Skip to content

COMMIT_AUTHORS_CACHING.md

Latest commit

 

History

History
47 lines (36 loc) · 2.42 KB

COMMIT_AUTHORS_CACHING.md

File metadata and controls

47 lines (36 loc) · 2.42 KB

EasyCLA: Author and Co-author Caching + Large-PR Support

  • Two-level caching for author and co-author identity & identity plus per-project signature decisions.
  • Caching for git co-authors parsed from commit messages.
  • GraphQL-based commit ingestion that comfortably handles PRs with 250+ commits (and beyond).

Why it matters

  • Faster PR checks and /easycla re-runs.
  • Lower DB/API load via memoized decisions.
  • Stable, deterministic output and accurate status posting on the PR head SHA.

Caching

  • Co-author cache keys are based on normalized email and name from the commit trailers (Co-authored-by:).
  • General cache key: (author_id, lower(login), lower(email)) → (user | None)
  • Per-project cache key: (project_id, author_id, lower(login), lower(email)) → (user | None, authorized, affiliated)
  • TTL policy: positives ~12h (~3h for per-project with signature status); negative/uncertain states use Negative TTL = 3m.
  • Flow: per-project cache → general cache → cold DB path. Results are stored back with the appropriate TTL.
  • When signature is signed: per-project and general caches are updated to reflect the new status (general cache is updated because given user could have no DynamoDB entry yet before signing the CLA).
  • Thread-safe with periodic expired entries cleanup (once per hour).
  • There are /v2/clear-cache and /v4/clear-cache endpoints to clear caches (testing & ops).

Large PR (250+) support

  • Switch to GitHub GraphQL for commits (pageSize=100) with cursor paging.
  • Parallel processing via thread pool; co-authors parsed from commit messages (Co-authored-by:).
  • Final actor lists are de-duplicated and sorted (login, name, email, sha) for stable comments.
  • PR comments are edited only when normalized body changes (prevents churn & size bloat).
  • Commit statuses are always posted to the true PR head SHA.

Operational notes

  • Expect noticeable latency reduction on large PRs and repeated checks.
  • Fallbacks remain safe; unknown users land in an “Unknown” bucket with guidance.
  • No behavior change to the core signing rules—only faster execution.

Constants

  • NEGATIVE_CACHE_TTL = 180 seconds (negative/uncertain states).
  • Default positive cache TTL ≈ 12 hours (3 hours for per-project with signature status).
  • GraphQL: pageSize=100, parallel workers tuned for throughput.