From 71d7a2d92cfc7ca45268ff3e3ca87361fc3a34f4 Mon Sep 17 00:00:00 2001
From: Patrick Ohly <patrick.ohly@intel.com>
Date: Wed, 3 Dec 2025 13:34:21 +0100
Subject: [PATCH] e2e podlogs: resume following logs based on last time stamp

If receiving the log output fails for whatever reason, it now uses the last
seen time stamp to continue where it got interrupted.
---
 test/e2e/storage/podlogs/podlogs.go | 52 +++++++++++++++++++++++------
 1 file changed, 42 insertions(+), 10 deletions(-)

diff --git a/test/e2e/storage/podlogs/podlogs.go b/test/e2e/storage/podlogs/podlogs.go
index 8ad723eca2d..7002bb9ef44 100644
--- a/test/e2e/storage/podlogs/podlogs.go
+++ b/test/e2e/storage/podlogs/podlogs.go
@@ -120,6 +120,8 @@ func CopyPodLogs(ctx context.Context, cs clientset.Interface, ns, podName string
 		active := map[string]bool{}
 		// Key is pod/container/container-id, true if we have ever started to capture its output.
 		started := map[string]bool{}
+		// Key is pod/container/container-id, value the time stamp of the last log line that has been seen.
+		latest := map[string]*meta.Time{}
 
 		check := func() {
 			m.Lock()
@@ -127,7 +129,7 @@ func CopyPodLogs(ctx context.Context, cs clientset.Interface, ns, podName string
 
 			pods, err := cs.CoreV1().Pods(ns).List(ctx, options)
 			if err != nil {
-				if to.StatusWriter != nil {
+				if ctx.Err() == nil && to.StatusWriter != nil {
 					_, _ = fmt.Fprintf(to.StatusWriter, "ERROR: get pod list in %s: %s\n", ns, err)
 				}
 				return
@@ -210,10 +212,13 @@ func CopyPodLogs(ctx context.Context, cs clientset.Interface, ns, podName string
 						defer maybeClose(output)
 					}
 
+					sinceTime := latest[id]
 					readCloser, err := logsForPod(ctx, cs, ns, pod.ObjectMeta.Name,
 						&v1.PodLogOptions{
-							Container: c.Name,
-							Follow:    true,
+							Container:  c.Name,
+							Follow:     true,
+							Timestamps: true,
+							SinceTime:  sinceTime,
 						})
 					if err != nil {
 						closeOutput()
@@ -227,6 +232,9 @@ func CopyPodLogs(ctx context.Context, cs clientset.Interface, ns, podName string
 						continue
 					}
 
+					active[name] = true
+					started[id] = true
+
 					go func() {
 						defer closeOutput()
 						first := true
@@ -235,10 +243,10 @@ func CopyPodLogs(ctx context.Context, cs clientset.Interface, ns, podName string
 							// If we never printed anything, then also skip the final message.
 							if !first {
 								if logWithPrefix != nil {
-									_, _ = fmt.Fprintf(logWithPrefix, "%s==== end of pod log ====\n", prefix)
+									_, _ = fmt.Fprintf(logWithPrefix, "%s==== end of pod log at %s ====\n", prefix, latest[id])
 								}
 								if logWithoutPrefix != nil {
-									_, _ = fmt.Fprintf(logWithoutPrefix, "==== end of pod log for container %s ====\n", name)
+									_, _ = fmt.Fprintf(logWithoutPrefix, "==== end of pod log for container %s at %s ====\n", name, latest[id])
 								}
 							}
 							active[name] = false
@@ -246,6 +254,7 @@ func CopyPodLogs(ctx context.Context, cs clientset.Interface, ns, podName string
 							readCloser.Close()
 						}()
 						scanner := bufio.NewScanner(readCloser)
+						var latestTS time.Time
 						for scanner.Scan() {
 							line := scanner.Text()
 							// Filter out the expected "end of stream" error message,
@@ -259,14 +268,28 @@ func CopyPodLogs(ctx context.Context, cs clientset.Interface, ns, podName string
 									// Because the same log might be written to multiple times
 									// in different test instances, log an extra line to separate them.
 									// Also provides some useful extra information.
+									since := "(initial part)"
+									if sinceTime != nil {
+										since = fmt.Sprintf("since %s", sinceTime)
+									}
 									if logWithPrefix != nil {
-										_, _ = fmt.Fprintf(logWithPrefix, "%s==== start of pod log ====\n", prefix)
+										_, _ = fmt.Fprintf(logWithPrefix, "%s==== start of pod log %s ====\n", prefix, since)
 									}
 									if logWithoutPrefix != nil {
-										_, _ = fmt.Fprintf(logWithoutPrefix, "==== start of pod log for container %s ====\n", name)
+										_, _ = fmt.Fprintf(logWithoutPrefix, "==== start of pod log for container %s %s ====\n", name, since)
 									}
 									first = false
 								}
+								index := strings.Index(line, " ")
+								if index > 0 {
+									ts, err := time.Parse(time.RFC3339, line[:index])
+									if err == nil {
+										latestTS = ts
+										// Log output typically has it's own log header with a time stamp,
+										// so let's strip the PodLogOptions time stamp.
+										line = line[index+1:]
+									}
+								}
 								if logWithPrefix != nil {
 									_, _ = fmt.Fprintf(logWithPrefix, "%s%s\n", prefix, line)
 								}
@@ -278,20 +301,29 @@ func CopyPodLogs(ctx context.Context, cs clientset.Interface, ns, podName string
 								}
 							}
 						}
+
+						if !latestTS.IsZero() {
+							m.Lock()
+							defer m.Unlock()
+							latest[id] = &meta.Time{Time: latestTS}
+						}
 					}()
-					active[name] = true
-					started[id] = true
 				}
 			}
 		}
 
 		// Watch events to see whether we can start logging
-		// and log interesting ones.
+		// and log interesting ones. Also check periodically,
+		// in case of failures which are not followed by
+		// some pod change.
+		ticker := time.NewTicker(30 * time.Second)
+		defer ticker.Stop()
 		check()
 		for {
 			select {
 			case <-watcher.ResultChan():
 				check()
+			case <-ticker.C:
 			case <-ctx.Done():
 				return
 			}