titan/atlas-iac
3
0
Fork 0
Code Issues Pull Requests 9 Actions Packages Projects Releases Wiki Activity

ops: report preemption and terminating service pods

Browse Source
This commit is contained in:
jenkins 2026-10-04 08:07:44 -05:00
parent eb1eb3698a
commit d85642adc1
2 changed files with 23 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
scripts/ops/cluster_settle_check.py
View File

@ -25,6 +25,7 @@ QUERIES = {
"helm": ["helmreleases.helm.toolkit.fluxcd.io", "-A"],
"volumes": ["volumes.longhorn.io", "-n", "longhorn-system"],
"warnings": ["events", "-A", "--field-selector=type=Warning"],
"preemptions": ["events", "-A", "--field-selector=reason=Preempted"],
}
@ -73,6 +74,7 @@ def snapshot() -> dict:
"uid": p["metadata"]["uid"], "node": p["spec"].get("nodeName"),
"owner": [(o["kind"], o["name"]) for o in owners],
"phase": p["status"]["phase"], "ready": conditions(p).get("Ready"),
"deleting": p["metadata"].get("deletionTimestamp"),
"restarts": {s["name"]: s.get("restartCount", 0) for s in statuses},
"waiting": {s["name"]: s["state"]["waiting"].get("reason")
for s in statuses if "waiting" in s.get("state", {})},
@ -111,7 +113,8 @@ def snapshot() -> dict:
"last_observed": e.get("series", {}).get("lastObservedTime")
or e.get("lastTimestamp") or e.get("eventTime"),
}
for e in raw["warnings"]
# Scheduler preemption can be a Normal event despite interrupting work.
for e in [*raw["warnings"], *raw["preemptions"]]
}
return result
@ -130,8 +133,10 @@ def summarize(current: dict, previous: dict | None = None) -> dict:
or v.get("rollout_failed", False)
or (v.get("observed_generation") or 0) < v["generation"]},
"unready_pods_on_reachable_nodes": [n for n, v in current["pods"].items()
if v["ready"] != "True" and v["node"] not in offline],
"unready_pods_on_offline_nodes": sum(v["ready"] != "True" and v["node"] in offline
if (v["ready"] != "True" or v.get("deleting"))
and v["node"] not in offline],
"unready_pods_on_offline_nodes": sum(bool(v["ready"] != "True" or v.get("deleting"))
and v["node"] in offline
for v in current["pods"].values()),
"unhealthy_attached_volumes": {n: v for n, v in current["volumes"].items()
if v["state"] == "attached" and v["robustness"] != "healthy"},

15
testing/tests/test_cluster_settle_check.py
View File

@ -77,6 +77,15 @@ def test_serving_old_replica_does_not_hide_failed_rollout():
assert list(summarize(current)["unready_workloads"]) == ["deployments/app/service"]
def test_terminating_ready_pod_is_not_reported_available():
"""A preempted pod can retain Ready while its termination grace runs."""
current = baseline()
current['pods']['app/worker'] = {
'node': 'worker', 'ready': 'True', 'deleting': '2026-10-04T09:00:00Z',
}
assert summarize(current)['unready_pods_on_reachable_nodes'] == ['app/worker']
def test_snapshot_omits_content_and_short_lived_jobs(monkeypatch):
"""Only operational metadata leaves raw API objects; Jobs are not service churn."""
import json
@ -100,11 +109,17 @@ def test_snapshot_omits_content_and_short_lived_jobs(monkeypatch):
'metadata': {'uid': 'event'}, 'involvedObject': {'namespace': 'app', 'name': 'service'},
'reason': 'Unhealthy', 'count': 1, 'message': hidden,
}]
data['preemptions'] = [{
'metadata': {'uid': 'scheduler-event'},
'involvedObject': {'namespace': 'app', 'name': 'service'},
'type': 'Normal', 'reason': 'Preempted', 'count': 1, 'message': hidden,
}]
monkeypatch.setattr(checker, 'fetch', lambda item: (item[0], data[item[0]]))
current = checker.snapshot()
assert list(current['pods']) == ['app/service']
assert hidden not in json.dumps(current)
assert current['warnings']['event']['reason'] == 'Unhealthy'
assert current['warnings']['scheduler-event']['reason'] == 'Preempted'
def test_failed_api_read_cannot_become_a_partial_health_report(monkeypatch):