fix(workspace-files): keep the file-search claim on the ordered pending index instead of a hash join

[waleedlatif1]

Summary

• The file-search dispatcher's claim query joined workspace_files with an INNER JOIN; under FOR UPDATE (no parallel plans) the planner mis-estimates the timestamp equi-join and picks a hash join over every file and every pending revision of the workspace before the top-N sort, which exceeded the 10 s statement timeout for a workspace with a large backlog and failed the dispatch task with SQLSTATE 57014
• The join is now a correlated LATERAL lookup with LIMIT 1, which cannot be flattened into a hash join, so the claim walks the ordered pending index and stops after the batch size: measured 4.6 s → 0.7 ms on the same workspace, same predicates and lock target

Type of Change

• Bug fix

Testing

• Added an integration test that captures the exact claim statement the dispatcher runs, pins the planner with SET LOCAL enable_nestloop = off (at fixture scale it already nests the join, so this reproduces production's choice), and asserts the LockRows subtree is fed by an index scan of workspace_file_search_revision_pending_idx with no sort over the backlog. Verified it fails against the old INNER JOIN spelling (LockRows -> Sort -> Hash Join) and passes with the LATERAL
• vitest run --mode integration lib/workspace-files/search/dispatcher.integration.ts against a local disposable PostgreSQL: 9 passed
• vitest run lib/workspace-files/search/dispatcher.test.ts lib/workspace-files/search/enqueue-dispatch.test.ts: 11 passed
• bun run type-check in apps/sim: clean
• bun run lint: clean
• bun run check:audits: every content audit passed (including check:api-validation:strict, check:sql-date-binding, check:utils, check:source-text); the only failures were audits that shell out to turbo/biome/tsc binaries missing from the isolated worktree, unrelated to this change

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
docs	Skipped		Sep 22, 2026 5:43am UTC

[cubic-dev-ai]

No issues found across 2 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge; the query rewrite preserves claim semantics and adds focused coverage for the production performance regression.

Summary

This PR rewrites the file-search claim query's live-file check as a correlated LATERAL lookup, preventing the claim from degrading into a hash join and backlog-wide sort while preserving candidate eligibility and row-locking behavior.

• Keeps candidate selection on workspace_file_search_revision_pending_idx.
• Retains the existing workspace, deletion, context, and content-version checks.
• Adds an integration regression test that captures the production claim SQL and verifies the ordered index-backed plan.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Select queued workspace] --> B[Walk ordered pending revision index]
  B --> C[Probe matching live workspace file via LATERAL LIMIT 1]
  C --> D{Eligible and unlocked?}
  D -- No --> B
  D -- Yes --> E[Lock revision with SKIP LOCKED]
  E --> F{Workspace and global capacity reached?}
  F -- No --> B
  F -- Yes --> G[Mark claimed revisions dispatched]

Loading

_{Reviews (1) · Last reviewed commit: "fix(workspace-files): keep the file-sear..."}