perf: reuse event loop in BatchSpanProcessor instead of creating one per export

[jy-tan]

Summary

BatchSpanProcessor previously called asyncio.new_event_loop() for every batch export when using async adapters. Creating and destroying event loops is expensive (involves allocating selectors, registering file descriptors, and other OS-level setup/teardown) and can show up as latency/jitter under load. This change creates a single long-lived event loop when the export thread starts and reuses it for all async exports.

Changes

• Add _thread_loop instance variable to store the long-lived event loop
• Create the event loop once at the start of _export_loop() instead of per-export
• Reuse the thread's event loop in _export_batch() for async adapter exports
• Properly clean up the loop in a finally block when the export thread exits
• Retain fallback to temporary loop creation for _force_flush() after shutdown (rare, one-time occurrence when the thread has already exited)
• Unit tests: 5 new tests in tests/unit/test_batch_processor.py

[cubic-dev-ai]

1 issue found across 1 file

Prompt for AI agents (all issues)

Check if these issues are valid — if so, understand the root cause of each and fix them.


<file name="drift/core/batch_processor.py">

<violation number="1" location="drift/core/batch_processor.py:203">
P2: Reusing the export thread’s event loop without verifying the current thread can break force_flush when stop() times out. In that case, _force_flush() runs on a different thread and will attempt to drive the export thread’s loop, causing a RuntimeError. Gate the reuse on running in the export thread (or always create a temporary loop off-thread).</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}