cluster

func (*Reader) Events ¶

func (r *Reader) Events(ctx context.Context, namespace, objectName string, warnOnly bool) ([]providers.KubeEvent, error)

Events returns recent Events in a namespace (optionally for one object, optionally Warning-only), most-recent first. It is EventsSince with no time window — kept for the KubeReader interface and existing callers.

func (*Reader) EventsSince ¶ added in v0.9.0

func (r *Reader) EventsSince(ctx context.Context, namespace, objectName string, warnOnly bool, sinceMinutes int) ([]providers.KubeEvent, error)

EventsSince returns recent Events in a namespace, dropping any older than sinceMinutes (0 = no lower bound), most-recent first.

K2: the old code fetched a single Limit:200 page and only sorted newest-first AFTER fetching — so in a busy namespace the newest events could sit on a page never fetched. When a window is set we walk pages (bounded by eventMaxPages) and keep every in-window event, so the newest in-window events are actually returned.

func (*Reader) PodLogs ¶

func (r *Reader) PodLogs(ctx context.Context, q providers.PodLogQuery) (providers.LogResult, error)

PodLogs returns recent log lines from up to maxPods pods selected by q, bounded to the last q.SinceMinutes. When q.Previous is true it reads each pod's last-terminated container (the crash output of a CrashLoopBackOff) instead of the running one. Each line is prefixed with "<pod>/<container>". Best-effort: a pod whose log stream fails still emits a marker line so a tool error reads as missing data, not silence.

B6 (CORE-706): every container of a pod is read explicitly. The old code set no Container in PodLogOptions, and the Kubernetes API REJECTS a log request on a pod with more than one container ("a container name must be specified…"); that error was swallowed, so every pod with an istio/linkerd/cloudsql sidecar silently yielded nothing.

func (*Reader) PodStatuses ¶

func (r *Reader) PodStatuses(ctx context.Context, namespace, labelSelector string) ([]providers.PodStatus, error)

PodStatuses returns pod health in a namespace (optional label selector), with per-container waiting/terminated reasons — surfacing pod-level failures (CreateContainerConfigError, ImagePullBackOff, CrashLoopBackOff) that never reach logs because the container never started. Unhealthy pods sort first.

func (*Reader) WorkloadOwnership ¶ added in v0.9.0

func (r *Reader) WorkloadOwnership(ctx context.Context, namespace, labelSelector, podName string) (providers.OwnerChain, error)

WorkloadOwnership resolves the owner chain for the pods selected by (namespace, labelSelector) — Pod → ReplicaSet → Deployment (or StatefulSet/DaemonSet/Job) — then reads the top controller's Flux/ArgoCD tracking labels to name the owning GitOps object, and computes the generic last-applied-configuration drift signal on the top controller. It picks podName when set, else the first pod the selector returns. A selector matching no pods is an error (missing data), not a silent empty chain.