fix(backup): 手动备份按服冷却、每服保留上限与独立保留期，容量驱逐不再删除回收世界的唯一副本

internal/api/k8sjobstatus.go

@@ -3,6 +3,8 @@ package api
 import (
 	"context"
 	"sort"
+	"strings"
+	"time"
 
 	batchv1 "k8s.io/api/batch/v1"
 	corev1 "k8s.io/api/core/v1"
@@ -53,9 +55,66 @@ func (k *K8sJobStatus) LatestJobs(ctx context.Context, serverName string) ([]Asy
 	if len(out) > 20 {
 		out = out[:20]
 	}
+	k.explainFailures(ctx, serverName, out)
 	return out, nil
 }
 
+// explainFailures replaces a failed Job's condition text ("Job has reached the
+// specified backoff limit") with the error its container exited on. The
+// executors set TerminationMessagePolicy FallbackToLogsOnError, so the
+// terminated state carries the tail of the log, whose last line is the
+// "felis backup: …" / "felis restore: …" error. The pods carry the Job's
+// labels, so one list covers every Job of the server; a pod already collected
+// by the Job TTL, or a list error, leaves the condition text in place.
+func (k *K8sJobStatus) explainFailures(ctx context.Context, serverName string, jobs []AsyncJob) {
+	failed := map[string]int{}
+	for i, j := range jobs {
+		if j.State == "failed" {
+			failed[j.Name] = i
+		}
+	}
+	if len(failed) == 0 {
+		return
+	}
+	var pods corev1.PodList
+	if err := k.c.List(ctx, &pods, client.InNamespace(k.namespace),
+		client.MatchingLabels{jobServerLabel: serverName}); err != nil {
+		return
+	}
+	newest := map[string]time.Time{}
+	for i := range pods.Items {
+		pod := &pods.Items[i]
+		idx, ok := failed[pod.Labels["job-name"]]
+		if !ok {
+			continue
+		}
+		msg := lastTerminationLine(pod)
+		if msg == "" || !pod.CreationTimestamp.Time.After(newest[pod.Labels["job-name"]]) {
+			continue
+		}
+		newest[pod.Labels["job-name"]] = pod.CreationTimestamp.Time
+		jobs[idx].Message = msg
+	}
+}
+
+// lastTerminationLine returns the last non-empty line of the pod's terminated
+// container message, capped for display.
+func lastTerminationLine(pod *corev1.Pod) string {
+	for _, cs := range pod.Status.ContainerStatuses {
+		t := cs.State.Terminated
+		if t == nil || t.ExitCode == 0 {
+			continue
+		}
+		lines := strings.Split(strings.TrimSpace(t.Message), "\n")
+		line := strings.TrimSpace(lines[len(lines)-1])
+		if len(line) > 400 {
+			line = line[:400] + "…"
+		}
+		return line
+	}
+	return ""
+}
+
 // jobToAsyncJob projects one Job onto its kind/state/message. Complete condition →
 // succeeded, Failed → failed with its reason (Job conditions carry the generic
 // "backoff limit exceeded" text; the pod log holds the underlying error), anything