[Nikhil Navakiran (nikhilNava)]

Summary

• Export canonical text_completion and generate_content inference spans through Agent365; retain existing identity requirements, unrelated-operation filtering, and legacy-name behavior.
• Continue durable replay for other identities after a retryable failure. Scan at most 100 records and attempt at most 10 per pass, releasing skipped/unattempted leases and preserving gate backoff, fatal-error handling, shutdown cleanup, and immediate draining of healthy backlogs.
• Add regressions for inference export, cross-tenant recovery, blocked prefixes, scan/attempt bounds, periodic backoff, shutdown, and real-file backlog replay.

Gate eviction is intentionally out of scope. The 100-record scan window is bounded, not an unbounded fairness guarantee.

Validation

Executed locally on Windows with Python 3.11, using this branch's src on PYTHONPATH:

• python -m pytest -q tests\a365 --ignore=tests\a365\integration: 711 passed, 21 skipped, 54 subtests passed.
• Final focused run of tests\a365\test_utils.py, test_exporter.py, test_replay_coordinator.py, and test_durable_restart.py: 140 passed, 2 skipped, 17 subtests passed.
• Confirmed the new non-chat and replay-isolation regressions fail before the corresponding production fixes.
• python -m black --check on all six modified Python files: passed.
• python -m pylint --output-format=parseable on both modified source files and all four modified test files (using the test allowances from tox.ini): passed, 10.00/10.
• python -m mypy --python-version 3.10 --show-error-codes --ignore-missing-imports on modified source files, plus --check-untyped-defs --follow-imports=silent on modified test files: passed.
• git diff --check: passed.

The full Agent365 run reports one pre-existing PytestUnhandledThreadExceptionWarning from A365EnrichingBatchSpanProcessor (cannot release un-acquired lock). Reproduced the identical warning on unchanged main: 704 passed, 21 skipped, 49 subtests passed. No unrelated processor changes are included.

Not run locally: other Python interpreters, Linux/macOS, or credential-dependent live-service integration tests.

[Copilot]

🟢 Approval recommended

No active review issues remain for the changes improving Agent365 inference-span export and durable replay isolation.

No open issues 🎉

What changed in this PR

This PR fixes two Agent365 export gaps: canonical non-chat inference spans are now accepted, and durable replay can continue processing healthy identities when another identity is retrying or blocked.

Changes:

• Accept text_completion and generate_content while preserving legacy-name filtering.
• Scan up to 100 queued records and attempt at most 10 deliveries per replay pass, releasing unattempted leases.
• Add regression coverage for identity isolation, replay limits, backoff, shutdown, and persisted backlogs.

No actionable findings. No validation commands were run during this review.

File	Description
tests/​a365/​test_utils.py	Tests canonical inference-operation filtering.
tests/​a365/​test_replay_coordinator.py	Tests cross-tenant recovery, replay bounds, backoff, and lease cleanup.
tests/​a365/​test_exporter.py	Verifies canonical operations export unchanged and legacy names remain filtered.
tests/​a365/​test_durable_restart.py	Tests multi-pass replay of a real-file backlog.
src/​microsoft/​opentelemetry/​a365/​core/​exporters/​utils.py	Adds canonical non-chat inference operations to accepted spans.
src/​microsoft/​opentelemetry/​a365/​core/​exporters/​replay_coordinator.py	Separates scan and attempt limits and continues past retryable failures.
CHANGELOG.md	Documents both Agent365 fixes and replay limits.

_{Review effort: Balanced}

[github-actions]

Performance comparison

Threshold: regressions >15.0% on gating scenarios fail the build. Higher ops/s is better; positive Δ means the PR is slower.

Scenario	Gating	Baseline (ops/s)	Candidate (ops/s)	Δ %	Status
azure_monitor_log	yes	28,198.4	28,119.1	+0.28%	✅
azure_monitor_span	yes	163,692.9	162,074.6	+1.00%	✅
otel_log	no	32,418.1	32,556.3	-0.42%	✅
otel_span	no	47,639.5	47,076.5	+1.20%	✅

[Copilot]

🟢 Approval recommended

No active review issues remain for the Agent365 export and durable replay changes.

No open issues 🎉

_{Review effort: Balanced}