fix(pii): load models only in workers and recycle them before memory runs out

[waleedlatif1]

Summary

• uvicorn's supervisor imports the app before spawning workers, so the module-level engine build left the parent holding an idle extra copy of every spaCy model. Engines now build in the FastAPI lifespan, which only serving processes run
• With more than one worker, each exits after a jittered request count (PII_MAX_REQUESTS, default 4000 ± PII_MAX_REQUESTS_JITTER 2000, health checks included) and the supervisor respawns it, so per-worker memory growth can't run the task into its hard limit between deploys
• A single worker never recycles (no supervisor to respawn it); --timeout-graceful-shutdown 300 matches the ALB idle timeout so in-flight redactions finish

Type of Change

• Bug fix

Testing

• AST scan: no import-time use of the engine globals remains; uvicorn 0.49 source confirms run() loads the app in the supervisor and keep_subprocess_alive respawns any exited worker
• CMD shell expansion checked for PII_WORKERS = 8, 1, unset, and malformed (malformed still fails uvicorn arg parsing)
• Validate on staging after deploy: one "building analyzer" log per worker (not workers + 1), healthy targets, lower starting memory, and workers restarting without 5xx

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing (new tests pass the test-audit authoring gate)
• 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.

Project	Deployment	Actions	Updated
docs	Ready	Preview	Oct 15, 2026 1:29am UTC

[cubic-dev-ai]

All reported issues were addressed across 2 files

<sub>Reply with feedback, questions, or to request a fix.

Turn on auto-fix | Re-trigger cubic</sub>

[greptile-apps]

Confidence Score: 5/5

[High impact] The PR appears safe to merge; no actionable new issue was established.

Summary

Moves PII engine setup into FastAPI’s lifespan and adds a Docker entrypoint for worker recycling.

• Each PII worker builds its models when it starts serving.
• Multi-worker PII containers recycle workers after varied request counts.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Docker entrypoint] --> B{More than one worker?}
    B -->|Yes| C[Add request limit and jitter]
    B -->|No| D[Leave recycling disabled]
    C --> E[Start uvicorn]
    D --> E
    E --> F[Worker runs FastAPI lifespan]
    F --> G[Build engines and blank artifacts]
    G --> H[Serve requests]
    H --> I{Request limit reached?}
    I -->|Multiple workers only| J[Drain and exit]
    J --> K[Supervisor starts replacement]
    K --> F

Loading

_{Reviews (2) · Last reviewed commit: "improvement(pii): move the start command..." · Reviewed by Greptile}

[waleedlatif1]

@greptile

[waleedlatif1]

@cubic-dev-ai review this PR

[cubic-dev-ai]

@cubic-dev-ai review this PR

@waleedlatif1 I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

No issues found across 3 files

Confidence score: 5/5

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

_{Turn on auto-fix | Re-trigger cubic}