cleanup(jenkins): harden pvc cleanup with dry-run metrics

This commit is contained in:
Brad Stein 2026-04-12 12:28:50 -03:00
parent ad99a83a98
commit 2ff3686700
4 changed files with 713 additions and 201 deletions

View File

@ -389,6 +389,8 @@ def _startup() -> None:
"metis_k3s_token_sync_cron": settings.metis_k3s_token_sync_cron,
"platform_quality_suite_probe_cron": settings.platform_quality_suite_probe_cron,
"jenkins_workspace_cleanup_cron": settings.jenkins_workspace_cleanup_cron,
"jenkins_workspace_cleanup_dry_run": settings.jenkins_workspace_cleanup_dry_run,
"jenkins_workspace_cleanup_max_deletions_per_run": settings.jenkins_workspace_cleanup_max_deletions_per_run,
"vault_k8s_auth_cron": settings.vault_k8s_auth_cron,
"vault_oidc_cron": settings.vault_oidc_cron,
"comms_guest_name_cron": settings.comms_guest_name_cron,

View File

@ -4,6 +4,8 @@ from dataclasses import dataclass
from datetime import datetime, timedelta, timezone
from typing import Any
from prometheus_client import Counter, Gauge
from ..k8s.client import delete_json, get_json
from ..settings import settings
from ..utils.logging import get_logger
@ -12,6 +14,48 @@ from ..utils.logging import get_logger
logger = get_logger(__name__)
JENKINS_WORKSPACE_CLEANUP_RUNS_TOTAL = Counter(
"ariadne_jenkins_workspace_cleanup_runs_total",
"Jenkins workspace cleanup runs by status and mode",
["status", "mode"],
)
JENKINS_WORKSPACE_CLEANUP_OBJECTS_TOTAL = Counter(
"ariadne_jenkins_workspace_cleanup_objects_total",
"Jenkins workspace cleanup objects by kind, action, and mode",
["kind", "action", "mode"],
)
JENKINS_WORKSPACE_CLEANUP_LAST_RUN_TS = Gauge(
"ariadne_jenkins_workspace_cleanup_last_run_timestamp_seconds",
"Last Jenkins workspace cleanup run timestamp",
)
JENKINS_WORKSPACE_CLEANUP_LAST_SUCCESS_TS = Gauge(
"ariadne_jenkins_workspace_cleanup_last_success_timestamp_seconds",
"Last successful Jenkins workspace cleanup timestamp",
)
JENKINS_WORKSPACE_CLEANUP_LAST_FAILURE_TS = Gauge(
"ariadne_jenkins_workspace_cleanup_last_failure_timestamp_seconds",
"Last failed Jenkins workspace cleanup timestamp",
)
JENKINS_WORKSPACE_CLEANUP_LAST_DELETED = Gauge(
"ariadne_jenkins_workspace_cleanup_last_deleted_total",
"Last Jenkins workspace cleanup deleted object count",
["kind"],
)
JENKINS_WORKSPACE_CLEANUP_LAST_PLANNED = Gauge(
"ariadne_jenkins_workspace_cleanup_last_planned_total",
"Last Jenkins workspace cleanup planned object count",
["kind"],
)
JENKINS_WORKSPACE_CLEANUP_LAST_SKIPPED = Gauge(
"ariadne_jenkins_workspace_cleanup_last_skipped_total",
"Last Jenkins workspace cleanup skipped object count",
)
JENKINS_WORKSPACE_CLEANUP_LAST_FAILURES = Gauge(
"ariadne_jenkins_workspace_cleanup_last_failures_total",
"Last Jenkins workspace cleanup failure count",
)
@dataclass(frozen=True)
class JenkinsWorkspaceCleanupSummary:
"""Summarize one Jenkins workspace-storage cleanup pass.
@ -20,11 +64,33 @@ class JenkinsWorkspaceCleanupSummary:
Outputs: deterministic counters for operator logs and metrics.
"""
pvs_planned: int
pvcs_planned: int
volumes_planned: int
pvs_deleted: int
pvcs_deleted: int
volumes_deleted: int
skipped: int
failures: int
dry_run: bool
@property
def planned(self) -> int:
return self.pvs_planned + self.pvcs_planned + self.volumes_planned
@property
def deleted(self) -> int:
return self.pvs_deleted + self.pvcs_deleted + self.volumes_deleted
@dataclass(frozen=True)
class _CleanupCandidate:
name: str
kind: str
path: str
created_at: datetime | None
related_pvc: str | None = None
pv_name: str | None = None
def _parse_timestamp(raw: str) -> datetime | None:
@ -37,30 +103,44 @@ def _parse_timestamp(raw: str) -> datetime | None:
return None
def _created_at(metadata: dict[str, Any]) -> datetime | None:
raw = metadata.get("creationTimestamp")
if not isinstance(raw, str) or not raw:
return None
return _parse_timestamp(raw)
def _is_old_enough(metadata: dict[str, Any]) -> bool:
"""Return true when an object age exceeds the configured cleanup threshold."""
raw = metadata.get("creationTimestamp")
if not isinstance(raw, str) or not raw:
return True
created_at = _parse_timestamp(raw)
created_at = _created_at(metadata)
if created_at is None:
return True
return False
min_age = timedelta(hours=settings.jenkins_workspace_cleanup_min_age_hours)
return datetime.now(timezone.utc) - created_at >= min_age
def _is_deleting(metadata: dict[str, Any]) -> bool:
deletion_ts = metadata.get("deletionTimestamp")
return isinstance(deletion_ts, str) and bool(deletion_ts.strip())
def _is_workspace_name(name: Any) -> bool:
return isinstance(name, str) and name.startswith(settings.jenkins_workspace_pvc_prefix)
def _active_workspace_claims() -> set[str]:
"""Collect currently referenced Jenkins workspace PVC names from pods."""
namespace = settings.jenkins_workspace_namespace
prefix = settings.jenkins_workspace_pvc_prefix
payload = get_json(f"/api/v1/namespaces/{namespace}/pods")
items = payload.get("items") if isinstance(payload.get("items"), list) else []
active: set[str] = set()
for pod in items:
if not isinstance(pod, dict):
continue
metadata = pod.get("metadata") if isinstance(pod.get("metadata"), dict) else {}
annotations = metadata.get("annotations") if isinstance(metadata.get("annotations"), dict) else {}
spec = pod.get("spec") if isinstance(pod.get("spec"), dict) else {}
volumes = spec.get("volumes") if isinstance(spec.get("volumes"), list) else []
for volume in volumes:
@ -70,19 +150,21 @@ def _active_workspace_claims() -> set[str]:
if not isinstance(claim, dict):
continue
claim_name = claim.get("claimName")
if isinstance(claim_name, str) and claim_name.startswith(prefix):
if _is_workspace_name(claim_name):
active.add(claim_name)
claim_name = annotations.get("jenkins.io/workspace-pvc")
if _is_workspace_name(claim_name):
active.add(claim_name)
return active
def _workspace_pv_candidates(active_claims: set[str]) -> tuple[list[dict[str, Any]], set[str]]:
def _workspace_pv_candidates(active_claims: set[str]) -> tuple[list[_CleanupCandidate], set[str]]:
"""Find releasable Jenkins workspace PVs and keep a set of all PV names."""
namespace = settings.jenkins_workspace_namespace
prefix = settings.jenkins_workspace_pvc_prefix
payload = get_json("/api/v1/persistentvolumes")
items = payload.get("items") if isinstance(payload.get("items"), list) else []
candidates: list[dict[str, Any]] = []
candidates: list[_CleanupCandidate] = []
all_pv_names: set[str] = set()
for pv in items:
@ -101,7 +183,9 @@ def _workspace_pv_candidates(active_claims: set[str]) -> tuple[list[dict[str, An
phase = status.get("phase")
if claim_namespace != namespace:
continue
if not isinstance(claim_name, str) or not claim_name.startswith(prefix):
if not _is_workspace_name(claim_name):
continue
if _is_deleting(metadata):
continue
if claim_name in active_claims:
continue
@ -109,18 +193,27 @@ def _workspace_pv_candidates(active_claims: set[str]) -> tuple[list[dict[str, An
continue
if not _is_old_enough(metadata):
continue
candidates.append(pv)
if not isinstance(name, str) or not name:
continue
candidates.append(
_CleanupCandidate(
name=name,
kind="pv",
path=f"/api/v1/persistentvolumes/{name}",
created_at=_created_at(metadata),
related_pvc=claim_name if isinstance(claim_name, str) else None,
)
)
return candidates, all_pv_names
def _workspace_pvc_candidates(active_claims: set[str]) -> list[dict[str, Any]]:
def _workspace_pvc_candidates(active_claims: set[str]) -> list[_CleanupCandidate]:
"""Find stale Jenkins workspace PVCs that are not actively referenced."""
namespace = settings.jenkins_workspace_namespace
prefix = settings.jenkins_workspace_pvc_prefix
payload = get_json(f"/api/v1/namespaces/{namespace}/persistentvolumeclaims")
items = payload.get("items") if isinstance(payload.get("items"), list) else []
candidates: list[dict[str, Any]] = []
candidates: list[_CleanupCandidate] = []
for pvc in items:
if not isinstance(pvc, dict):
@ -129,7 +222,9 @@ def _workspace_pvc_candidates(active_claims: set[str]) -> list[dict[str, Any]]:
status = pvc.get("status") if isinstance(pvc.get("status"), dict) else {}
claim_name = metadata.get("name")
phase = status.get("phase")
if not isinstance(claim_name, str) or not claim_name.startswith(prefix):
if not _is_workspace_name(claim_name):
continue
if _is_deleting(metadata):
continue
if claim_name in active_claims:
continue
@ -137,134 +232,369 @@ def _workspace_pvc_candidates(active_claims: set[str]) -> list[dict[str, Any]]:
continue
if not _is_old_enough(metadata):
continue
candidates.append(pvc)
if not isinstance(claim_name, str) or not claim_name:
continue
candidates.append(
_CleanupCandidate(
name=claim_name,
kind="pvc",
path=f"/api/v1/namespaces/{namespace}/persistentvolumeclaims/{claim_name}",
created_at=_created_at(metadata),
)
)
return candidates
def _delete_stale_pvcs(stale_pvcs: list[dict[str, Any]], namespace: str) -> tuple[int, int, int]:
"""Delete stale workspace PVCs and return deleted/skipped/failure counters."""
deleted = 0
skipped = 0
failures = 0
for pvc in stale_pvcs:
metadata = pvc.get("metadata") if isinstance(pvc.get("metadata"), dict) else {}
claim_name = metadata.get("name")
if not isinstance(claim_name, str) or not claim_name:
skipped += 1
continue
try:
delete_json(f"/api/v1/namespaces/{namespace}/persistentvolumeclaims/{claim_name}")
deleted += 1
except Exception as exc:
failures += 1
logger.info(
"jenkins workspace pvc delete failed",
extra={"event": "jenkins_workspace_cleanup", "claim": claim_name, "detail": str(exc)},
)
return deleted, skipped, failures
def _delete_stale_pvs(stale_pvs: list[dict[str, Any]]) -> tuple[int, int, int, set[str]]:
"""Delete stale workspace PVs and return counters plus removed PV names."""
deleted = 0
skipped = 0
failures = 0
removed_pv_names: set[str] = set()
for pv in stale_pvs:
metadata = pv.get("metadata") if isinstance(pv.get("metadata"), dict) else {}
pv_name = metadata.get("name")
if not isinstance(pv_name, str) or not pv_name:
skipped += 1
continue
try:
delete_json(f"/api/v1/persistentvolumes/{pv_name}")
removed_pv_names.add(pv_name)
deleted += 1
except Exception as exc:
failures += 1
logger.info(
"jenkins workspace pv delete failed",
extra={"event": "jenkins_workspace_cleanup", "pv": pv_name, "detail": str(exc)},
)
return deleted, skipped, failures, removed_pv_names
def _volume_should_delete(
metadata: dict[str, Any],
removed_pv_names: set[str],
all_pv_names: set[str],
prefix: str,
) -> bool:
"""Return true when a Longhorn volume belongs to stale Jenkins workspace storage."""
name = metadata.get("name")
if not isinstance(name, str) or not name:
return False
if name in removed_pv_names:
return True
labels = metadata.get("labels") if isinstance(metadata.get("labels"), dict) else {}
pvc_name = labels.get("kubernetes.io/created-for/pvc/name")
return isinstance(pvc_name, str) and pvc_name.startswith(prefix) and name not in all_pv_names
def _delete_stale_longhorn_volumes(
removed_pv_names: set[str],
all_pv_names: set[str],
prefix: str,
) -> tuple[int, int, int]:
"""Delete stale Longhorn volumes tied to Jenkins workspace PVCs."""
deleted = 0
skipped = 0
failures = 0
def _workspace_longhorn_candidates(all_pv_names: set[str], removed_pv_names: set[str]) -> list[_CleanupCandidate]:
namespace = "longhorn-system"
payload = get_json("/apis/longhorn.io/v1beta2/namespaces/longhorn-system/volumes")
items = payload.get("items") if isinstance(payload.get("items"), list) else []
candidates: list[_CleanupCandidate] = []
for volume in items:
if not isinstance(volume, dict):
continue
metadata = volume.get("metadata") if isinstance(volume.get("metadata"), dict) else {}
status = volume.get("status") if isinstance(volume.get("status"), dict) else {}
spec = volume.get("spec") if isinstance(volume.get("spec"), dict) else {}
name = metadata.get("name")
if not isinstance(name, str) or not name:
skipped += 1
continue
if not _volume_should_delete(metadata, removed_pv_names, all_pv_names, prefix):
labels = metadata.get("labels") if isinstance(metadata.get("labels"), dict) else {}
pvc_name = labels.get("kubernetes.io/created-for/pvc/name")
robust_state = status.get("robustness")
state = status.get("state")
attached = status.get("isAttached")
frontend = spec.get("frontend")
should_delete = False
if name in removed_pv_names:
should_delete = True
elif _is_workspace_name(pvc_name) and name not in all_pv_names:
should_delete = True
if not should_delete:
continue
if _is_deleting(metadata):
continue
if not _is_old_enough(metadata):
continue
if state not in {None, "detached", "faulted", "unknown"}:
continue
if attached is True:
continue
if robust_state not in {None, "unknown", "faulted", "degraded"}:
continue
if frontend not in {None, "", "blockdev"}:
continue
candidates.append(
_CleanupCandidate(
name=name,
kind="longhorn_volume",
path=f"/apis/longhorn.io/v1beta2/namespaces/{namespace}/volumes/{name}",
created_at=_created_at(metadata),
pv_name=name,
)
)
return candidates
def _validate_cleanup_settings() -> tuple[str, str, bool, int]:
namespace = settings.jenkins_workspace_namespace
prefix = settings.jenkins_workspace_pvc_prefix.strip()
dry_run = settings.jenkins_workspace_cleanup_dry_run
max_deletions = settings.jenkins_workspace_cleanup_max_deletions_per_run
if not namespace.strip():
raise ValueError("jenkins workspace cleanup namespace is empty")
if not prefix:
raise ValueError("jenkins workspace cleanup pvc prefix is empty")
if settings.jenkins_workspace_cleanup_min_age_hours < 1.0:
raise ValueError("jenkins workspace cleanup min age must be >= 1 hour")
if max_deletions < 1:
raise ValueError("jenkins workspace cleanup max deletions must be >= 1")
return namespace, prefix, dry_run, max_deletions
def _planned_removed_pv_names_dry_run(
stale_pvcs: list[_CleanupCandidate],
stale_pvs: list[_CleanupCandidate],
max_deletions: int,
) -> set[str]:
remaining = max(max_deletions - len(stale_pvcs), 0)
if remaining == 0:
return set()
names = [candidate.name for candidate in stale_pvs if candidate.name]
return set(names[:remaining])
def _delete_candidates(
candidates: list[_CleanupCandidate],
*,
deletion_budget: int | None,
failure_log: str,
failure_field: str,
removed_pv_names: set[str] | None = None,
) -> tuple[int, int, int, int | None]:
deleted = 0
skipped = 0
failures = 0
budget = deletion_budget
for candidate in candidates:
if not candidate.name:
skipped += 1
continue
if budget is not None and budget <= 0:
skipped += 1
continue
if budget is not None:
budget -= 1
try:
delete_json(f"/apis/longhorn.io/v1beta2/namespaces/longhorn-system/volumes/{name}")
delete_json(candidate.path)
deleted += 1
if removed_pv_names is not None:
removed_pv_names.add(candidate.name)
except Exception as exc:
failures += 1
logger.info(
"jenkins workspace longhorn volume delete failed",
extra={"event": "jenkins_workspace_cleanup", "volume": name, "detail": str(exc)},
failure_log,
extra={"event": "jenkins_workspace_cleanup", failure_field: candidate.name, "detail": str(exc)},
)
return deleted, skipped, failures
return deleted, skipped, failures, budget
def _record_guard_cap(
*,
max_deletions: int,
stale_pvcs: list[_CleanupCandidate],
stale_pvs: list[_CleanupCandidate],
stale_volumes: list[_CleanupCandidate],
dry_run: bool,
) -> None:
planned_total = len(stale_pvcs) + len(stale_pvs) + len(stale_volumes)
if planned_total <= max_deletions:
return
logger.warning(
"jenkins workspace cleanup capped by max deletions guard",
extra={
"event": "jenkins_workspace_cleanup",
"status": "guard_capped",
"namespace": settings.jenkins_workspace_namespace,
"dry_run": dry_run,
"planned_total": planned_total,
"max_deletions": max_deletions,
"planned_pvs": len(stale_pvs),
"planned_pvcs": len(stale_pvcs),
"planned_volumes": len(stale_volumes),
},
)
def _dry_run_summary(
*,
namespace: str,
max_deletions: int,
stale_pvcs: list[_CleanupCandidate],
stale_pvs: list[_CleanupCandidate],
all_pv_names: set[str],
) -> JenkinsWorkspaceCleanupSummary:
simulated_removed = _planned_removed_pv_names_dry_run(stale_pvcs, stale_pvs, max_deletions)
stale_volumes = _workspace_longhorn_candidates(all_pv_names, simulated_removed)
_record_guard_cap(
max_deletions=max_deletions,
stale_pvcs=stale_pvcs,
stale_pvs=stale_pvs,
stale_volumes=stale_volumes,
dry_run=True,
)
logger.info(
"jenkins workspace cleanup dry-run enabled",
extra={
"event": "jenkins_workspace_cleanup",
"status": "dry_run",
"namespace": namespace,
"dry_run": True,
"planned_pvs": len(stale_pvs),
"planned_pvcs": len(stale_pvcs),
"planned_volumes": len(stale_volumes),
"max_deletions": max_deletions,
},
)
return JenkinsWorkspaceCleanupSummary(
pvs_planned=len(stale_pvs),
pvcs_planned=len(stale_pvcs),
volumes_planned=len(stale_volumes),
pvs_deleted=0,
pvcs_deleted=0,
volumes_deleted=0,
skipped=0,
failures=0,
dry_run=True,
)
def _delete_run_summary(
*,
namespace: str,
max_deletions: int,
stale_pvcs: list[_CleanupCandidate],
stale_pvs: list[_CleanupCandidate],
all_pv_names: set[str],
) -> JenkinsWorkspaceCleanupSummary:
removed_pv_names: set[str] = set()
deletion_budget: int | None = max_deletions
pvcs_deleted, pvc_skipped, pvc_failures, deletion_budget = _delete_candidates(
stale_pvcs,
deletion_budget=deletion_budget,
failure_log="jenkins workspace pvc delete failed",
failure_field="claim",
)
pvs_deleted, pv_skipped, pv_failures, deletion_budget = _delete_candidates(
stale_pvs,
deletion_budget=deletion_budget,
failure_log="jenkins workspace pv delete failed",
failure_field="pv",
removed_pv_names=removed_pv_names,
)
stale_volumes = _workspace_longhorn_candidates(all_pv_names, removed_pv_names)
_record_guard_cap(
max_deletions=max_deletions,
stale_pvcs=stale_pvcs,
stale_pvs=stale_pvs,
stale_volumes=stale_volumes,
dry_run=False,
)
volumes_deleted, volume_skipped, volume_failures, _ = _delete_candidates(
stale_volumes,
deletion_budget=deletion_budget,
failure_log="jenkins workspace longhorn volume delete failed",
failure_field="volume",
)
return JenkinsWorkspaceCleanupSummary(
pvs_planned=len(stale_pvs),
pvcs_planned=len(stale_pvcs),
volumes_planned=len(stale_volumes),
pvs_deleted=pvs_deleted,
pvcs_deleted=pvcs_deleted,
volumes_deleted=volumes_deleted,
skipped=pvc_skipped + pv_skipped + volume_skipped,
failures=pvc_failures + pv_failures + volume_failures,
dry_run=False,
)
def _record_metrics(summary: JenkinsWorkspaceCleanupSummary) -> None:
mode = "dry_run" if summary.dry_run else "delete"
status = "ok" if summary.failures == 0 else "error"
JENKINS_WORKSPACE_CLEANUP_RUNS_TOTAL.labels(status=status, mode=mode).inc()
if summary.failures:
JENKINS_WORKSPACE_CLEANUP_LAST_FAILURE_TS.set(datetime.now(timezone.utc).timestamp())
else:
JENKINS_WORKSPACE_CLEANUP_LAST_SUCCESS_TS.set(datetime.now(timezone.utc).timestamp())
JENKINS_WORKSPACE_CLEANUP_LAST_RUN_TS.set(datetime.now(timezone.utc).timestamp())
JENKINS_WORKSPACE_CLEANUP_LAST_DELETED.labels(kind="pvc").set(summary.pvcs_deleted)
JENKINS_WORKSPACE_CLEANUP_LAST_DELETED.labels(kind="pv").set(summary.pvs_deleted)
JENKINS_WORKSPACE_CLEANUP_LAST_DELETED.labels(kind="longhorn_volume").set(summary.volumes_deleted)
JENKINS_WORKSPACE_CLEANUP_LAST_PLANNED.labels(kind="pvc").set(summary.pvcs_planned)
JENKINS_WORKSPACE_CLEANUP_LAST_PLANNED.labels(kind="pv").set(summary.pvs_planned)
JENKINS_WORKSPACE_CLEANUP_LAST_PLANNED.labels(kind="longhorn_volume").set(summary.volumes_planned)
JENKINS_WORKSPACE_CLEANUP_LAST_SKIPPED.set(summary.skipped)
JENKINS_WORKSPACE_CLEANUP_LAST_FAILURES.set(summary.failures)
for kind, planned, deleted in (
("pvc", summary.pvcs_planned, summary.pvcs_deleted),
("pv", summary.pvs_planned, summary.pvs_deleted),
("longhorn_volume", summary.volumes_planned, summary.volumes_deleted),
):
if planned:
JENKINS_WORKSPACE_CLEANUP_OBJECTS_TOTAL.labels(kind=kind, action="planned", mode=mode).inc(planned)
if deleted:
JENKINS_WORKSPACE_CLEANUP_OBJECTS_TOTAL.labels(kind=kind, action="deleted", mode=mode).inc(deleted)
if summary.skipped:
JENKINS_WORKSPACE_CLEANUP_OBJECTS_TOTAL.labels(
kind="cleanup",
action="skipped",
mode=mode,
).inc(summary.skipped)
if summary.failures:
JENKINS_WORKSPACE_CLEANUP_OBJECTS_TOTAL.labels(
kind="cleanup",
action="failed",
mode=mode,
).inc(summary.failures)
def cleanup_jenkins_workspace_storage() -> JenkinsWorkspaceCleanupSummary:
"""Delete stale Jenkins workspace PVC/PV artifacts and orphan Longhorn volumes."""
namespace = settings.jenkins_workspace_namespace
prefix = settings.jenkins_workspace_pvc_prefix
active_claims = _active_workspace_claims()
stale_pvs, all_pv_names = _workspace_pv_candidates(active_claims)
stale_pvcs = _workspace_pvc_candidates(active_claims)
pvcs_deleted, pvc_skipped, pvc_failures = _delete_stale_pvcs(stale_pvcs, namespace)
pvs_deleted, pv_skipped, pv_failures, removed_pv_names = _delete_stale_pvs(stale_pvs)
volumes_deleted, volume_skipped, volume_failures = _delete_stale_longhorn_volumes(
removed_pv_names, all_pv_names, prefix
summary = JenkinsWorkspaceCleanupSummary(
pvs_planned=0,
pvcs_planned=0,
volumes_planned=0,
pvs_deleted=0,
pvcs_deleted=0,
volumes_deleted=0,
skipped=0,
failures=0,
dry_run=settings.jenkins_workspace_cleanup_dry_run,
)
skipped = pvc_skipped + pv_skipped + volume_skipped
failures = pvc_failures + pv_failures + volume_failures
return JenkinsWorkspaceCleanupSummary(
pvs_deleted=pvs_deleted,
pvcs_deleted=pvcs_deleted,
volumes_deleted=volumes_deleted,
skipped=skipped,
failures=failures,
try:
namespace, _prefix, dry_run, max_deletions = _validate_cleanup_settings()
active_claims = _active_workspace_claims()
stale_pvs, all_pv_names = _workspace_pv_candidates(active_claims)
stale_pvcs = _workspace_pvc_candidates(active_claims)
if dry_run:
summary = _dry_run_summary(
namespace=namespace,
max_deletions=max_deletions,
stale_pvcs=stale_pvcs,
stale_pvs=stale_pvs,
all_pv_names=all_pv_names,
)
else:
summary = _delete_run_summary(
namespace=namespace,
max_deletions=max_deletions,
stale_pvcs=stale_pvcs,
stale_pvs=stale_pvs,
all_pv_names=all_pv_names,
)
except Exception as exc:
logger.exception(
"jenkins workspace cleanup failed",
extra={
"event": "jenkins_workspace_cleanup",
"status": "error",
"namespace": settings.jenkins_workspace_namespace,
"detail": str(exc),
},
)
summary = JenkinsWorkspaceCleanupSummary(
pvs_planned=summary.pvs_planned,
pvcs_planned=summary.pvcs_planned,
volumes_planned=summary.volumes_planned,
pvs_deleted=summary.pvs_deleted,
pvcs_deleted=summary.pvcs_deleted,
volumes_deleted=summary.volumes_deleted,
skipped=summary.skipped,
failures=summary.failures + 1,
dry_run=summary.dry_run,
)
_record_metrics(summary)
raise
_record_metrics(summary)
logger.info(
"jenkins workspace cleanup finished",
extra={
"event": "jenkins_workspace_cleanup",
"status": "ok" if summary.failures == 0 else "error",
"dry_run": summary.dry_run,
"namespace": namespace,
"planned_pvs": summary.pvs_planned,
"planned_pvcs": summary.pvcs_planned,
"planned_volumes": summary.volumes_planned,
"deleted_pvs": summary.pvs_deleted,
"deleted_pvcs": summary.pvcs_deleted,
"deleted_volumes": summary.volumes_deleted,
"skipped": summary.skipped,
"failures": summary.failures,
},
)
return summary

View File

@ -171,6 +171,8 @@ class Settings:
jenkins_workspace_namespace: str
jenkins_workspace_pvc_prefix: str
jenkins_workspace_cleanup_min_age_hours: float
jenkins_workspace_cleanup_dry_run: bool
jenkins_workspace_cleanup_max_deletions_per_run: int
vaultwarden_namespace: str
vaultwarden_pod_label: str
@ -469,6 +471,11 @@ class Settings:
"jenkins_workspace_namespace": _env("JENKINS_WORKSPACE_NAMESPACE", "jenkins"),
"jenkins_workspace_pvc_prefix": _env("JENKINS_WORKSPACE_PVC_PREFIX", "pvc-workspace-"),
"jenkins_workspace_cleanup_min_age_hours": _env_float("JENKINS_WORKSPACE_CLEANUP_MIN_AGE_HOURS", 12.0),
"jenkins_workspace_cleanup_dry_run": _env_bool("JENKINS_WORKSPACE_CLEANUP_DRY_RUN", "false"),
"jenkins_workspace_cleanup_max_deletions_per_run": _env_int(
"JENKINS_WORKSPACE_CLEANUP_MAX_DELETIONS_PER_RUN",
20,
),
}
@classmethod

View File

@ -3,101 +3,228 @@ from __future__ import annotations
from datetime import datetime, timezone
import types
from prometheus_client import REGISTRY
from ariadne.services import jenkins_workspace_cleanup as cleanup_module
def test_cleanup_jenkins_workspace_storage(monkeypatch) -> None:
dummy_settings = types.SimpleNamespace(
def _metric_value(name: str, labels: dict[str, str]) -> float:
value = REGISTRY.get_sample_value(name, labels)
return float(value) if value is not None else 0.0
def _dummy_settings(*, dry_run: bool, max_deletions: int = 20) -> types.SimpleNamespace:
return types.SimpleNamespace(
jenkins_workspace_namespace="jenkins",
jenkins_workspace_pvc_prefix="pvc-workspace-",
jenkins_workspace_cleanup_min_age_hours=1.0,
jenkins_workspace_cleanup_dry_run=dry_run,
jenkins_workspace_cleanup_max_deletions_per_run=max_deletions,
)
monkeypatch.setattr(cleanup_module, "settings", dummy_settings)
def _fake_payloads(now_iso: str, old_iso: str) -> dict[str, dict[str, object]]:
return {
"/api/v1/namespaces/jenkins/pods": {
"items": [
{
"metadata": {
"annotations": {
"jenkins.io/workspace-pvc": "pvc-workspace-annotated-active",
}
},
"spec": {
"volumes": [
{"persistentVolumeClaim": {"claimName": "pvc-workspace-active"}},
]
},
}
]
},
"/api/v1/namespaces/jenkins/persistentvolumeclaims": {
"items": [
{
"metadata": {"name": "pvc-workspace-stale", "creationTimestamp": old_iso},
"status": {"phase": "Lost"},
},
{
"metadata": {"name": "pvc-workspace-active", "creationTimestamp": old_iso},
"status": {"phase": "Bound"},
},
{
"metadata": {"name": "pvc-workspace-annotated-active", "creationTimestamp": old_iso},
"status": {"phase": "Lost"},
},
{
"metadata": {"name": "pvc-workspace-fresh", "creationTimestamp": now_iso},
"status": {"phase": "Lost"},
},
{
"metadata": {
"name": "pvc-workspace-deleting",
"creationTimestamp": old_iso,
"deletionTimestamp": old_iso,
},
"status": {"phase": "Lost"},
},
]
},
"/api/v1/persistentvolumes": {
"items": [
{
"metadata": {"name": "pvc-old", "creationTimestamp": old_iso},
"status": {"phase": "Released"},
"spec": {"claimRef": {"namespace": "jenkins", "name": "pvc-workspace-stale"}},
},
{
"metadata": {"name": "pvc-active", "creationTimestamp": old_iso},
"status": {"phase": "Released"},
"spec": {"claimRef": {"namespace": "jenkins", "name": "pvc-workspace-active"}},
},
{
"metadata": {"name": "pvc-annotated", "creationTimestamp": old_iso},
"status": {"phase": "Released"},
"spec": {"claimRef": {"namespace": "jenkins", "name": "pvc-workspace-annotated-active"}},
},
{
"metadata": {"name": "pvc-fresh", "creationTimestamp": now_iso},
"status": {"phase": "Released"},
"spec": {"claimRef": {"namespace": "jenkins", "name": "pvc-workspace-fresh"}},
},
{
"metadata": {
"name": "pvc-deleting",
"creationTimestamp": old_iso,
"deletionTimestamp": old_iso,
},
"status": {"phase": "Released"},
"spec": {"claimRef": {"namespace": "jenkins", "name": "pvc-workspace-deleting"}},
},
]
},
"/apis/longhorn.io/v1beta2/namespaces/longhorn-system/volumes": {
"items": [
{"metadata": {"name": "pvc-old", "creationTimestamp": old_iso}},
{
"metadata": {
"name": "pvc-orphan",
"creationTimestamp": old_iso,
"labels": {
"kubernetes.io/created-for/pvc/name": "pvc-workspace-orphan",
},
}
},
{
"metadata": {
"name": "pvc-attached",
"creationTimestamp": old_iso,
"labels": {
"kubernetes.io/created-for/pvc/name": "pvc-workspace-annotated-active",
},
},
"status": {"state": "attached", "isAttached": True, "robustness": "healthy"},
"spec": {"frontend": "blockdev"},
},
{
"metadata": {
"name": "pvc-orphan-fresh",
"creationTimestamp": now_iso,
"labels": {
"kubernetes.io/created-for/pvc/name": "pvc-workspace-fresh",
},
}
},
{
"metadata": {
"name": "pvc-vol-deleting",
"creationTimestamp": old_iso,
"deletionTimestamp": old_iso,
"labels": {
"kubernetes.io/created-for/pvc/name": "pvc-workspace-orphan",
},
}
},
]
},
}
def test_cleanup_jenkins_workspace_storage_dry_run(monkeypatch) -> None:
monkeypatch.setattr(cleanup_module, "settings", _dummy_settings(dry_run=True))
now_iso = datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
old_iso = "2020-01-01T00:00:00Z"
payloads = _fake_payloads(now_iso, old_iso)
deleted_paths: list[str] = []
def fake_get_json(path: str):
if path == "/api/v1/namespaces/jenkins/pods":
return {
"items": [
{
"spec": {
"volumes": [
{"persistentVolumeClaim": {"claimName": "pvc-workspace-active"}},
]
}
}
]
}
if path == "/api/v1/namespaces/jenkins/persistentvolumeclaims":
return {
"items": [
{
"metadata": {"name": "pvc-workspace-stale", "creationTimestamp": old_iso},
"status": {"phase": "Lost"},
},
{
"metadata": {"name": "pvc-workspace-active", "creationTimestamp": old_iso},
"status": {"phase": "Bound"},
},
{
"metadata": {"name": "pvc-workspace-fresh", "creationTimestamp": now_iso},
"status": {"phase": "Lost"},
},
]
}
if path == "/api/v1/persistentvolumes":
return {
"items": [
{
"metadata": {"name": "pvc-old", "creationTimestamp": old_iso},
"status": {"phase": "Released"},
"spec": {"claimRef": {"namespace": "jenkins", "name": "pvc-workspace-stale"}},
},
{
"metadata": {"name": "pvc-active", "creationTimestamp": old_iso},
"status": {"phase": "Released"},
"spec": {"claimRef": {"namespace": "jenkins", "name": "pvc-workspace-active"}},
},
{
"metadata": {"name": "pvc-fresh", "creationTimestamp": now_iso},
"status": {"phase": "Released"},
"spec": {"claimRef": {"namespace": "jenkins", "name": "pvc-workspace-fresh"}},
},
]
}
if path == "/apis/longhorn.io/v1beta2/namespaces/longhorn-system/volumes":
return {
"items": [
{"metadata": {"name": "pvc-old", "creationTimestamp": old_iso}},
{
"metadata": {
"name": "pvc-orphan",
"creationTimestamp": old_iso,
"labels": {
"kubernetes.io/created-for/pvc/name": "pvc-workspace-orphan",
},
}
},
{
"metadata": {
"name": "pvc-orphan-fresh",
"creationTimestamp": now_iso,
"labels": {
"kubernetes.io/created-for/pvc/name": "pvc-workspace-fresh",
},
}
},
]
}
if path in payloads:
return payloads[path]
raise AssertionError(f"unexpected path: {path}")
def fake_delete_json(path: str):
deleted_paths.append(path)
return {"status": "Success"}
before_runs = _metric_value(
"ariadne_jenkins_workspace_cleanup_runs_total",
{"status": "ok", "mode": "dry_run"},
)
before_planned = _metric_value(
"ariadne_jenkins_workspace_cleanup_objects_total",
{"kind": "pvc", "action": "planned", "mode": "dry_run"},
)
monkeypatch.setattr(cleanup_module, "get_json", fake_get_json)
monkeypatch.setattr(cleanup_module, "delete_json", fake_delete_json)
summary = cleanup_module.cleanup_jenkins_workspace_storage()
assert summary.dry_run is True
assert summary.pvcs_planned == 1
assert summary.pvs_planned == 1
assert summary.volumes_planned == 2
assert summary.pvcs_deleted == 0
assert summary.pvs_deleted == 0
assert summary.volumes_deleted == 0
assert summary.failures == 0
assert deleted_paths == []
assert _metric_value(
"ariadne_jenkins_workspace_cleanup_runs_total",
{"status": "ok", "mode": "dry_run"},
) == before_runs + 1
assert _metric_value(
"ariadne_jenkins_workspace_cleanup_objects_total",
{"kind": "pvc", "action": "planned", "mode": "dry_run"},
) == before_planned + 1
def test_cleanup_jenkins_workspace_storage(monkeypatch) -> None:
monkeypatch.setattr(cleanup_module, "settings", _dummy_settings(dry_run=False))
now_iso = datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
old_iso = "2020-01-01T00:00:00Z"
deleted_paths: list[str] = []
payloads = _fake_payloads(now_iso, old_iso)
def fake_get_json(path: str):
if path in payloads:
return payloads[path]
raise AssertionError(f"unexpected path: {path}")
def fake_delete_json(path: str):
deleted_paths.append(path)
return {"status": "Success"}
before_runs = _metric_value(
"ariadne_jenkins_workspace_cleanup_runs_total",
{"status": "ok", "mode": "delete"},
)
before_deleted = _metric_value(
"ariadne_jenkins_workspace_cleanup_objects_total",
{"kind": "longhorn_volume", "action": "deleted", "mode": "delete"},
)
monkeypatch.setattr(cleanup_module, "get_json", fake_get_json)
monkeypatch.setattr(cleanup_module, "delete_json", fake_delete_json)
@ -111,15 +238,19 @@ def test_cleanup_jenkins_workspace_storage(monkeypatch) -> None:
assert "/api/v1/persistentvolumes/pvc-old" in deleted_paths
assert "/apis/longhorn.io/v1beta2/namespaces/longhorn-system/volumes/pvc-old" in deleted_paths
assert "/apis/longhorn.io/v1beta2/namespaces/longhorn-system/volumes/pvc-orphan" in deleted_paths
assert "/apis/longhorn.io/v1beta2/namespaces/longhorn-system/volumes/pvc-attached" not in deleted_paths
assert _metric_value(
"ariadne_jenkins_workspace_cleanup_runs_total",
{"status": "ok", "mode": "delete"},
) == before_runs + 1
assert _metric_value(
"ariadne_jenkins_workspace_cleanup_objects_total",
{"kind": "longhorn_volume", "action": "deleted", "mode": "delete"},
) == before_deleted + 2
def test_cleanup_jenkins_workspace_storage_failure(monkeypatch) -> None:
dummy_settings = types.SimpleNamespace(
jenkins_workspace_namespace="jenkins",
jenkins_workspace_pvc_prefix="pvc-workspace-",
jenkins_workspace_cleanup_min_age_hours=1.0,
)
monkeypatch.setattr(cleanup_module, "settings", dummy_settings)
monkeypatch.setattr(cleanup_module, "settings", _dummy_settings(dry_run=False))
def fake_get_json(path: str):
if path == "/api/v1/namespaces/jenkins/pods":
@ -142,9 +273,51 @@ def test_cleanup_jenkins_workspace_storage_failure(monkeypatch) -> None:
def fake_delete_json(_path: str):
raise RuntimeError("boom")
before_failures = _metric_value(
"ariadne_jenkins_workspace_cleanup_objects_total",
{"kind": "cleanup", "action": "failed", "mode": "delete"},
)
monkeypatch.setattr(cleanup_module, "get_json", fake_get_json)
monkeypatch.setattr(cleanup_module, "delete_json", fake_delete_json)
summary = cleanup_module.cleanup_jenkins_workspace_storage()
assert summary.failures == 1
assert summary.pvcs_deleted == 0
assert _metric_value(
"ariadne_jenkins_workspace_cleanup_objects_total",
{"kind": "cleanup", "action": "failed", "mode": "delete"},
) == before_failures + 1
def test_cleanup_jenkins_workspace_storage_guard_caps_mass_delete(monkeypatch) -> None:
monkeypatch.setattr(cleanup_module, "settings", _dummy_settings(dry_run=False, max_deletions=1))
now_iso = datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
old_iso = "2020-01-01T00:00:00Z"
payloads = _fake_payloads(now_iso, old_iso)
deleted_paths: list[str] = []
def fake_get_json(path: str):
if path in payloads:
return payloads[path]
raise AssertionError(f"unexpected path: {path}")
def fake_delete_json(path: str):
deleted_paths.append(path)
return {"status": "Success"}
monkeypatch.setattr(cleanup_module, "get_json", fake_get_json)
monkeypatch.setattr(cleanup_module, "delete_json", fake_delete_json)
summary = cleanup_module.cleanup_jenkins_workspace_storage()
assert summary.failures == 0
assert summary.pvcs_planned == 1
assert summary.pvs_planned == 1
assert summary.volumes_planned == 1
assert summary.pvcs_deleted == 1
assert summary.pvs_deleted == 0
assert summary.volumes_deleted == 0
assert summary.skipped == 2
assert deleted_paths == ["/api/v1/namespaces/jenkins/persistentvolumeclaims/pvc-workspace-stale"]