mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-05-29 06:31:32 +00:00
2d422720b5
Codex / Responses-API requests had three latent timeout bugs that combined into the long silent hangs reported on #21444: 1. The non-stream stale-call detector estimated context tokens from ``api_kwargs["messages"]`` only. Codex / Responses-API payloads carry their conversational load in ``input`` (with ``instructions`` and ``tools``), so every Codex turn logged ``context=~0 tokens`` and the detector never applied its >50k / >100k tier bumps. 2. ``providers.<id>.request_timeout_seconds`` was silently dropped on the main Codex path. The chat_completions path and the auxiliary Codex adapter both forwarded it; the main path skipped it through three places (``build_api_kwargs``, ``ResponsesApiTransport.build_kwargs``, ``_preflight_codex_api_kwargs``). 3. The streaming stale detector had the same payload-shape bug for ``codex_responses`` requests, which route through the non-streaming detector (it's the path that emits the user-facing "No response from provider for 300s (non-streaming, ...)" warning that reporters keep pasting). This commit: - Adds ``estimate_request_context_tokens`` in ``chat_completion_helpers``, used by both the non-stream and stream detectors. Handles ``messages`` (Chat Completions), ``input + instructions + tools`` (Responses API), bare lists, and an unknown-dict fallback. - Forwards ``timeout`` through ``ResponsesApiTransport.build_kwargs`` and ``_preflight_codex_api_kwargs`` (with guards against zero/negative/inf/bool values), and wires ``_resolved_api_call_timeout()`` into the Codex branch of ``build_api_kwargs``. - Lowers the implicit non-stream stale defaults so fallback providers kick in faster when upstream stalls: * base 300s -> 90s * >50k 450s -> 150s * >100k 600s -> 240s These only apply when the user has *not* set ``providers.<id>.stale_timeout_seconds`` or ``HERMES_API_CALL_STALE_TIMEOUT``. Explicit config still wins. - Adds regression tests for the estimator shapes, the new defaults, the context-tier scaling, transport timeout pass-through, and preflight timeout pass-through / rejection of invalid values. Closes #21444 Supersedes #21652 #24126 #31855 Co-authored-by: Hoang V. Pham <26063003+hehehe0803@users.noreply.github.com>
515 lines
20 KiB
Python
515 lines
20 KiB
Python
"""Tests for the ResponsesApiTransport (Codex)."""
|
|
|
|
import json
|
|
import pytest
|
|
from types import SimpleNamespace
|
|
|
|
from agent.transports import get_transport
|
|
from agent.transports.types import NormalizedResponse, ToolCall
|
|
|
|
|
|
@pytest.fixture
|
|
def transport():
|
|
import agent.transports.codex # noqa: F401
|
|
return get_transport("codex_responses")
|
|
|
|
|
|
class TestCodexTransportBasic:
|
|
|
|
def test_api_mode(self, transport):
|
|
assert transport.api_mode == "codex_responses"
|
|
|
|
def test_registered_on_import(self, transport):
|
|
assert transport is not None
|
|
|
|
def test_convert_tools(self, transport):
|
|
tools = [{
|
|
"type": "function",
|
|
"function": {
|
|
"name": "terminal",
|
|
"description": "Run a command",
|
|
"parameters": {"type": "object", "properties": {"command": {"type": "string"}}},
|
|
}
|
|
}]
|
|
result = transport.convert_tools(tools)
|
|
assert len(result) == 1
|
|
assert result[0]["type"] == "function"
|
|
assert result[0]["name"] == "terminal"
|
|
|
|
|
|
class TestCodexBuildKwargs:
|
|
|
|
def test_basic_kwargs(self, transport):
|
|
messages = [
|
|
{"role": "system", "content": "You are helpful."},
|
|
{"role": "user", "content": "Hello"},
|
|
]
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.4",
|
|
messages=messages,
|
|
tools=[],
|
|
)
|
|
assert kw["model"] == "gpt-5.4"
|
|
assert kw["instructions"] == "You are helpful."
|
|
assert "input" in kw
|
|
assert kw["store"] is False
|
|
|
|
def test_system_extracted_from_messages(self, transport):
|
|
messages = [
|
|
{"role": "system", "content": "Custom system prompt"},
|
|
{"role": "user", "content": "Hi"},
|
|
]
|
|
kw = transport.build_kwargs(model="gpt-5.4", messages=messages, tools=[])
|
|
assert kw["instructions"] == "Custom system prompt"
|
|
|
|
def test_no_system_uses_default(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(model="gpt-5.4", messages=messages, tools=[])
|
|
assert kw["instructions"] # should be non-empty default
|
|
|
|
def test_reasoning_config(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.4", messages=messages, tools=[],
|
|
reasoning_config={"effort": "high"},
|
|
)
|
|
assert kw.get("reasoning", {}).get("effort") == "high"
|
|
|
|
def test_reasoning_disabled(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.4", messages=messages, tools=[],
|
|
reasoning_config={"enabled": False},
|
|
)
|
|
assert "reasoning" not in kw or kw.get("include") == []
|
|
|
|
def test_session_id_sets_cache_key(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.4", messages=messages, tools=[],
|
|
session_id="test-session-123",
|
|
)
|
|
assert kw.get("prompt_cache_key") == "test-session-123"
|
|
|
|
def test_github_responses_no_cache_key(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.4", messages=messages, tools=[],
|
|
session_id="test-session",
|
|
is_github_responses=True,
|
|
)
|
|
assert "prompt_cache_key" not in kw
|
|
|
|
def test_xai_responses_sends_cache_key_via_extra_body(self, transport):
|
|
"""xAI's Responses API documents ``prompt_cache_key`` as the
|
|
body-level cache-routing key (the ``x-grok-conv-id`` header is
|
|
Chat-Completions-only). Passing it via ``extra_body`` is robust
|
|
against openai SDK builds whose ``Responses.stream()`` kwarg
|
|
signature ever drops the field — the body field still serializes
|
|
and reaches xAI either way. The ``x-grok-conv-id`` header is kept
|
|
as a belt-and-braces fallback so cache routing survives even
|
|
when the body field would be stripped by an intermediate proxy.
|
|
Ref: https://docs.x.ai/developers/advanced-api-usage/prompt-caching/maximizing-cache-hits
|
|
"""
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="grok-4.3", messages=messages, tools=[],
|
|
session_id="conv-xai-1",
|
|
is_xai_responses=True,
|
|
)
|
|
assert "prompt_cache_key" not in kw
|
|
assert kw.get("extra_body", {}).get("prompt_cache_key") == "conv-xai-1"
|
|
assert kw.get("extra_headers", {}).get("x-grok-conv-id") == "conv-xai-1"
|
|
|
|
def test_xai_responses_extra_body_preserves_caller_fields(self, transport):
|
|
"""When the caller already supplies ``extra_body`` (e.g. via
|
|
request_overrides), the xAI cache-key injection must merge into
|
|
the existing dict instead of overwriting it. Caller-supplied
|
|
``prompt_cache_key`` wins (setdefault semantics) so user overrides
|
|
aren't silently clobbered by the transport."""
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="grok-4.3", messages=messages, tools=[],
|
|
session_id="conv-xai-1",
|
|
is_xai_responses=True,
|
|
request_overrides={"extra_body": {"prompt_cache_key": "caller-override", "other_field": 42}},
|
|
)
|
|
eb = kw.get("extra_body", {})
|
|
assert eb.get("prompt_cache_key") == "caller-override"
|
|
assert eb.get("other_field") == 42
|
|
|
|
def test_max_tokens(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.4", messages=messages, tools=[],
|
|
max_tokens=4096,
|
|
)
|
|
assert kw.get("max_output_tokens") == 4096
|
|
|
|
def test_codex_backend_no_max_output_tokens(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.4", messages=messages, tools=[],
|
|
max_tokens=4096,
|
|
is_codex_backend=True,
|
|
)
|
|
assert "max_output_tokens" not in kw
|
|
|
|
def test_xai_headers(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="grok-3", messages=messages, tools=[],
|
|
session_id="conv-123",
|
|
is_xai_responses=True,
|
|
)
|
|
assert kw.get("extra_headers", {}).get("x-grok-conv-id") == "conv-123"
|
|
|
|
def test_xai_headers_preserve_request_override_headers(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="grok-3", messages=messages, tools=[],
|
|
session_id="conv-123",
|
|
is_xai_responses=True,
|
|
request_overrides={"extra_headers": {"X-Test": "1", "X-Trace": "abc"}},
|
|
)
|
|
assert kw.get("extra_headers") == {
|
|
"X-Test": "1",
|
|
"X-Trace": "abc",
|
|
"x-grok-conv-id": "conv-123",
|
|
}
|
|
|
|
def test_minimal_effort_clamped(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.4", messages=messages, tools=[],
|
|
reasoning_config={"effort": "minimal"},
|
|
)
|
|
# "minimal" should be clamped to "low"
|
|
assert kw.get("reasoning", {}).get("effort") == "low"
|
|
|
|
def test_xai_reasoning_effort_passed(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="grok-4.3", messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"effort": "high"},
|
|
)
|
|
# xAI Responses receives reasoning.effort on the allowlisted models.
|
|
assert kw.get("reasoning") == {"effort": "high"}
|
|
# As of May 2026 (post-revert of PR #26644) we DO request
|
|
# reasoning.encrypted_content back from xAI so we can replay it
|
|
# across turns for cross-turn coherence — xAI explicitly relies
|
|
# on this for their partnership integration. See
|
|
# tests/run_agent/test_codex_xai_oauth_recovery.py for the
|
|
# full history.
|
|
assert "reasoning.encrypted_content" in kw.get("include", [])
|
|
|
|
def test_xai_reasoning_disabled_no_reasoning_key(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="grok-4.3", messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"enabled": False},
|
|
)
|
|
# When reasoning is disabled, do not send the reasoning key at all
|
|
assert "reasoning" not in kw
|
|
|
|
def test_xai_minimal_effort_clamped(self, transport):
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="grok-4.3", messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"effort": "minimal"},
|
|
)
|
|
# "minimal" should be clamped to "low" for xAI as well
|
|
assert kw.get("reasoning", {}).get("effort") == "low"
|
|
|
|
# --- Grok reasoning-effort capability allowlist ---
|
|
# api.x.ai 400s with "Model X does not support parameter reasoningEffort"
|
|
# on grok-4 / grok-4-fast / grok-3 / grok-code-fast / grok-4.20-0309-*.
|
|
# Those models reason natively but don't expose the dial. The transport
|
|
# must omit the `reasoning` key for them. As of May 2026 we DO request
|
|
# ``reasoning.encrypted_content`` back from xAI on every model —
|
|
# see test_xai_reasoning_effort_passed for the rationale.
|
|
|
|
def test_xai_grok_4_omits_reasoning_effort(self, transport):
|
|
"""grok-4 / grok-4-0709 reject reasoning.effort with HTTP 400."""
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
for model in ("grok-4", "grok-4-0709"):
|
|
kw = transport.build_kwargs(
|
|
model=model, messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"effort": "high"},
|
|
)
|
|
assert "reasoning" not in kw, (
|
|
f"{model} must not receive a reasoning key (xAI rejects it)"
|
|
)
|
|
# Even without the effort dial we still ask xAI to echo back
|
|
# encrypted reasoning content so it can be replayed next turn.
|
|
assert "reasoning.encrypted_content" in kw.get("include", [])
|
|
|
|
def test_xai_grok_4_fast_omits_reasoning_effort(self, transport):
|
|
"""grok-4-fast and grok-4-1-fast variants reject reasoning.effort."""
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
for model in (
|
|
"grok-4-fast-reasoning",
|
|
"grok-4-fast-non-reasoning",
|
|
"grok-4-1-fast-reasoning",
|
|
"grok-4-1-fast-non-reasoning",
|
|
):
|
|
kw = transport.build_kwargs(
|
|
model=model, messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"effort": "low"},
|
|
)
|
|
assert "reasoning" not in kw, (
|
|
f"{model} must not receive a reasoning key (xAI rejects it)"
|
|
)
|
|
|
|
def test_xai_grok_3_non_mini_omits_reasoning_effort(self, transport):
|
|
"""Plain grok-3 rejects reasoning.effort — only grok-3-mini accepts it."""
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="grok-3", messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"effort": "medium"},
|
|
)
|
|
assert "reasoning" not in kw
|
|
|
|
def test_xai_grok_3_mini_keeps_reasoning_effort(self, transport):
|
|
"""grok-3-mini and -fast variants do accept the effort dial."""
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
for model in ("grok-3-mini", "grok-3-mini-fast"):
|
|
kw = transport.build_kwargs(
|
|
model=model, messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"effort": "high"},
|
|
)
|
|
assert kw.get("reasoning") == {"effort": "high"}
|
|
|
|
def test_xai_grok_4_20_0309_variants_omit_reasoning_effort(self, transport):
|
|
"""grok-4.20-0309-(non-)reasoning reject the effort dial.
|
|
|
|
Counterintuitively, only grok-4.20-multi-agent-0309 accepts it.
|
|
"""
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
for model in ("grok-4.20-0309-reasoning", "grok-4.20-0309-non-reasoning"):
|
|
kw = transport.build_kwargs(
|
|
model=model, messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"effort": "high"},
|
|
)
|
|
assert "reasoning" not in kw, f"{model} must not receive reasoning"
|
|
|
|
def test_xai_grok_4_20_multi_agent_keeps_reasoning_effort(self, transport):
|
|
"""grok-4.20-multi-agent-0309 is the one grok-4.20 variant that accepts effort."""
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="grok-4.20-multi-agent-0309", messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"effort": "low"},
|
|
)
|
|
assert kw.get("reasoning") == {"effort": "low"}
|
|
|
|
def test_xai_grok_code_fast_omits_reasoning_effort(self, transport):
|
|
"""grok-code-fast-1 rejects reasoning.effort."""
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
kw = transport.build_kwargs(
|
|
model="grok-code-fast-1", messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"effort": "high"},
|
|
)
|
|
assert "reasoning" not in kw
|
|
|
|
def test_xai_aggregator_prefix_stripped(self, transport):
|
|
"""`x-ai/grok-3-mini` (OpenRouter-style slug) still resolves correctly."""
|
|
messages = [{"role": "user", "content": "Hi"}]
|
|
# Effort-capable
|
|
kw = transport.build_kwargs(
|
|
model="x-ai/grok-3-mini", messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"effort": "high"},
|
|
)
|
|
assert kw.get("reasoning") == {"effort": "high"}
|
|
# Effort-incapable
|
|
kw = transport.build_kwargs(
|
|
model="x-ai/grok-4-0709", messages=messages, tools=[],
|
|
is_xai_responses=True,
|
|
reasoning_config={"effort": "high"},
|
|
)
|
|
assert "reasoning" not in kw
|
|
|
|
|
|
class TestCodexValidateResponse:
|
|
|
|
def test_none_response(self, transport):
|
|
assert transport.validate_response(None) is False
|
|
|
|
def test_empty_output(self, transport):
|
|
r = SimpleNamespace(output=[], output_text=None)
|
|
assert transport.validate_response(r) is False
|
|
|
|
def test_valid_output(self, transport):
|
|
r = SimpleNamespace(output=[{"type": "message", "content": []}])
|
|
assert transport.validate_response(r) is True
|
|
|
|
def test_output_text_fallback_not_valid(self, transport):
|
|
"""validate_response is strict — output_text doesn't make it valid.
|
|
The caller handles output_text fallback with diagnostic logging."""
|
|
r = SimpleNamespace(output=None, output_text="Some text")
|
|
assert transport.validate_response(r) is False
|
|
|
|
|
|
class TestCodexMapFinishReason:
|
|
|
|
def test_completed(self, transport):
|
|
assert transport.map_finish_reason("completed") == "stop"
|
|
|
|
def test_incomplete(self, transport):
|
|
assert transport.map_finish_reason("incomplete") == "length"
|
|
|
|
def test_failed(self, transport):
|
|
assert transport.map_finish_reason("failed") == "stop"
|
|
|
|
def test_unknown(self, transport):
|
|
assert transport.map_finish_reason("unknown_status") == "stop"
|
|
|
|
|
|
class TestCodexNormalizeResponse:
|
|
|
|
def test_text_response(self, transport):
|
|
"""Normalize a simple text Codex response."""
|
|
r = SimpleNamespace(
|
|
output=[
|
|
SimpleNamespace(
|
|
type="message",
|
|
role="assistant",
|
|
content=[SimpleNamespace(type="output_text", text="Hello world")],
|
|
status="completed",
|
|
),
|
|
],
|
|
status="completed",
|
|
incomplete_details=None,
|
|
usage=SimpleNamespace(input_tokens=10, output_tokens=5,
|
|
input_tokens_details=None, output_tokens_details=None),
|
|
)
|
|
nr = transport.normalize_response(r)
|
|
assert isinstance(nr, NormalizedResponse)
|
|
assert nr.content == "Hello world"
|
|
assert nr.finish_reason == "stop"
|
|
|
|
def test_message_items_preserved_in_provider_data(self, transport):
|
|
"""Codex assistant message item ids/phases must survive transport normalization."""
|
|
r = SimpleNamespace(
|
|
output=[
|
|
SimpleNamespace(
|
|
type="message",
|
|
role="assistant",
|
|
id="msg_abc",
|
|
phase="final_answer",
|
|
content=[SimpleNamespace(type="output_text", text="Hello world")],
|
|
status="completed",
|
|
),
|
|
],
|
|
status="completed",
|
|
incomplete_details=None,
|
|
usage=SimpleNamespace(input_tokens=10, output_tokens=5,
|
|
input_tokens_details=None, output_tokens_details=None),
|
|
)
|
|
nr = transport.normalize_response(r)
|
|
assert nr.codex_message_items == [
|
|
{
|
|
"type": "message",
|
|
"role": "assistant",
|
|
"status": "completed",
|
|
"content": [{"type": "output_text", "text": "Hello world"}],
|
|
"id": "msg_abc",
|
|
"phase": "final_answer",
|
|
}
|
|
]
|
|
|
|
def test_tool_call_response(self, transport):
|
|
"""Normalize a Codex response with tool calls."""
|
|
r = SimpleNamespace(
|
|
output=[
|
|
SimpleNamespace(
|
|
type="function_call",
|
|
call_id="call_abc123",
|
|
name="terminal",
|
|
arguments=json.dumps({"command": "ls"}),
|
|
id="fc_abc123",
|
|
status="completed",
|
|
),
|
|
],
|
|
status="completed",
|
|
incomplete_details=None,
|
|
usage=SimpleNamespace(input_tokens=10, output_tokens=20,
|
|
input_tokens_details=None, output_tokens_details=None),
|
|
)
|
|
nr = transport.normalize_response(r)
|
|
assert nr.finish_reason == "tool_calls"
|
|
assert len(nr.tool_calls) == 1
|
|
tc = nr.tool_calls[0]
|
|
assert tc.name == "terminal"
|
|
assert '"command"' in tc.arguments
|
|
|
|
|
|
|
|
class TestCodexTransportTimeout:
|
|
"""Forward per-request timeout from build_kwargs to the SDK kwargs."""
|
|
|
|
def test_positive_timeout_preserved(self, transport):
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.5",
|
|
messages=[{"role": "user", "content": "hi"}],
|
|
tools=[],
|
|
timeout=600.0,
|
|
)
|
|
assert kw.get("timeout") == 600.0
|
|
|
|
def test_zero_timeout_dropped(self, transport):
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.5",
|
|
messages=[{"role": "user", "content": "hi"}],
|
|
tools=[],
|
|
timeout=0,
|
|
)
|
|
assert "timeout" not in kw
|
|
|
|
def test_none_timeout_omitted(self, transport):
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.5",
|
|
messages=[{"role": "user", "content": "hi"}],
|
|
tools=[],
|
|
timeout=None,
|
|
)
|
|
assert "timeout" not in kw
|
|
|
|
def test_inf_timeout_dropped(self, transport):
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.5",
|
|
messages=[{"role": "user", "content": "hi"}],
|
|
tools=[],
|
|
timeout=float("inf"),
|
|
)
|
|
assert "timeout" not in kw
|
|
|
|
def test_bool_timeout_dropped(self, transport):
|
|
"""``True`` is technically int but must not survive — caller bug guard."""
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.5",
|
|
messages=[{"role": "user", "content": "hi"}],
|
|
tools=[],
|
|
timeout=True,
|
|
)
|
|
assert "timeout" not in kw
|
|
|
|
def test_request_overrides_can_supply_timeout(self, transport):
|
|
"""request_overrides["timeout"] is honored when no explicit kwarg passed."""
|
|
kw = transport.build_kwargs(
|
|
model="gpt-5.5",
|
|
messages=[{"role": "user", "content": "hi"}],
|
|
tools=[],
|
|
request_overrides={"timeout": 450.0},
|
|
)
|
|
assert kw.get("timeout") == 450.0
|