From: Adam C. Emerson <aemerson@redhat.com>
Date: Wed, 12 Mar 2025 15:53:43 +0000 (-0400)
Subject: rgw: Try to handle unwatch errors sensibly
X-Git-Tag: testing/wip-khiremat-testing-20250422.120708-squid-debug~34^2
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=ff248d7ed94cc441a2e7f3254cb8c0d53d3997d1;p=ceph-ci.git

rgw: Try to handle unwatch errors sensibly

IF we get `-ENOENT` from unwatch just stop trying to renew.

Otherwise, schedule retry on watch failure, but add a maximum retry
timeout.

Fixes: https://tracker.ceph.com/issues/70422

Signed-off-by: Adam C. Emerson <aemerson@redhat.com>
(cherry picked from commit 34366f0f0d8e13eb8fef1e253794303b699e08b2)

Fixes: https://tracker.ceph.com/issues/70526
Signed-off-by: Adam C. Emerson <aemerson@redhat.com>
---

diff --git a/src/rgw/services/svc_notify.cc b/src/rgw/services/svc_notify.cc
index 5593dee9ae2..baa3512c1b7 100644
--- a/src/rgw/services/svc_notify.cc
+++ b/src/rgw/services/svc_notify.cc
@@ -33,6 +33,7 @@ class RGWWatcher : public DoutPrefixProvider , public librados::WatchCtx2 {
   int register_ret{0};
   bool unregister_done{false};
   librados::AioCompletion *register_completion{nullptr};
+  uint64_t retries = 0;
 
   class C_ReinitWatch : public Context {
     RGWWatcher *watcher;
@@ -86,15 +87,28 @@ public:
   }
 
   void reinit() {
+    if (retries > 100) {
+      lderr(cct) << "ERROR: Looping in attempt to reinit watch. Halting."
+		 << dendl;
+      abort();
+    }
     if(!unregister_done) {
       int ret = unregister_watch();
       if (ret < 0) {
         ldout(cct, 0) << "ERROR: unregister_watch() returned ret=" << ret << dendl;
+	if (-2 == ret) {
+	  // Going down there is no such watch.
+	  return;
+	} else {
+	  ++retries;
+	  svc->schedule_context(new C_ReinitWatch(this));
+	}
       }
     }
     int ret = register_watch();
     if (ret < 0) {
       ldout(cct, 0) << "ERROR: register_watch() returned ret=" << ret << dendl;
+      ++retries;
       svc->schedule_context(new C_ReinitWatch(this));
       return;
     }