[F041] park grok exit-78 (auth missing/expired) instead of crash-retrying

A one-shot grok container whose entrypoint ran grok_auth --check and found
the token missing/expired exits 78 (EX_CONFIG). Crash-retrying 3x burns
tokens for zero progress — the agent cannot start without a valid token.
Park the provider with kind=auth_missing (same shape as the 429 exit-75
path) so the probe-resume loop revives the task once grok_auth.refresh_if_stale
mints a fresh token; if still expired, the next exit 78 re-parks (no burn).

Also fixes a latent F035 regression: _park_provider_unavailable now registers
a WaitingRecord, so the bare-__new__ rate-limit park test had to set
_waiting_records + stub _persist_waiting_record (mirrors the overload-test
fixture).
This commit is contained in:
Renn F
2026-06-28 11:15:52 +02:00
parent 003f8a3e26
commit adbb98b5c6
2 changed files with 132 additions and 6 deletions
+60 -6
View File
@@ -126,6 +126,8 @@ def _system_api_headers() -> dict[str, str]:
_SYSTEM_API_HEADERS["X-Agent-ID"], "system", ""
),
}
# Consecutive failed recovery probes before the CEO is notified once per episode.
_CEO_NOTIFY_THRESHOLD = 10
@@ -270,6 +272,17 @@ _GROK_INTERACTIVE_DOCKERFILES = {
# the retry window (unknown-provider time-expiry fallback in _probe_target).
_GROK_RATE_LIMIT_EXIT_CODE = 75
_GROK_RATE_LIMIT_RETRY_AFTER_S = 60.0
# A one-shot Grok container exits with this code (EX_CONFIG) when the
# entrypoint's `grok_auth --check` backstop found the access token missing or
# expired (it can't be refreshed headlessly, so the CLI would hang at an
# interactive login prompt). Park the provider instead of crash-retrying 3x —
# the agent cannot start without a valid token, so respawning burns tokens for
# zero progress. The probe-resume loop revives the task once
# grok_auth.refresh_if_stale (run once per dispatch tick) mints a fresh token
# from the offline-access refresh token; if still expired, the next exit 78
# re-parks (no token burn). Same shape as the 429 exit-75 path (F041).
_GROK_AUTH_EXIT_CODE = 78
_GROK_AUTH_RETRY_AFTER_S = 60.0
# =============================================================================
@@ -5599,6 +5612,14 @@ Start by:
if self._is_grok_rate_limit_exit(instance, exit_code):
await self._park_grok_rate_limited(agent_id, instance)
return
# Grok auth-missing parking (F041): a one-shot grok run whose entrypoint
# found the token missing/expired exits 78 (EX_CONFIG). Park the provider
# instead of crash-retrying — the agent can't start without a valid token,
# so respawning burns tokens for zero progress. The probe-resume loop
# revives the task once grok_auth.refresh_if_stale mints a fresh token.
if self._is_grok_auth_exit(instance, exit_code):
await self._park_grok_auth_unavailable(agent_id, instance)
return
graceful = exit_code == 0
# Session/usage-limit parking: the Claude session ("5-hour") limit is a
# 429 the SDK does not retry — the container exits non-zero with a
@@ -6376,6 +6397,23 @@ Start by:
and instance.config.provider_type == ModelProvider.GROK.value
)
@staticmethod
def _is_grok_auth_exit(instance: Any, exit_code: int | None) -> bool:
"""True for a one-shot grok container that exited 78 (auth missing/expired).
The entrypoint runs ``grok_auth --check`` as a backstop and exits 78
(EX_CONFIG) when the access token is missing or expired the CLI cannot
refresh it headlessly and would otherwise hang at an interactive login
prompt. See ``_GROK_AUTH_EXIT_CODE`` for the full rationale (F041).
"""
from roboco.models.base import ModelProvider
return (
exit_code == _GROK_AUTH_EXIT_CODE
and instance.config is not None
and instance.config.provider_type == ModelProvider.GROK.value
)
@staticmethod
async def _tail_container_logs(container_name: str, lines: int = 80) -> str:
"""Return the last ``lines`` of a container's combined output, '' on error.
@@ -6530,9 +6568,7 @@ Start by:
# call ``mark_waiting_long`` itself — the container is already dead (it
# exited), so there is nothing to stop, and parking keeps OFFLINE (not
# WAITING_LONG) so the reaper's live-skip / health loop ignore it.
task_id = (
str(instance.current_task_id) if instance.current_task_id else None
)
task_id = str(instance.current_task_id) if instance.current_task_id else None
record = WaitingRecord(
agent_id=agent_id,
task_id=task_id,
@@ -6563,6 +6599,26 @@ Start by:
kind="rate_limited",
)
async def _park_grok_auth_unavailable(self, agent_id: str, instance: Any) -> None:
"""Park a grok agent whose token was missing/expired (entrypoint exit 78).
Same park-and-probe shape as the 429 exit-75 path, but with
``kind="auth_missing"``: the agent cannot start without a valid token, so
crash-retrying burns tokens for zero progress. The probe-resume loop
revives the task once ``grok_auth.refresh_if_stale`` mints a fresh token
(run once per dispatch tick); if still expired, the next exit 78 re-parks
(no token burn). See ``_GROK_AUTH_EXIT_CODE`` (F041).
"""
from roboco.models.base import ModelProvider
await self._park_provider_unavailable(
agent_id,
instance,
provider=ModelProvider.GROK.value,
retry_after=_GROK_AUTH_RETRY_AFTER_S,
kind="auth_missing",
)
@staticmethod
def _too_early_to_probe(state: dict[str, Any]) -> bool:
"""True while the estimated lift time (activated_at + retry_after) is future.
@@ -7835,9 +7891,7 @@ Start now: evidence(task_id="{task_id}")
# ACTIVE; the reaper's Docker-liveness fallback covers it).
container_id: str | None = None
try:
container_id = await self._resolve_container_id(
f"roboco-agent-{slug}"
)
container_id = await self._resolve_container_id(f"roboco-agent-{slug}")
except Exception:
container_id = None
self._instances[slug] = AgentInstance(
@@ -15,6 +15,7 @@ from unittest.mock import AsyncMock
import pytest
from roboco.models.runtime import AgentInstance
from roboco.runtime.orchestrator import (
_GROK_AUTH_EXIT_CODE,
_GROK_RATE_LIMIT_EXIT_CODE,
AgentOrchestrator,
AgentState,
@@ -107,12 +108,18 @@ async def test_park_grok_rate_limited_activates_and_offlines(
monkeypatch: pytest.MonkeyPatch,
) -> None:
orch = AgentOrchestrator.__new__(AgentOrchestrator)
# _park_provider_unavailable registers a WaitingRecord (F035) so the
# probe-resume loop can revive the task; the bare __new__ orchestrator
# needs the dict + persist stub to exercise that without AttributeError.
orch._waiting_records = {}
orch._rate_limit_ceo_notified = set()
inst = _grok_instance()
inst.error_count = 2 # pretend prior crashes — parking must NOT count one
tracker = _FakeTracker()
monkeypatch.setattr(orch, "_make_tracker", lambda _p: tracker)
finalize = AsyncMock()
monkeypatch.setattr(orch, "_finalize_spawn_session", finalize)
monkeypatch.setattr(orch, "_persist_waiting_record", AsyncMock())
await orch._park_grok_rate_limited("be-dev-1", inst)
@@ -143,3 +150,68 @@ async def test_handle_stopped_container_parks_on_grok_429(
park.assert_awaited_once_with("be-dev-1", inst)
# Early-return: the normal crash/graceful finalize path never runs.
finalize.assert_not_awaited()
# ---------------------------------------------------------------------------
# F041: exit 78 (auth missing/expired) parks instead of crash-retrying
# ---------------------------------------------------------------------------
def test_is_grok_auth_exit() -> None:
inst = _grok_instance()
assert AgentOrchestrator._is_grok_auth_exit(inst, _GROK_AUTH_EXIT_CODE)
# Wrong exit code, or a non-grok provider, is not a grok-auth exit.
assert not AgentOrchestrator._is_grok_auth_exit(inst, 0)
assert not AgentOrchestrator._is_grok_auth_exit(inst, 1)
assert not AgentOrchestrator._is_grok_auth_exit(
_grok_instance(provider_type="anthropic"), _GROK_AUTH_EXIT_CODE
)
@pytest.mark.asyncio
async def test_handle_stopped_container_parks_on_grok_auth_exit(
monkeypatch: pytest.MonkeyPatch,
) -> None:
# F041: a grok container whose entrypoint ran `grok_auth --check` and found
# the token missing/expired exits 78 (EX_CONFIG). Crash-retrying 3x burns
# tokens for zero progress (the agent can't start without a valid token);
# park it like the 429 exit-75 path so the probe-resume loop revives the
# task once grok_auth.refresh_if_stale mints a fresh token.
orch = AgentOrchestrator.__new__(AgentOrchestrator)
inst = _grok_instance()
park = AsyncMock()
finalize = AsyncMock()
monkeypatch.setattr(orch, "_park_grok_auth_unavailable", park)
monkeypatch.setattr(orch, "_finalize_spawn_session", finalize)
await orch._handle_stopped_container("be-dev-1", inst, _GROK_AUTH_EXIT_CODE)
park.assert_awaited_once_with("be-dev-1", inst)
# Early-return: the crash-retry path never runs (no token burn).
finalize.assert_not_awaited()
@pytest.mark.asyncio
async def test_park_grok_auth_unavailable_activates_with_auth_missing_kind(
monkeypatch: pytest.MonkeyPatch,
) -> None:
orch = AgentOrchestrator.__new__(AgentOrchestrator)
orch._waiting_records = {}
orch._rate_limit_ceo_notified = set()
inst = _grok_instance()
inst.error_count = 2 # prior crashes — parking must NOT count one
tracker = _FakeTracker()
monkeypatch.setattr(orch, "_make_tracker", lambda _p: tracker)
monkeypatch.setattr(orch, "_finalize_spawn_session", AsyncMock())
monkeypatch.setattr(orch, "_persist_waiting_record", AsyncMock())
await orch._park_grok_auth_unavailable("be-dev-1", inst)
assert inst.state == AgentState.OFFLINE
assert inst.container_id is None
assert inst.error_count == 0 # an auth-missing exit is not a crash
assert tracker.activated_with == {
"retry_after": pytest.approx(60.0),
"affected_agents": ["be-dev-1"],
"kind": "auth_missing",
}