320 lines
12 KiB
Python

from __future__ import annotations
from .common import *
from .nodes import *
from .k8s import *
def _node_pods_top(node_pods: list[dict[str, Any]], limit: int = 5) -> list[dict[str, Any]]:
output: list[dict[str, Any]] = []
for entry in node_pods[:limit]:
if not isinstance(entry, dict):
continue
output.append(
{
"node": entry.get("node"),
"pods_total": entry.get("pods_total"),
"pods_running": entry.get("pods_running"),
"namespaces_top": entry.get("namespaces_top") or [],
}
)
return output
def _record_pending_pod(
pending_oldest: list[dict[str, Any]],
info: dict[str, Any],
) -> bool:
age_hours = info.get("age_hours")
if age_hours is None:
return False
pending_oldest.append(info)
return age_hours >= _PENDING_15M_HOURS
def _update_pod_issue(
pod: dict[str, Any],
acc: dict[str, Any],
) -> None:
metadata = pod.get("metadata") if isinstance(pod.get("metadata"), dict) else {}
status = pod.get("status") if isinstance(pod.get("status"), dict) else {}
spec = pod.get("spec") if isinstance(pod.get("spec"), dict) else {}
namespace = metadata.get("namespace") if isinstance(metadata.get("namespace"), str) else ""
name = metadata.get("name") if isinstance(metadata.get("name"), str) else ""
created_at = (
metadata.get("creationTimestamp")
if isinstance(metadata.get("creationTimestamp"), str)
else ""
)
age_hours = _age_hours(created_at)
if not name or not namespace:
return
phase = status.get("phase") if isinstance(status.get("phase"), str) else ""
restarts = 0
waiting_reasons: list[str] = []
for container in status.get("containerStatuses") or []:
if not isinstance(container, dict):
continue
restarts += int(container.get("restartCount") or 0)
state = container.get("state") if isinstance(container.get("state"), dict) else {}
waiting = state.get("waiting") if isinstance(state.get("waiting"), dict) else {}
reason = waiting.get("reason")
if isinstance(reason, str) and reason:
waiting_reasons.append(reason)
acc["waiting_reasons"][reason] = acc["waiting_reasons"].get(reason, 0) + 1
phase_reason = status.get("reason")
if isinstance(phase_reason, str) and phase_reason:
acc["phase_reasons"][phase_reason] = acc["phase_reasons"].get(phase_reason, 0) + 1
if phase in acc["counts"]:
acc["counts"][phase] += 1
if phase in _PHASE_SEVERITY or restarts > 0:
acc["items"].append(
{
"namespace": namespace,
"pod": name,
"node": spec.get("nodeName") or "",
"phase": phase,
"reason": status.get("reason") or "",
"restarts": restarts,
"waiting_reasons": sorted(set(waiting_reasons)),
"created_at": created_at,
"age_hours": age_hours,
}
)
if phase == "Pending":
info = {
"namespace": namespace,
"pod": name,
"node": spec.get("nodeName") or "",
"age_hours": age_hours,
"reason": status.get("reason") or "",
}
if _record_pending_pod(acc["pending_oldest"], info):
acc["pending_over_15m"] += 1
def _summarize_pod_issues(payload: dict[str, Any]) -> dict[str, Any]:
acc = {
"items": [],
"counts": {key: 0 for key in _PHASE_SEVERITY},
"pending_oldest": [],
"pending_over_15m": 0,
"waiting_reasons": {},
"phase_reasons": {},
}
for pod in _items(payload):
if isinstance(pod, dict):
_update_pod_issue(pod, acc)
items = acc["items"]
items.sort(
key=lambda item: (
-_PHASE_SEVERITY.get(item.get("phase") or "", 0),
-(item.get("restarts") or 0),
item.get("namespace") or "",
item.get("pod") or "",
)
)
pending_oldest = acc["pending_oldest"]
pending_oldest.sort(key=lambda item: -(item.get("age_hours") or 0.0))
return {
"counts": acc["counts"],
"items": items[:20],
"pending_oldest": pending_oldest[:10],
"pending_over_15m": acc["pending_over_15m"],
"waiting_reasons": acc["waiting_reasons"],
"phase_reasons": acc["phase_reasons"],
}
def _summarize_jobs(payload: dict[str, Any]) -> dict[str, Any]:
totals = {"total": 0, "active": 0, "failed": 0, "succeeded": 0}
by_namespace: dict[str, dict[str, int]] = {}
failing: list[dict[str, Any]] = []
active_oldest: list[dict[str, Any]] = []
for job in _items(payload):
metadata = job.get("metadata") if isinstance(job.get("metadata"), dict) else {}
status = job.get("status") if isinstance(job.get("status"), dict) else {}
name = metadata.get("name") if isinstance(metadata.get("name"), str) else ""
namespace = metadata.get("namespace") if isinstance(metadata.get("namespace"), str) else ""
created_at = (
metadata.get("creationTimestamp")
if isinstance(metadata.get("creationTimestamp"), str)
else ""
)
if not name or not namespace:
continue
active = int(status.get("active") or 0)
failed = int(status.get("failed") or 0)
succeeded = int(status.get("succeeded") or 0)
totals["total"] += 1
totals["active"] += active
totals["failed"] += failed
totals["succeeded"] += succeeded
entry = by_namespace.setdefault(namespace, {"active": 0, "failed": 0, "succeeded": 0})
entry["active"] += active
entry["failed"] += failed
entry["succeeded"] += succeeded
age_hours = _age_hours(created_at)
if failed > 0:
failing.append(
{
"namespace": namespace,
"job": name,
"failed": failed,
"age_hours": age_hours,
}
)
if active > 0 and age_hours is not None:
active_oldest.append(
{
"namespace": namespace,
"job": name,
"active": active,
"age_hours": age_hours,
}
)
failing.sort(
key=lambda item: (
-(item.get("failed") or 0),
-(item.get("age_hours") or 0.0),
item.get("namespace") or "",
item.get("job") or "",
)
)
active_oldest.sort(key=lambda item: -(item.get("age_hours") or 0.0))
namespace_summary = [
{
"namespace": ns,
"active": stats.get("active", 0),
"failed": stats.get("failed", 0),
"succeeded": stats.get("succeeded", 0),
}
for ns, stats in by_namespace.items()
]
namespace_summary.sort(
key=lambda item: (
-(item.get("active") or 0),
-(item.get("failed") or 0),
item.get("namespace") or "",
)
)
return {
"totals": totals,
"by_namespace": namespace_summary[:20],
"failing": failing[:20],
"active_oldest": active_oldest[:20],
}
def _summarize_deployments(payload: dict[str, Any]) -> dict[str, Any]:
items = _items(payload)
unhealthy: list[dict[str, Any]] = []
for dep in items:
metadata = dep.get("metadata") if isinstance(dep.get("metadata"), dict) else {}
spec = dep.get("spec") if isinstance(dep.get("spec"), dict) else {}
status = dep.get("status") if isinstance(dep.get("status"), dict) else {}
name = metadata.get("name") if isinstance(metadata.get("name"), str) else ""
namespace = metadata.get("namespace") if isinstance(metadata.get("namespace"), str) else ""
desired = int(spec.get("replicas") or 0)
ready = int(status.get("readyReplicas") or 0)
available = int(status.get("availableReplicas") or 0)
updated = int(status.get("updatedReplicas") or 0)
if desired <= 0:
continue
if ready < desired or available < desired:
unhealthy.append(
{
"name": name,
"namespace": namespace,
"desired": desired,
"ready": ready,
"available": available,
"updated": updated,
}
)
unhealthy.sort(key=lambda item: (item.get("namespace") or "", item.get("name") or ""))
return {
"total": len(items),
"not_ready": len(unhealthy),
"items": unhealthy,
}
def _summarize_statefulsets(payload: dict[str, Any]) -> dict[str, Any]:
items = _items(payload)
unhealthy: list[dict[str, Any]] = []
for st in items:
metadata = st.get("metadata") if isinstance(st.get("metadata"), dict) else {}
spec = st.get("spec") if isinstance(st.get("spec"), dict) else {}
status = st.get("status") if isinstance(st.get("status"), dict) else {}
name = metadata.get("name") if isinstance(metadata.get("name"), str) else ""
namespace = metadata.get("namespace") if isinstance(metadata.get("namespace"), str) else ""
desired = int(spec.get("replicas") or 0)
ready = int(status.get("readyReplicas") or 0)
current = int(status.get("currentReplicas") or 0)
updated = int(status.get("updatedReplicas") or 0)
if desired <= 0:
continue
if ready < desired:
unhealthy.append(
{
"name": name,
"namespace": namespace,
"desired": desired,
"ready": ready,
"current": current,
"updated": updated,
}
)
unhealthy.sort(key=lambda item: (item.get("namespace") or "", item.get("name") or ""))
return {
"total": len(items),
"not_ready": len(unhealthy),
"items": unhealthy,
}
def _summarize_daemonsets(payload: dict[str, Any]) -> dict[str, Any]:
items = _items(payload)
unhealthy: list[dict[str, Any]] = []
for ds in items:
metadata = ds.get("metadata") if isinstance(ds.get("metadata"), dict) else {}
status = ds.get("status") if isinstance(ds.get("status"), dict) else {}
name = metadata.get("name") if isinstance(metadata.get("name"), str) else ""
namespace = metadata.get("namespace") if isinstance(metadata.get("namespace"), str) else ""
desired = int(status.get("desiredNumberScheduled") or 0)
ready = int(status.get("numberReady") or 0)
updated = int(status.get("updatedNumberScheduled") or 0)
if desired <= 0:
continue
if ready < desired:
unhealthy.append(
{
"name": name,
"namespace": namespace,
"desired": desired,
"ready": ready,
"updated": updated,
}
)
unhealthy.sort(key=lambda item: (item.get("namespace") or "", item.get("name") or ""))
return {
"total": len(items),
"not_ready": len(unhealthy),
"items": unhealthy,
}
def _summarize_workload_health(
deployments: dict[str, Any],
statefulsets: dict[str, Any],
daemonsets: dict[str, Any],
) -> dict[str, Any]:
return {
"deployments": deployments,
"statefulsets": statefulsets,
"daemonsets": daemonsets,
}
__all__ = [name for name in globals() if not name.startswith("__")]