Aggregator is not thread-safe — flush vs. ingest race

[AndresL230]

Summary

recost/_aggregator.py:127-132's flush() iterates self._buckets.values() while user threads concurrently call ingest() (via interceptor wrappers, which run on whatever thread issued the HTTP request). The flush is driven by a background threading.Timer. Concurrent dict mutation during iteration raises RuntimeError: dictionary changed size during iteration.

This is a real race under any non-trivial load. There is no lock anywhere in Aggregator.

Fix

Wrap ingest, flush, and would_overflow with a threading.RLock. Add a regression test that pumps N worker threads into ingest while the timer thread calls flush.

Files

• recost/_aggregator.py
• tests/test_aggregator.py

Priority

P0 — corrupts telemetry / crashes the timer thread under load.