feat(hermes): triage the branches inside multibranch folders
All checks were successful
Tests / Declarative: Post Actions passed: 1192

A multibranch project is a folder, not a job: its /api/json carries no
lastBuild at all, only a jobs array with one child per branch. Detection read
lastBuild, so hermes-code-demo-branches was returned as skipped on every tick
since it was created and no branch could ever be triaged.

Expand a folder into its branch jobs and process each. Jenkins addresses them
as parent/job/branch, which the existing fetch already builds correctly, and
branch names stay URL-encoded because re-encoding them yields a 404.

Expansion is capped by ARIADNE_HERMES_MAX_BRANCHES (default 5): a repository
with many active branches would otherwise multiply the watched job count
without limit and could open one incident per failing branch in a single tick.
A name containing a slash is refused outright, since it would escape the
parent and address an unrelated job.

The Jenkins transport moves to its own module so the orchestrator reads as
decision logic, and jobs[name] rides along on the existing request rather than
costing a second call per job per tick.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
codex 2026-08-06 11:01:57 -03:00
parent bdd0254523
commit 281f06dccd
9 changed files with 267 additions and 43 deletions

View File

@ -11,6 +11,8 @@ from ..utils.logging import get_logger
from . import hermes_agent_client, hermes_autotriage_repair, hermes_code_flow, hermes_incident_issue
from . import hermes_triage_prompt
from . import hermes_hung_builds
from . import hermes_jenkins_client
from . import hermes_multibranch
from . import hermes_infra_signals
from . import hermes_autotriage_decision as hermes_decision
from . import hermes_autotriage_events as hermes_events
@ -36,7 +38,6 @@ RETRY_ACTION = "retry_transient_infra"
REBUILD_FAILED_REASON = "repair rebuild failed"
CODE_FIX_PROPOSED_REASON = "code_fix_proposed"
_LAST_BUILD_TREE = "lastBuild[number,result,building,timestamp,duration,url]"
_RUN_COMPLETED = "completed"
_UNKNOWN_ACTION_LABEL = "unknown"
@ -57,6 +58,8 @@ def run_hermes_autotriage(storage: Any) -> dict[str, Any]:
tick_state: dict[str, Any] = {}
for job in settings.hermes_autotriage_job_allowlist:
jobs[job] = _process_job(storage, job, incidents, tick_state)
for branch in jobs[job].get("branches") or []:
jobs[branch] = _process_job(storage, branch, incidents, tick_state)
HERMES_TRIAGE_DURATION_SECONDS.labels(phase="total").set(time.time() - started)
logger.info(
"hermes autotriage tick finished",
@ -70,7 +73,11 @@ def _process_job(
) -> dict[str, Any]:
"""Inspect one allowlisted job's last build and advance its incidents."""
last_build = _fetch_last_build(job)
payload = hermes_jenkins_client.fetch_job_payload(job)
if hermes_multibranch.is_folder(payload):
branches = hermes_multibranch.branch_job_paths(payload, job, settings.hermes_max_branches)
return {"status": "folder", "branches": branches}
last_build = hermes_jenkins_client.last_build(payload)
if last_build is None or last_build.get("number") is None:
return {"status": "skipped"}
if last_build.get("building"):
@ -109,41 +116,6 @@ def _handle_running_build(
return {"status": "human_required", "incident_id": incident_id, "reason": reason}
def _fetch_last_build(job: str) -> dict[str, Any] | None:
"""Fetch the job's lastBuild summary from Jenkins, or None on failure."""
base_url = settings.jenkins_base_url.strip().rstrip("/")
if not base_url:
return None
try:
with httpx.Client(**_jenkins_client_kwargs()) as client:
response = client.get(f"{base_url}/job/{job}/api/json", params={"tree": _LAST_BUILD_TREE})
response.raise_for_status()
payload = response.json()
except Exception as exc:
logger.info(
"hermes autotriage jenkins fetch failed",
extra={"event": "hermes_autotriage", "status": "jenkins_error", "job": job, "detail": str(exc)},
)
return None
last_build = payload.get("lastBuild") if isinstance(payload, dict) else None
return last_build if isinstance(last_build, dict) else None
def _jenkins_client_kwargs() -> dict[str, Any]:
"""Build httpx client kwargs with Ariadne's Jenkins read credential."""
kwargs: dict[str, Any] = {
"timeout": settings.jenkins_api_timeout_sec,
"follow_redirects": True,
}
username = settings.jenkins_api_user.strip()
token = settings.jenkins_api_token.strip()
if username and token:
kwargs["auth"] = (username, token)
return kwargs
def _resolve_on_success(
storage: Any, job: str, last_build: dict[str, Any], incidents: dict[str, dict[str, Any]]
) -> dict[str, Any]:

View File

@ -0,0 +1,78 @@
"""Read-only Jenkins access for the auto-triage tick.
Kept apart from the orchestrator so the HTTP concerns - credentials, the field
selector, and failure handling - stay in one place, and so the orchestrator
reads as decision logic rather than transport.
"""
from __future__ import annotations
from typing import Any
import httpx
from ..settings import settings
from ..utils.logging import get_logger
logger = get_logger(__name__)
# `jobs[name]` is requested alongside the build fields so a multibranch folder,
# which has no lastBuild of its own, can be recognised from the same call
# rather than costing a second request per job per tick.
JOB_TREE = "lastBuild[number,result,building,timestamp,duration,url],jobs[name]"
def base_url() -> str:
"""Return the configured Jenkins base URL without a trailing slash."""
return settings.jenkins_base_url.strip().rstrip("/")
def client_kwargs() -> dict[str, Any]:
"""Build httpx client kwargs carrying Ariadne's Jenkins read credential."""
kwargs: dict[str, Any] = {
"timeout": settings.jenkins_api_timeout_sec,
"follow_redirects": True,
}
username = settings.jenkins_api_user.strip()
token = settings.jenkins_api_token.strip()
if username and token:
kwargs["auth"] = (username, token)
return kwargs
def fetch_job_payload(job: str) -> dict[str, Any] | None:
"""Fetch one job's build summary, or None when Jenkins cannot be read.
Inputs: a job name, which may be a nested path such as `parent/job/branch`
because Jenkins addresses branch jobs that way. Outputs: the parsed
payload, or None on any transport, status, or parse failure - a tick must
survive Jenkins being briefly unavailable.
"""
root = base_url()
if not root:
return None
try:
with httpx.Client(**client_kwargs()) as client:
response = client.get(f"{root}/job/{job}/api/json", params={"tree": JOB_TREE})
response.raise_for_status()
payload = response.json()
except Exception as exc:
logger.info(
"hermes autotriage jenkins fetch failed",
extra={"event": "hermes_autotriage", "status": "jenkins_error", "job": job, "detail": str(exc)},
)
return None
return payload if isinstance(payload, dict) else None
def last_build(payload: Any) -> dict[str, Any] | None:
"""Return the lastBuild object from a job payload, if it has one."""
if not isinstance(payload, dict):
return None
build = payload.get("lastBuild")
return build if isinstance(build, dict) else None

View File

@ -0,0 +1,65 @@
"""Expand Jenkins multibranch folders into the branch jobs they contain.
A multibranch project is a folder, not a job: its `/api/json` carries no
`lastBuild` at all, only a `jobs` array with one child per branch. Detection
reads `lastBuild`, so such a job was silently skipped on every tick and no
branch could ever be triaged.
Expansion is capped rather than unbounded. A repository with many active
branches would otherwise multiply the watched job count without limit and
could open one incident per failing branch in a single tick.
"""
from __future__ import annotations
from typing import Any
FOLDER_CLASS_MARKERS = ("MultiBranchProject", "Folder")
DEFAULT_MAX_BRANCHES = 5
_JOBS_KEY = "jobs"
def is_folder(payload: Any) -> bool:
"""Report whether a job payload is a folder rather than a buildable job.
Inputs: the parsed `/api/json` payload for one job. Outputs: True when it
exposes child jobs and no `lastBuild` of its own. Both conditions are
required so a buildable job that merely reports children is never mistaken
for a folder. Never raises.
"""
if not isinstance(payload, dict):
return False
if payload.get("lastBuild") is not None:
return False
return isinstance(payload.get(_JOBS_KEY), list)
def branch_job_paths(payload: Any, parent: str, max_branches: int = DEFAULT_MAX_BRANCHES) -> list[str]:
"""Return the API paths of the branches inside a multibranch folder.
Inputs: the folder's `/api/json` payload, the parent job name, and the cap
on how many branches to expand. Outputs: paths shaped `parent/job/branch`,
which is what Jenkins' nested job URLs expect and what the existing
last-build fetch already builds correctly.
Branch names arrive URL-encoded (`hermes-repair%2F4`) and are passed
through untouched: re-encoding or decoding them yields a 404. Never
raises; returns [] for anything that is not a folder.
"""
if not is_folder(payload) or max_branches <= 0:
return []
paths: list[str] = []
for child in payload.get(_JOBS_KEY) or []:
if not isinstance(child, dict):
continue
name = str(child.get("name") or "").strip()
if not name or "/" in name:
continue
paths.append(f"{parent}/job/{name}")
if len(paths) >= max_branches:
break
return paths

View File

@ -190,6 +190,7 @@ class Settings:
hermes_parameterized_jobs: list[str]
hermes_min_confidence: float
hermes_hung_build_minutes: float
hermes_max_branches: int
hermes_max_actions_per_incident: int
hermes_api_url: str
hermes_api_key: str

View File

@ -290,6 +290,7 @@ def _hermes_autotriage_config() -> dict[str, Any]:
],
"hermes_min_confidence": _env_float("ARIADNE_HERMES_MIN_CONFIDENCE", 0.85),
"hermes_hung_build_minutes": _env_float("ARIADNE_HERMES_HUNG_BUILD_MINUTES", 45.0),
"hermes_max_branches": _env_int("ARIADNE_HERMES_MAX_BRANCHES", 5),
"hermes_max_actions_per_incident": _env_int("ARIADNE_HERMES_MAX_ACTIONS_PER_INCIDENT", 1),
"hermes_api_url": _env(
"ARIADNE_HERMES_API_URL",

View File

@ -59,6 +59,7 @@ def _settings(**overrides) -> SimpleNamespace: # type: ignore[no-untyped-def]
"hermes_parameterized_jobs": [JOB],
"hermes_min_confidence": 0.85,
"hermes_hung_build_minutes": 45.0,
"hermes_max_branches": 5,
"hermes_max_actions_per_incident": 1,
"hermes_api_url": "http://hermes:8642",
"hermes_api_key": "key",
@ -168,7 +169,7 @@ def _install_jenkins(monkeypatch, calls, last_build, exc) -> None: # type: igno
raise exc
return FakeResponse({"lastBuild": last_build})
monkeypatch.setattr(module.httpx, "Client", FakeClient)
monkeypatch.setattr(module.hermes_jenkins_client.httpx, "Client", FakeClient)
def _prepare( # type: ignore[no-untyped-def] # noqa: PLR0913
@ -187,7 +188,11 @@ def _prepare( # type: ignore[no-untyped-def] # noqa: PLR0913
):
storage = storage if storage is not None else FakeStorage()
calls: dict = {"gets": [], "triage": [], "repairs": [], "rebuilds": [], "retries": []}
monkeypatch.setattr(module, "settings", cfg if cfg is not None else _settings())
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)
_install_jenkins(
monkeypatch, calls, last_build if last_build is not None else _build(12, "FAILURE"), jenkins_exc
)

View File

@ -341,7 +341,11 @@ def test_jenkins_request_uses_basic_auth_and_tree(monkeypatch) -> None:
assert env.calls["client_kwargs"]["auth"] == ("user", "token")
url, params = env.calls["gets"][0]
assert url == f"https://ci.example/job/{JOB}/api/json"
assert params == {"tree": "lastBuild[number,result,building,timestamp,duration,url]"}
tree = params["tree"]
assert "lastBuild[number,result,building,timestamp,duration,url]" in tree
# jobs[name] rides along on the same request so a multibranch folder, which
# carries no lastBuild, is recognisable without a second call per tick.
assert "jobs[name]" in tree
def _issue_settings(**overrides): # type: ignore[no-untyped-def]

View File

@ -369,9 +369,16 @@ def _prepare_orchestrator(monkeypatch, flow_result, **setting_overrides): # typ
calls: list = []
monkeypatch.setattr(autotriage, "settings", _orchestrator_settings(**setting_overrides))
monkeypatch.setattr(
autotriage,
"_fetch_last_build",
lambda job: {"number": 7, "result": "FAILURE", "building": False, "url": "https://ci.example/7/"},
autotriage.hermes_jenkins_client,
"fetch_job_payload",
lambda job: {
"lastBuild": {
"number": 7,
"result": "FAILURE",
"building": False,
"url": "https://ci.example/7/",
}
},
)
monkeypatch.setattr(autotriage.hermes_evidence, "collect_evidence", lambda i, j, b: dict(BUNDLE))

View File

@ -0,0 +1,91 @@
"""Tests for expanding Jenkins multibranch folders into branch jobs."""
from __future__ import annotations
from ariadne.services import hermes_multibranch as module
def _folder(*names: str) -> dict:
return {"jobs": [{"name": n} for n in names]}
def test_a_folder_has_child_jobs_and_no_last_build() -> None:
assert module.is_folder(_folder("master")) is True
def test_a_buildable_job_is_never_treated_as_a_folder() -> None:
"""Both conditions are required, or a real job could be expanded away."""
assert module.is_folder({"lastBuild": {"number": 1}, "jobs": [{"name": "x"}]}) is False
assert module.is_folder({"lastBuild": {"number": 1}}) is False
assert module.is_folder({}) is False
assert module.is_folder(None) is False
assert module.is_folder("not-a-dict") is False
def test_branches_expand_to_nested_jenkins_paths() -> None:
"""Jenkins addresses a branch as parent/job/branch."""
paths = module.branch_job_paths(_folder("master", "develop"), "demo-branches")
assert paths == ["demo-branches/job/master", "demo-branches/job/develop"]
def test_encoded_branch_names_are_passed_through_untouched() -> None:
"""Re-encoding or decoding a branch name yields a 404."""
paths = module.branch_job_paths(_folder("hermes-repair%2F4"), "demo")
assert paths == ["demo/job/hermes-repair%2F4"]
def test_expansion_is_capped() -> None:
"""A busy repository must not multiply the watched job count without limit."""
folder = _folder(*[f"b{i}" for i in range(20)])
assert len(module.branch_job_paths(folder, "demo")) == module.DEFAULT_MAX_BRANCHES
assert len(module.branch_job_paths(folder, "demo", max_branches=2)) == 2
assert module.branch_job_paths(folder, "demo", max_branches=0) == []
assert module.branch_job_paths(folder, "demo", max_branches=-1) == []
def test_malformed_children_are_skipped_not_fatal() -> None:
folder = {"jobs": [None, {"name": ""}, {"name": " "}, {"nope": 1}, {"name": "ok"}]}
assert module.branch_job_paths(folder, "demo") == ["demo/job/ok"]
def test_a_name_containing_a_slash_is_refused() -> None:
"""A raw slash would escape the parent and address an unrelated job."""
assert module.branch_job_paths(_folder("a/b"), "demo") == []
def test_non_folders_expand_to_nothing() -> None:
assert module.branch_job_paths({"lastBuild": {"number": 3}}, "demo") == []
assert module.branch_job_paths(None, "demo") == []
def test_tick_expands_a_folder_into_its_branches(monkeypatch) -> None:
"""End to end: a multibranch job must stop being silently skipped."""
from ariadne.services import hermes_autotriage as autotriage
from tests.hermes_autotriage_harness import _prepare
env = _prepare(monkeypatch)
payloads = {
"hermes-triage-demo": {"jobs": [{"name": "master"}, {"name": "hermes-repair%2F4"}]},
"hermes-triage-demo/job/master": {
"lastBuild": {"number": 3, "result": "SUCCESS", "building": False, "url": "u"}
},
"hermes-triage-demo/job/hermes-repair%2F4": {
"lastBuild": {"number": 4, "result": "SUCCESS", "building": False, "url": "u"}
},
}
monkeypatch.setattr(
autotriage.hermes_jenkins_client, "fetch_job_payload", lambda job: payloads.get(job)
)
jobs = autotriage.run_hermes_autotriage(env.storage)["jobs"]
assert jobs["hermes-triage-demo"]["status"] == "folder"
assert jobs["hermes-triage-demo/job/master"]["status"] == "healthy"
assert jobs["hermes-triage-demo/job/hermes-repair%2F4"]["status"] == "healthy"