fix(build): land the batching commit's unpushed consumers - #681
fix(build): land the batching commit's unpushed consumers#681ScriptedAlchemy wants to merge 3 commits into
Conversation
Sizes the code-index worker pool from a per-worker resident budget and host headroom instead of raw core count, and reports the resulting plan (configured selection, effective workers, limiting reason) so callers can see why a width was chosen. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
perf(observation) landed an import of tracedecay_runtime_core::background_cpu, but neither the module file nor its declaration was added, so the pushed integration branch does not compile: every branch cut from it fails on an unresolved import. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
perf(observation) added an ObservationApplicationError variant and a HostAdmissionOutcome field without updating four consumers, and code_index_scheduler references four parallelism items whose file was never git added. The pushed branch therefore does not compile, so every branch cut from it fails before its own work is even built. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 537a041e5a
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".
| pub fn install_process_background_cpu( | ||
| width: NonZeroUsize, | ||
| ) -> Result<Arc<ProcessBackgroundCpuV1>, BackgroundCpuInstallErrorV1> { |
There was a problem hiding this comment.
A repo-wide search of this commit finds no production caller of either install_worker_plan or install_process_background_cpu, so this process-wide singleton remains unset. HostAdmissionFacade::application requires process_background_cpu() and otherwise returns background_cpu_unavailable, meaning every otherwise-supported production observation capture is rejected; migrate daemon startup to the new installer rather than leaving the authority unmounted.
AGENTS.md reference: AGENTS.md:L82-L84
Useful? React with 👍 / 👎.
| let pool = rayon::ThreadPoolBuilder::new() | ||
| .num_threads(indexing_worker_target(detected_cores())) |
There was a problem hiding this comment.
When a standalone or one-shot caller has no installed runtime, this fallback ignores both TRACEDECAY_INDEX_WORKERS and the new memory-safe plan and instead uses half the detected CPUs. On a 96-core host, even an explicit one-worker override therefore creates 48 parser/artifact workers—about 6 GiB under this module's own 128 MiB-per-worker model—reintroducing the swap/OOM risk the previous max-eight ceiling avoided; malformed overrides also bypass the new typed refusal.
AGENTS.md reference: AGENTS.md:L102-L104
Useful? React with 👍 / 👎.
|
Closing in favour of #688, which is the same fix minus one commit that cannot land on this base. #681 also carried #688 covers |
1 participant
Merge this first — it supersedes #676
cursor/simplify-pr421-hot-pathsdoes not compile.perf(observation): batch admission and preparationlanded changes whose consumers and one whole file were nevergit added, so the work exists only in one local working tree. Every branch cut from the pushed head fails before its own code is built — that is currently every open PR and every agent working against this branch.Four separate breakages, all the same root cause:
crates/tracedecay-runtime-core/src/background_cpu.rs(whole file +modline)observation.rsimportObservationApplicationError::BatchWorkerStoppedarmruntime/ingest/failure.rs:517HostAdmissionOutcome.recoveryfieldkiro.rs,cline_like.rs,ingest/tests.rs×2CodeIndexWorkerPlanInstallErrorV1,worker_reservation_bytes,installed_worker_status,with_background_cpu_permitsrc/daemon/code_index_scheduler.rsThe last one is a ~694-added/170-removed-line delta in
crates/tracedecay-code-index/src/parallelism.rs— the in-flight worker-planning work. It is carried here as its original commit rather than reconstructed.One judgement call, flagged for the author
BatchWorkerStoppedneeded a classification and none was written. It is classified retryable /Unavailablewith reason codeobservation_batch_worker_stopped: a worker that went away before reaching a verdict decided nothing about the payload, so re-running the same input can succeed. Classifying itpermanentwould silently drop observations on a transient worker fault. If the author intended different semantics, this is the line to change.Verification
cargo check -p tracedecay-sessions --all-targetsclean.Cargo.lockis updated because the carried commit adds dependencies.Two agents independently hit these and reported rather than repairing, which is how the scope got established rather than guessed.
🤖 Generated with Claude Code