release: v0.2.5850 - #735
Conversation
release: v0.2.5850 concurrent warm queries
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 163420b9fb
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| defer cliWorkerSlotRelease(active_queries); | ||
| defer std.heap.c_allocator.destroy(job); | ||
|
|
||
| const retire = cliServeConn(job.io, std.heap.c_allocator, job.explorer, job.store, job.abs_root, job.conn); |
There was a problem hiding this comment.
When two snapshot/reindex notifications arrive together, both workers can enter watcher.refreshIndex for the same Explorer and Store. Each refresh seeds an independent pre-refresh known map, and applyKnownFile records a snapshot before parsing and committing content; consequently both refreshes can append duplicate change events, or a slower refresh that read an older file version can commit after a newer refresh and leave the live index stale. Refresh sentinels need an exclusive gate rather than the concurrent read-query worker path.
Useful? React with 👍 / 👎.
| defer cliWorkerSlotRelease(active_queries); | ||
| defer std.heap.c_allocator.destroy(job); | ||
|
|
||
| const retire = cliServeConn(job.io, std.heap.c_allocator, job.explorer, job.store, job.abs_root, job.conn); |
There was a problem hiding this comment.
With simultaneous search or context requests, this call runs Explorer searches concurrently even though Explorer.last_search_breakdown is a plain shared struct: searchContentUncached writes it while holding only the shared Explorer lock, and cache insertion, MCP telemetry, and JSON provenance subsequently read it. One request can therefore cache or report another request's tier data, with unsynchronized writes constituting a data race; keep the breakdown request-local or protect it before enabling parallel dispatch.
AGENTS.md reference: AGENTS.md:L18-L18
Useful? React with 👍 / 👎.
CodeDB 0.2.5850
Ships bounded concurrent warm CLI query dispatch for multi-agent and multi-terminal workloads on macOS and Linux.
Measured on eight concurrent hybrid context calls: 6.44 s to 2.69 s batch wall time (58% reduction), with 8/8 successful and parity preserved.
Validation: 263 unit tests passed (4 skipped), 76/76 MCP E2E, five repeated optimized live-reindex passes, macOS resource regression green, paired benchmark/parity green, default hybrid 8-request smoke green, and codedeebee@0.2.5850 package dry-run green.