Repository navigation
One embed_names job per base: a running copy counts as queued - #46
Merged
Merged
Conversation
The per-document enqueue only looked at queued copies, so once the first embed_names claimed, every document extracted while it ran queued another, and on a base whose pending scan takes hours (48k identity names, deeplethe#1472) 14 copies sat behind it, each scanning the same facts. The enqueue now skips a running copy too, and the job re-enqueues itself whenever it filled anything, so names written while it ran are picked up by the next one. Refs deeplethe#1472. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
gedaiu
added a commit
to re-cinq/lore
that referenced
this pull request
Nov 1, 2026
…job per base (#4062) Bumps the Utopia pin to `0.1.0-re.27` (digest `sha256:55321d80…`), the re-dev head after two fork merges: - re-cinq/utopia#45: `signature_keys_of_document` reaches `fact_evidence` by its own `document_id` column (indexed) instead of through `chunks`, which seq-scanned all 8.3M evidence rows per extracted document (12 s each, 28 to 30 at once). Measured 0.16 s with the index. Upstream: deeplethe/utopia#1471. - re-cinq/utopia#46: one `embed_names` job per base. A per-document enqueue now counts a running copy as queued, so 14 parallel 1 to 3 h backfills over the same 48k identity names no longer pile up. Upstream: deeplethe/utopia#1472. No new migration. The terraform apply is Bogdan's, from main. Related: #4025, #4057. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
embed_names_jobis fork-only (0c56fd1), and its per-document enqueue usedenqueue_unless_queued, which looks at queued copies only. Once the first copy was claimed, every document extracted while it ran queued another one; on the lore base, wherepending()walks 48,045 identity names to find 20 prose ones (deeplethe#1472), 14 copies ran at once on 2026-10-31, each 1 to 3 h, each blocked inMultiXactOffsetSLRU.enqueue_unless_pending(queued or running, no debounce), the helper the ontology page already uses for the same pile-up.enqueue_unless_queuedwhenever it filled anything, not only when a batch was full, so names written while it ran are picked up by the next copy; a run that fills nothing ends the chain. At most one idle extra run per burst.Tests:
a_name_embedding_runs_while_model_work_holds_every_slotandextraction_open(18) pass against a fresh database. The real fix for the scan itself is upstream deeplethe#1472.Related: #45 (the signature scan beside it).
🤖 Generated with Claude Code