Summary

Splits the span batch-size / linger tuning out of #394 (which is now scoped to end-only ingest) so each change ships and reverts independently.

Changes in span_queue.py:

• _DEFAULT_BATCH_SIZE 50 → 200 — coalesce more spans per upsert_batch HTTP call (one INSERT … ON CONFLICT server-side), amortizing the per-request round trip + per-statement parse/plan + index maintenance that dominates at high span volume. Kept well under the EGP backend's 1000-row cap.
• _DEFAULT_LINGER_MS 100 → 250 — let a batch fill (spans arrive a few ms apart) instead of shipping near-size-1 batches; bounded so worst-case ingest latency / in-flight loss window stays sub-second.
• batch_size is now env-tunable via AGENTEX_SPAN_QUEUE_BATCH_SIZE (arg > env > default), clamped to ≥1.

Tests: test_span_queue.py — batch_size resolution (explicit arg > env > default, min-clamp).

Why separate

#394 is the end-only ingest change (skip span-start upsert); this is independent batching throughput tuning. Splitting keeps each reviewable and revertable on its own.

🤖 Generated with Claude Code

Greptile Summary

This PR tunes the AsyncSpanQueue's default batch parameters and makes batch_size configurable at runtime via an environment variable. The implementation is straightforward and well-tested.

• _DEFAULT_BATCH_SIZE raised from 50 → 200 and _DEFAULT_LINGER_MS from 100 → 250, with detailed inline rationale for each value.
• batch_size in AsyncSpanQueue.__init__ changed from int = _DEFAULT_BATCH_SIZE to int | None = None, resolved via the new AGENTEX_SPAN_QUEUE_BATCH_SIZE env var (arg > env > default, clamped to ≥1) using the existing _read_int_env helper.
• Six new unit tests cover all resolution paths: default, explicit arg, env var, arg-beats-env, zero-clamp, and invalid-env fallback.

Confidence Score: 5/5

Safe to merge; the changes are bounded config tuning with no logic alterations to the drain loop or export path.

The batch size and linger defaults are raised within documented safe limits, the env-var resolution follows the exact same pattern already used for linger_ms, max_retries, concurrency, and max_size, and all resolution paths are covered by dedicated unit tests. No callers outside the test suite pass batch_size explicitly, so the signature change is non-breaking.

No files require special attention.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[AsyncSpanQueue.__init__] --> B{batch_size arg provided?}
    B -->|No - None| C[_read_int_env AGENTEX_SPAN_QUEUE_BATCH_SIZE default=200 min=1]
    C --> D{Env var set?}
    D -->|Yes valid int| E[max 1 int env]
    D -->|Yes invalid| F[log warning use 200]
    D -->|Not set| G[use 200]
    B -->|Yes explicit int| H[max 1 batch_size]
    E --> I[self._batch_size]
    F --> I
    G --> I
    H --> I

Reviews (2): Last reviewed commit: "perf(tracing): larger span batch (50→200..." | Re-trigger Greptile