Prevent stale raw-data archive links

[Freika]

Production signal

Root cause

The archiver compressed a raw-data snapshot, then unconditionally linked point rows after upload and verification. A concurrent tracker upsert in that interval could change a point before it was linked to the stale snapshot.

Fix

• Record a checksum for every raw-data snapshot while compressing
• Row-lock flagging batches in conflict-key order and link only rows whose current raw data still matches
• Retry transient database contention while flagging
• Leave changed rows eligible for a later archive pass using the current snapshot
• Restore/cache only snapshots still linked to their source archive, with database rows locked through restore
• Namespace temporary recovery cache entries by archive so detached/relinked points cannot read stale snapshots
• Report actual linked counts in stats and metrics

Verification

• asdf exec bundle exec rspec spec/models/concerns/archivable_spec.rb spec/services/points/raw_data/chunk_compressor_spec.rb spec/services/points/raw_data/archiver_spec.rb spec/services/points/raw_data/restorer_spec.rb — 80 examples, 0 failures
• asdf exec bundle exec rubocop app/models/concerns/archivable.rb app/services/points/raw_data/chunk_compressor.rb app/services/points/raw_data/archiver.rb app/services/points/raw_data/restorer.rb spec/models/concerns/archivable_spec.rb spec/services/points/raw_data/chunk_compressor_spec.rb spec/services/points/raw_data/archiver_spec.rb spec/services/points/raw_data/restorer_spec.rb — 8 files, no offenses
• git diff --check — passed

Existing legacy stale links still require an explicitly approved backup-first repair procedure. No production data mutation is included.

Review status

REQUEST_CHANGES — do not merge yet. Stable-head engineering and QA review still find lock-order deadlock risk: PostgreSQL geography ordering differs from ingestion's normalized numeric longitude/latitude order, and restore retains locks across independently ordered batches without contention retry. The two permitted fix/re-review cycles were exhausted.

GitHub writes: none.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 9cd1270c-34bf-4a98-8f2e-36a763e49068

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/raw-data-archive-flag-race

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}