Keep automatic GC requests local to each interpreter - #8902

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 9 remain after this review.

Walkthrough

Threaded automatic GC requests are now associated with individual interpreters and run through the VM bytecode-loop breaker. GC count decrements use a checked atomic update to avoid decrementing an already-zero count.

Changes

Garbage collection scheduling

Priority: ⬇️ Low

Estimated code review effort: 3 (Moderate) | ~20 minutes

Change: Bug fix

Sequence Diagram(s)

sequenceDiagram
  participant AllocatingInterpreter
  participant GcInterpreterState
  participant VirtualMachine
  participant GlobalCollectionLock
  AllocatingInterpreter->>GcInterpreterState: Schedule request at gen0 threshold
  VirtualMachine->>GcInterpreterState: Check collection_ready()
  GcInterpreterState->>GlobalCollectionLock: Check collector lock
  GcInterpreterState-->>VirtualMachine: Report readiness when unlocked
  VirtualMachine->>GcInterpreterState: Run collect(0)

Suggested reviewers: youknowone

Security Blast Radius

• inferred — Code able to drive allocations can trigger an automatic GC request for its interpreter. A resulting collection still coordinates with threads process-wide, while object selection remains ownership-filtered.

Trust Boundaries and Controls

• observed — The new dispatch checks the current VM's GC state. Existing enablement and collector-lock checks remain in the collection path; the two-interpreter test checks that servicing one VM does not consume the other's request.

Resilience and Maintainability Implications

• inferred — A pending request is retryable after collector contention, but automatic servicing requires a later bytecode safepoint on that interpreter. The reviewed evidence does not establish prompt finalization for an interpreter that executes no further bytecode.

Hardening Proposals

• proposed — For hosts that rely on prompt GC finalization, establish a resource-bound policy for interpreters that stop reaching bytecode safepoints.

• X
• Mastodon
• Reddit
• LinkedIn