the full suite was running 4 test files in parallel. I did not capture the assertion output (found count) at the time.

Likely causes (hypotheses, not proven)

  1. Timing under CPU contention: the test publishes 40 notes then waits on drained(); the fake embedder awaits a 2 ms Future.delayed per note and the test has a 2-minute timeout. Under 4-way parallel isolates plus tostore's own isolate, drained() may return before the last vector is flushed, or ask() can run while the index is still being built.
  2. tostore index state: AGENTS/CLAUDE notes that tostore 3.5 ANN search returns 99-100% recall (not 100%) on stored vectors; with 40 near-identical "alpha beta" tails an ANN miss would give found == 39. A genuine 99% recall figure makes this assertion (found == 40) inherently probabilistic.

Suggested next steps

  • Re-run the file in a loop (50x) under --concurrency=4 and record found on failure to tell (1) from (2).
  • If it is ANN recall, relax the assertion to a threshold (for example ≥ 38 of 40) or make the test deterministic by querying with the exact stored vector.
  • If it is a drain race, have drained() also wait for the store's flush.

Not a product bug as far as I can tell; it is CI flakiness risk. Refs #88

Dominant language
Dart
Stars
2
Forks
2
Avg merge
1d 35m
Merged PRs (30d)
36

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from basictech01/uniun

All issues in basictech01/uniun

Similar issues

More Dart issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.