ariadne/tests/hermes_autotriage_harness.py
codex ca2324ab95
All checks were successful
Tests / Declarative: Post Actions passed: 1251
feat(hermes): reclaim exhausted workspace storage, and stop hung builds
Two real actions, both for failures that interrupted this project while it was
being built, and both backed by code Ariadne already had.

reclaim_workspace_storage closes a gap left open deliberately: 'no space left
on device' was excluded from the transient retry because a rebuild lands on
the same full volume and either fails identically or hides a capacity problem.
Reclaiming first makes the retry meaningful. The reclaim is the existing
scheduled cleanup, with its own deletion budget, so no new capability is
granted - it is only reachable from triage now. Its signature requires both a
storage marker and a workspace hint, because a full disk elsewhere in the
cluster is a different failure that reclaiming Jenkins workspaces would not
address.

Hung-build detection previously filed an issue and left the build running,
holding one of five Jenkins agent slots and starving every other job - the
actual harm. Ariadne now stops it as well, which is reversible: the job can
simply be built again.

The action executors move to their own module. They are the only code in
triage that changes anything outside Ariadne and should be reviewable as one
unit.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-06 20:55:35 -03:00

333 lines
12 KiB
Python

"""Shared fakes for the Hermes auto-triage orchestrator tests.
Not collected by pytest; imported by test_hermes_autotriage.py and
test_hermes_autotriage_retry.py so both drive the same fake Jenkins, storage,
and remediation executors.
"""
from __future__ import annotations
import json
from types import SimpleNamespace
from ariadne.services import hermes_autotriage as module
from ariadne.services.hermes_agent_client import HermesRunResult
JOB = "hermes-triage-demo"
INCIDENT_ID = f"{JOB}/12"
INFRA_MARKER = "temporary failure in name resolution"
class FakeStorage:
def __init__(self) -> None:
self.events: list[dict] = []
def record_event(self, event_type, detail) -> None: # type: ignore[no-untyped-def]
self.events.append({"event_type": event_type, "detail": detail})
def list_events(self, limit=200, event_type=None): # type: ignore[no-untyped-def]
rows = [
dict(row)
for row in reversed(self.events)
if event_type is None or row["event_type"] == event_type
]
return rows[:limit]
class FakeResponse:
def __init__(self, payload) -> None: # type: ignore[no-untyped-def]
self.payload = payload
def raise_for_status(self) -> None:
return None
def json(self): # type: ignore[no-untyped-def]
return self.payload
def _settings(**overrides) -> SimpleNamespace: # type: ignore[no-untyped-def]
values = {
"hermes_autotriage_enabled": True,
"hermes_autotriage_job_allowlist": [JOB],
"hermes_autoremediation_enabled": True,
"hermes_allowed_actions": ["repair_demo_fixture"],
"hermes_action_classifications": {
"known_demo_fixture_failure": "repair_demo_fixture",
"transient_infra_failure": "retry_transient_infra",
},
"hermes_parameterized_jobs": [JOB],
"hermes_min_confidence": 0.85,
"hermes_hung_build_minutes": 45.0,
"hermes_max_branches": 5,
"hermes_job_namespaces": {},
"hermes_max_actions_per_incident": 1,
"hermes_api_url": "http://hermes:8642",
"hermes_api_key": "key",
"hermes_run_timeout_seconds": 420.0,
"hermes_demo_namespace": "hermes-triage-demo",
"hermes_demo_fixture_configmap": "hermes-triage-demo-fixture",
"hermes_code_enabled": False,
"hermes_code_job": "hermes-code-demo",
"hermes_code_owner": "bstein",
"hermes_code_repo": "hermes-code-demo",
"hermes_code_base_branch": "master",
"hermes_code_candidate_path": "src/discount.py",
"hermes_code_allowed_prefixes": ["src/"],
"hermes_code_allowed_suffixes": [".py"],
"hermes_code_max_patch_bytes": 4000,
"hermes_code_max_changed_lines": 20,
"hermes_gitea_base_url": "https://scm.example",
"hermes_gitea_token": "gitea-token",
"hermes_issues_enabled": False,
"hermes_issue_repos": {},
"hermes_issue_dedupe_scope": "classification",
"hermes_issue_max_per_tick": 2,
"jenkins_base_url": "https://ci.example",
"jenkins_api_user": "user",
"jenkins_api_token": "token",
"jenkins_api_timeout_sec": 5.0,
}
values.update(overrides)
return SimpleNamespace(**values)
def _build(number: int, result, job: str = JOB, **overrides): # type: ignore[no-untyped-def]
payload = {
"number": number,
"result": result,
"building": False,
"timestamp": 1720000000000,
"duration": 60000,
"url": f"https://ci.example/job/{job}/{number}/",
}
payload.update(overrides)
return payload
def _model_output(**overrides) -> str: # type: ignore[no-untyped-def]
payload = {
"incident_id": INCIDENT_ID,
"classification": "known_demo_fixture_failure",
"confidence": 0.95,
"facts": [
{"statement": "fixture-state-check failed", "source": "jenkins", "reference": "console"}
],
"inferences": [],
"first_failed_gate": "fixture-state-check",
"requested_action": {"type": "run_ariadne_job", "id": "repair_demo_fixture"},
"human_required": False,
"reason": "known fixture failure",
}
payload.update(overrides)
return json.dumps(payload)
def _retry_output(job: str = JOB, build_number: int = 12, **overrides) -> str: # type: ignore[no-untyped-def]
payload = {
"incident_id": f"{job}/{build_number}",
"classification": "transient_infra_failure",
"confidence": 0.95,
"facts": [
{"statement": "pip install hit a DNS failure", "source": "jenkins", "reference": "console"}
],
"inferences": [],
"first_failed_gate": "dependencies",
"requested_action": {"type": "run_ariadne_job", "id": "retry_transient_infra"},
"human_required": False,
"reason": "transient name resolution failure on the agent",
}
payload.update(overrides)
return json.dumps(payload)
def _run(status: str = "completed", output=None, error=None) -> HermesRunResult: # type: ignore[no-untyped-def]
return HermesRunResult(
status=status,
output=output,
run_id="run-1",
session_id="sess-1",
error=error,
duration_seconds=1.5,
denied_approvals=0,
)
def _install_jenkins(monkeypatch, calls, last_build, exc) -> None: # type: ignore[no-untyped-def]
class FakeClient:
def __init__(self, **kwargs) -> None: # type: ignore[no-untyped-def]
calls["client_kwargs"] = kwargs
def __enter__(self): # type: ignore[no-untyped-def]
return self
def __exit__(self, *args) -> None: # type: ignore[no-untyped-def]
return None
def get(self, url, params=None): # type: ignore[no-untyped-def]
calls["gets"].append((url, params))
if exc is not None:
raise exc
return FakeResponse({"lastBuild": last_build})
monkeypatch.setattr(module.hermes_jenkins_client.httpx, "Client", FakeClient)
def _prepare( # type: ignore[no-untyped-def] # noqa: PLR0913
monkeypatch,
*,
cfg=None,
last_build=None,
jenkins_exc=None,
run=None,
repair=None,
rebuild=None,
retry=None,
signature=True,
infra=None,
storage=None,
):
storage = storage if storage is not None else FakeStorage()
calls: dict = {"gets": [], "triage": [], "repairs": [], "rebuilds": [], "retries": []}
resolved = cfg if cfg is not None else _settings()
monkeypatch.setattr(module, "settings", resolved)
# The Jenkins transport holds its own settings reference, so patching
# only the orchestrator leaves it reading the real configuration.
monkeypatch.setattr(module.hermes_jenkins_client, "settings", resolved)
# The action executors live in their own module with their own references,
# so patching only the orchestrator leaves them on the real settings and
# the real Jenkins client.
from ariadne.services import hermes_autotriage_actions as _actions
monkeypatch.setattr(_actions, "settings", resolved)
_install_jenkins(
monkeypatch, calls, last_build if last_build is not None else _build(12, "FAILURE"), jenkins_exc
)
monkeypatch.setattr(
module.hermes_evidence,
"collect_evidence",
lambda incident_id, job, build: {
"incident_id": incident_id,
"jenkins": {"job": job},
"log_evidence": {"records": []},
},
)
monkeypatch.setattr(
module.hermes_evidence, "evidence_has_signature", lambda bundle, incident_id: signature
)
monkeypatch.setattr(
module.hermes_infra_signals,
"has_transient_infra_signature",
lambda bundle: infra if infra is not None else (True, INFRA_MARKER),
)
def fake_run_triage(config, prompt): # type: ignore[no-untyped-def]
calls["triage"].append((config, prompt))
return run if run is not None else _run(output=_model_output())
def fake_execute_repair(repair_cfg, incident_id, build_number): # type: ignore[no-untyped-def]
calls["repairs"].append((repair_cfg, incident_id, build_number))
if repair is not None:
return repair
return {
"action": "configmap_patch",
"target": "hermes-triage-demo/hermes-triage-demo-fixture",
"succeeded": True,
"error": None,
}
def fake_trigger_rebuild(rebuild_cfg, job): # type: ignore[no-untyped-def]
calls["rebuilds"].append(job)
return rebuild if rebuild is not None else {"requested": True, "error": None}
def fake_retry_build(retry_cfg, job, parameterized): # type: ignore[no-untyped-def]
calls["retries"].append((job, parameterized))
return retry if retry is not None else {"requested": True, "error": None}
monkeypatch.setattr(module.hermes_agent_client, "run_triage", fake_run_triage)
monkeypatch.setattr(module.hermes_autotriage_repair, "execute_repair", fake_execute_repair)
monkeypatch.setattr(_actions.hermes_autotriage_repair, "execute_repair", fake_execute_repair)
monkeypatch.setattr(module.hermes_autotriage_repair, "trigger_rebuild", fake_trigger_rebuild)
monkeypatch.setattr(module.hermes_autotriage_repair, "retry_build", fake_retry_build)
return SimpleNamespace(storage=storage, calls=calls)
def _code_settings(**overrides) -> SimpleNamespace: # type: ignore[no-untyped-def]
values = {
"hermes_code_enabled": True,
"hermes_code_repos": f"{JOB}=bstein/{JOB}",
"hermes_issues_enabled": True,
"hermes_issue_repos": {JOB: ("bstein", JOB)},
}
values.update(overrides)
return _settings(**values)
def _pr_opened(**overrides) -> dict: # type: ignore[no-untyped-def]
payload = {
"status": "pr_opened",
"branch": "hermes-repair/12",
"pr_number": 8,
"url": "https://scm.example/pulls/8",
}
payload.update(overrides)
return payload
def _prepare_code(monkeypatch, *, result=None, error=None, cfg=None, **kwargs): # type: ignore[no-untyped-def]
env = _prepare(monkeypatch, cfg=cfg if cfg is not None else _code_settings(), **kwargs)
env.calls.update({"proposals": [], "gitea": [], "issues": []})
def fake_propose(storage, incident_id, job, build_number, bundle, hermes_cfg, code_cfg): # type: ignore[no-untyped-def] # noqa: PLR0913
env.calls["proposals"].append((incident_id, job, build_number, bundle, code_cfg))
if error is not None:
raise error
return result if result is not None else _pr_opened()
def fake_lookup(lookup_cfg): # type: ignore[no-untyped-def]
env.calls["gitea"].append(lookup_cfg)
return {"found": False, "error": None}
def fake_create(issue_cfg, context): # type: ignore[no-untyped-def]
env.calls["issues"].append(context)
return {"issue_number": 7, "url": "https://scm.example/issues/7", "error": None}
monkeypatch.setattr(module.hermes_code_flow, "propose_code_fix", fake_propose)
monkeypatch.setattr(module.hermes_code_flow.hermes_code_repair, "find_open_proposal", fake_lookup)
monkeypatch.setattr(
module.hermes_incident_issue,
"find_open_incident_issue",
lambda *args: {"found": False, "issue_number": None, "url": None, "error": None},
)
monkeypatch.setattr(module.hermes_incident_issue, "create_incident_issue", fake_create)
return env
def _events(storage: FakeStorage, event_type: str) -> list[dict]:
return [row["detail"] for row in storage.events if row["event_type"] == event_type]
def _statuses(storage: FakeStorage) -> list[str]:
return [detail["status"] for detail in _events(storage, module.INCIDENT_EVENT_TYPE)]
def _seed_incident( # type: ignore[no-untyped-def]
storage: FakeStorage, status: str, build_number: int = 12, as_json: bool = False, job: str = JOB
) -> None:
detail = {
"incident_id": f"{job}/{build_number}",
"job": job,
"build_number": build_number,
"status": status,
"phase": {},
}
storage.record_event(module.INCIDENT_EVENT_TYPE, json.dumps(detail) if as_json else detail)
def _gauge(build: str, status: str, job: str = JOB) -> float:
return module.HERMES_TRIAGE_INCIDENT.labels(jenkins_job=job, build=build, status=status)._value.get()
def _counter(action: str, result: str) -> float:
return module.HERMES_TRIAGE_ACTION_TOTAL.labels(action=action, result=result)._value.get()