From: Yuval Lifshitz <ylifshit@redhat.com>
Date: Thu, 1 Jun 2023 06:13:06 +0000 (+0000)
Subject: rgw/kafka: make sure that destroy is called after connection is removed
X-Git-Tag: v19.0.0~1045^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=refs%2Fpull%2F51864%2Fhead;p=ceph.git

rgw/kafka: make sure that destroy is called after connection is removed

in addition. the dtor calls "destroy()", so there is no need to call
it explictly

Fixes: https://tracker.ceph.com/issues/61540

Signed-off-by: Yuval Lifshitz <ylifshit@redhat.com>
---

diff --git a/src/rgw/rgw_kafka.cc b/src/rgw/rgw_kafka.cc
index 88b4f16932a5..7bb65aef587c 100644
--- a/src/rgw/rgw_kafka.cc
+++ b/src/rgw/rgw_kafka.cc
@@ -100,10 +100,12 @@ struct connection_t {
     // fire all remaining callbacks (if not fired by rd_kafka_flush)
     std::for_each(callbacks.begin(), callbacks.end(), [this](auto& cb_tag) {
         cb_tag.cb(status);
-        ldout(cct, 20) << "Kafka destroy: invoking callback with tag=" << cb_tag.tag << dendl;
+        ldout(cct, 20) << "Kafka destroy: invoking callback with tag=" << cb_tag.tag << 
+          " for: " << broker << dendl;
       });
     callbacks.clear();
     delivery_tag = 1;
+    ldout(cct, 20) << "Kafka destroy: complete for: " << broker << dendl;
   }
 
   bool is_ok() const {
@@ -459,7 +461,6 @@ private:
         // Checking the connection idlesness
         if(conn->timestamp.sec() + max_idle_time < ceph_clock_now()) {
           ldout(conn->cct, 20) << "kafka run: deleting a connection due to idle behaviour: " << ceph_clock_now() << dendl;
-          conn->destroy(STATUS_CONNECTION_IDLE);
           std::lock_guard lock(connections_lock);
           conn_it = connections.erase(conn_it);
           --connection_count; \