Skip to content

embed_names runs on the fast lane, so name vectors keep up with extraction - #16

Merged
gedaiu merged 1 commit into
re-devfrom
feat/embed-names-fast-lane
Oct 22, 2026
Merged

gedaiu merged 1 commit into
re-devfrom
feat/embed-names-fast-lane

Conversation

@gedaiu

@gedaiu gedaiu commented Oct 21, 2026

Copy link
Copy Markdown

The embed_names job only calls the embedding model, but job_priority (migration 0116) does not list it, so it gets priority 1. It waits behind extraction and the reserved worker slots never claim it. Migration 9003 adds it to the priority-0 list. The index jobs_priority_claim_idx is on job_priority(kind) and stores the computed value, so changing the function alone would leave existing entries stale. The migration rebuilds the index too, and only when the function does not yet list embed_names, so a second run changes nothing.

I ran the migrate example twice against a database that already had the 107 earlier migrations, a new test (embed_names has priority 0 and is claimed while an extraction holds the only regular slot; it fails without the migration), cargo test -p utopia-store --test store (330 passed), cargo test -p utopia-cli (schema version now 108), fmt and clippy with -D warnings.

🤖 Generated with Claude Code

…ehind extraction.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Signed-off-by: Bogdan Szabo <[email protected]>
@gedaiu
gedaiu merged commit 05011bd into re-dev Oct 22, 2026
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant