fix(streaming): route Anthropic stream cleanup to the Anthropic client, not OpenAI rebuild (#28161)

[teknium1]

Summary

Stream-cleanup on Anthropic-native sessions no longer rebuilds the OpenAI primary client — eliminating a ~15-minute hang on stuck streams.

Root cause: three stream-cleanup paths in agent/chat_completion_helpers.py called _replace_primary_openai_client() unconditionally. On api_mode == "anthropic_messages" (Anthropic-only configs, no OPENAI_API_KEY) that rebuild fails with a missing-credentials error AND the wedged in-flight httpx stream is never closed, so the worker thread blocks on the dead socket until the 900s httpx read-timeout fires. The _interrupt_requested branch in the same function already did the right thing for Anthropic (_anthropic_client.close() + _rebuild_anthropic_client()); the three cleanup sites now mirror it.

Changes

• agent/chat_completion_helpers.py: branch each of the three pool-cleanup sites on api_mode — Anthropic mode closes + rebuilds the Anthropic client; everything else rebuilds the OpenAI primary client as before:
  • stream_mid_tool_retry_pool_cleanup
  • stream_retry_pool_cleanup
  • stale_stream_pool_cleanup
• tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py: new tests for the two reachable Anthropic cleanup sites (stream-retry, stale-stream) asserting the Anthropic close+rebuild fires and the OpenAI rebuild does not.
• tests/run_agent/test_streaming.py: re-point test_anthropic_stream_parser_valueerror_retries_before_delivery — it previously asserted _replace_primary_openai_client was called once (passing because the bug was live). It now asserts the corrected close+rebuild-Anthropic behavior.

Validation

	Before	After
Anthropic stale/retry stream cleanup	rebuilds OpenAI client → fails, stream left open, ~15 min hang	closes + rebuilds Anthropic client, stream torn down immediately
_replace_primary_openai_client calls on Anthropic path	1	0
Streaming test suite	green while bug live	40/40 pass, assertions reflect fix

E2E (real imports, OPENAI_API_KEY unset, api_mode=anthropic_messages): a transient stream error triggers cleanup → _replace_primary_openai_client called 0 times, _rebuild_anthropic_client + _anthropic_client.close() each called once, retry succeeds.

Salvaged from #28240 by @EloquentBrush0x (earliest dedicated fix for this issue); conflict resolved against current main's newer request-client cleanup helper, sibling tests + the bug-encoding assertion fixed on top.

Infographic

[github-actions]

🔎 Lint report: hermes/hermes-f1139e66 vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 11581 on HEAD, 11524 on base (🆕 +57)

🆕 New issues (14):

Rule	Count
invalid-argument-type	11
unresolved-import	2
unresolved-attribute	1

First entries

tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:40: [invalid-argument-type] invalid-argument-type: Argument to `AIAgent.__init__` is incorrect: Expected `list[dict[str, Any]]`, found `str | bool`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:40: [invalid-argument-type] invalid-argument-type: Argument to `AIAgent.__init__` is incorrect: Expected `list[str] | None`, found `str | bool`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:40: [invalid-argument-type] invalid-argument-type: Argument to `AIAgent.__init__` is incorrect: Expected `int | float | None`, found `str | bool`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:19: [unresolved-import] unresolved-import: Cannot resolve imported module `httpx`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:40: [invalid-argument-type] invalid-argument-type: Argument to `AIAgent.__init__` is incorrect: Expected `list[str]`, found `str | bool`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:20: [unresolved-import] unresolved-import: Cannot resolve imported module `pytest`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:40: [invalid-argument-type] invalid-argument-type: Argument to `AIAgent.__init__` is incorrect: Expected `str`, found `str | bool`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:40: [invalid-argument-type] invalid-argument-type: Argument to `AIAgent.__init__` is incorrect: Expected `dict[str, Any]`, found `str | bool`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:40: [invalid-argument-type] invalid-argument-type: Argument to `AIAgent.__init__` is incorrect: Expected `bool`, found `str | bool`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:40: [invalid-argument-type] invalid-argument-type: Argument to `AIAgent.__init__` is incorrect: Expected `IterationBudget`, found `str | bool`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:41: [unresolved-attribute] unresolved-attribute: Unresolved attribute `api_mode` on type `AIAgent`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:40: [invalid-argument-type] invalid-argument-type: Argument to `AIAgent.__init__` is incorrect: Expected `int`, found `str | bool`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:40: [invalid-argument-type] invalid-argument-type: Argument to `AIAgent.__init__` is incorrect: Expected `((str, dict[Unknown, Unknown], /) -> None) | None`, found `str | bool`
tests/run_agent/test_28161_anthropic_stream_pool_cleanup.py:40: [invalid-argument-type] invalid-argument-type: Argument to `AIAgent.__init__` is incorrect: Expected `int | float`, found `str | bool`

✅ Fixed issues: none

Unchanged: 6069 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.

[alt-glitch]

This was generated by AI during triage.

Saturated competing cluster for #28161 (all open): #28240, #33855, #44076, #47797 — all fix the unconditional _replace_primary_openai_client() rebuild on api_mode==anthropic_messages stream-cleanup paths (the ~15-min hang). This maintainer PR is the authoritative three-site version. Flagging for a human to pick one and close the rest; not marking any as duplicate.