Conversation
This PR completes the ingestion workflow improvement tasks for the MultiDoc RAG project. The changes improve the PDF ingestion pipeline across backend validation, background worker reliability, batch upload support, frontend upload UX, and workflow documentation/reporting.
Fix token budget lock contention for streaming queries and Supabase setup
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
This PR fixes token budget lock contention that could cause streaming query failures with PostgreSQL statement
timeouts.
The issue showed up around
workspace_daily_usageupdates during query execution. The previous flow could keep thetoken budget row lock open too long while a request was still doing retrieval, LLM work, or SSE streaming. Under
concurrent requests for the same workspace, that created blocking and timeout failures.
What changed
workspace_daily_usagerow lock across long-running query and streaming workserver/scripts/bootstrap_schema.pyserver/scripts/inspect_schema.pydocs/task/ingestion-follow-up-hardening-and-validation.mdRoot cause
The budget reservation path used the request DB session. In API flows, that session could already be inside an open
transaction before
reserve_tokens()was called. That meant the row lock onworkspace_daily_usagecould remain openmuch longer than intended, especially in
/query/stream, where the request stays alive while tokens are streamedback.
Concurrent requests for the same workspace/day could then block on the same row and eventually fail with a statement
timeout.
Result
This change keeps budget mutations atomic, but limits the lock lifetime to the budget update itself instead of the
full request lifecycle.
Validation
regression from this change