Skip to content

v0.9.6 — Stop write churn in search, schedules and crawls

Choose a tag to compare

@willgriffin willgriffin released this 03 Nov 04:03
df3bb6c

Stops continuous rewrites that drove ~9 GB/day of WAL on jobgeni_prod (#288, fixes #286):

  • Search documents are content-hashed; only changed docs are upserted and only removed ones deleted. Fixes a race where public search briefly showed a partial catalogue.
  • public_search_generation bumps only on real writes.
  • Schedule sync saves a source only when nextCheckAt changes, and keeps the crawl job's value.
  • Provenance and crawl-item normalizers skip rows already in their target state.
    First run after deploy backfills content_hash once (~6.4k rows).

Resume image candidates

Site: ghcr.io/willgriffin/iolaus/site@sha256:80e7e96a593e29d9ceab4f93f9e2667d41fd2c2646ad320cd5fc5d3bc4a02c7c

Extractor: ghcr.io/willgriffin/iolaus/resume-import-extraction@sha256:112482985632b901d46443becf23f908dfe5a4cf3eaf2191a3c346ce6e06c599

Source: df3bb6c267756e3f4db3388b0fe71e924dd73634;
lock SHA-256: d3831e1c67837e74326118b05520b6966fa079e4d1baa8682ebded2ff0107ecc.

Envelope tested. Production requires qualification:
docs/resume-extractor-release.md.