From: Jos Collin <jcollin@redhat.com>
Date: Thu, 4 Apr 2024 11:28:36 +0000 (+0530)
Subject: cephfs_mirror: update peer status for invalid metadata in remote snapshot
X-Git-Tag: v20.0.0~1233^2~2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=8055255c73f5eaadbe507999ff091cb269d72712;p=ceph.git

cephfs_mirror: update peer status for invalid metadata in remote snapshot

Fixes: https://tracker.ceph.com/issues/65317
Signed-off-by: Jos Collin <jcollin@redhat.com>
---

diff --git a/src/tools/cephfs_mirror/PeerReplayer.cc b/src/tools/cephfs_mirror/PeerReplayer.cc
index ce35b72ed0e9..b56ca9a2f1cc 100644
--- a/src/tools/cephfs_mirror/PeerReplayer.cc
+++ b/src/tools/cephfs_mirror/PeerReplayer.cc
@@ -528,8 +528,9 @@ int PeerReplayer::build_snap_map(const std::string &dir_root,
     uint64_t snap_id;
     if (is_remote) {
       if (!info.nr_snap_metadata) {
-        derr << ": snap_path=" << snap_path << " has invalid metadata in remote snapshot"
-             << dendl;
+        std::string failed_reason = "snapshot '" + snap  + "' has invalid metadata";
+        derr << ": " << failed_reason << dendl;
+        m_snap_sync_stats.at(dir_root).last_failed_reason = failed_reason;
         rv = -EINVAL;
       } else {
         auto metadata = decode_snap_metadata(info.snap_metadata, info.nr_snap_metadata);
@@ -1807,6 +1808,9 @@ void PeerReplayer::peer_status(Formatter *f) {
     f->open_object_section(dir_root);
     if (sync_stat.failed) {
       f->dump_string("state", "failed");
+      if (sync_stat.last_failed_reason) {
+	f->dump_string("failure_reason", *sync_stat.last_failed_reason);
+      }
     } else if (!sync_stat.current_syncing_snap) {
       f->dump_string("state", "idle");
     } else {
diff --git a/src/tools/cephfs_mirror/PeerReplayer.h b/src/tools/cephfs_mirror/PeerReplayer.h
index 6d04cf0cd838..933cb182635b 100644
--- a/src/tools/cephfs_mirror/PeerReplayer.h
+++ b/src/tools/cephfs_mirror/PeerReplayer.h
@@ -141,6 +141,7 @@ private:
   struct SnapSyncStat {
     uint64_t nr_failures = 0; // number of consecutive failures
     boost::optional<monotime> last_failed; // lat failed timestamp
+    boost::optional<std::string> last_failed_reason;
     bool failed = false; // hit upper cap for consecutive failures
     boost::optional<std::pair<uint64_t, std::string>> last_synced_snap;
     boost::optional<std::pair<uint64_t, std::string>> current_syncing_snap;
@@ -177,6 +178,7 @@ private:
     sync_stat.nr_failures = 0;
     sync_stat.failed = false;
     sync_stat.last_failed = boost::none;
+    sync_stat.last_failed_reason = boost::none;
   }
 
   void _set_last_synced_snap(const std::string &dir_root, uint64_t snap_id,