rgw/notification: add exception handling for persistent notification thread

Fixes: https://tracker.ceph.com/issues/49322
Signed-off-by: Yuval Lifshitz <ylifshit@redhat.com>

diff --git a/src/rgw/rgw_notify.cc b/src/rgw/rgw_notify.cc
index c21c986f011d5538d7bc9928aa547b4eab021751..10fd3b57baceaae3b080914a7174d4007ad21d1c 100644 (file)
--- a/src/rgw/rgw_notify.cc
+++ b/src/rgw/rgw_notify.cc
@@ -360,7 +360,6 @@ class Manager : public DoutPrefixProvider {
           << queue_name << dendl;
         }
       }
-
     }
   }
 
@@ -381,6 +380,8 @@ class Manager : public DoutPrefixProvider {
     const auto max_jitter = 500; // ms
     std::uniform_int_distribution<> duration_jitter(min_jitter, max_jitter);
 
+    std::vector<std::string> queue_gc;
+    std::mutex queue_gc_lock;
     while (true) {
       Timer timer(io_context);
       const auto duration = (has_error ? 
@@ -399,8 +400,6 @@ class Manager : public DoutPrefixProvider {
         continue;
       }
 
-      std::vector<std::string> queue_gc;
-      std::mutex queue_gc_lock;
       for (const auto& queue_name : queues) {
         // try to lock the queue to check if it is owned by this rgw
         // or if ownershif needs to be taken
@@ -493,9 +492,16 @@ public:
       // start the worker threads to do the actual queue processing
       const std::string WORKER_THREAD_NAME = "notif-worker";
       for (auto worker_id = 0U; worker_id < worker_count; ++worker_id) {
-        workers.emplace_back([this]() noexcept { io_context.run(); });
+        workers.emplace_back([this]() {
+          try {
+            io_context.run(); 
+          } catch (const std::exception& err) {
+            ldpp_dout(this, 10) << "Notification worker failed with error: " << err.what() << dendl;
+            throw(err);
+          }
+        });
         const auto rc = ceph_pthread_setname(workers.back().native_handle(), 
-            (WORKER_THREAD_NAME+std::to_string(worker_id)).c_str());
+          (WORKER_THREAD_NAME+std::to_string(worker_id)).c_str());
         ceph_assert(rc == 0);
       }
       ldpp_dout(this, 10) << "Started notification manager with: " << worker_count << " workers" << dendl;