Merge pull request #13070 from zmstone/0518-improve-kafka-connection-error-logs

0518 improve kafka connection error logs

apps/emqx_bridge_kafka/src/emqx_bridge_kafka_impl_producer.erl

@@ -572,33 +572,54 @@ check_client_connectivity(ClientId) ->
             {error, {find_client, Reason}}
     end.
 
+is_alive(Pid) ->
+    is_pid(Pid) andalso erlang:is_process_alive(Pid).
+
+error_summary(Map, [Error]) ->
+    Map#{error => Error};
+error_summary(Map, [Error | More]) ->
+    Map#{first_error => Error, total_errors => length(More) + 1}.
+
 check_if_healthy_leaders(ClientId, ClientPid, KafkaTopic, MaxPartitions) when is_pid(ClientPid) ->
-    Leaders =
+    case wolff_client:get_leader_connections(ClientPid, KafkaTopic, MaxPartitions) of
-        case wolff_client:get_leader_connections(ClientPid, KafkaTopic, MaxPartitions) of
+        {ok, Leaders} ->
-            {ok, LeadersToCheck} ->
+            %% Kafka is considered healthy as long as any of the partition leader is reachable.
-                %% Kafka is considered healthy as long as any of the partition leader is reachable.
+            case lists:partition(fun({_Partition, Pid}) -> is_alive(Pid) end, Leaders) of
-                lists:filtermap(
+                {[], Errors} ->
-                    fun({_Partition, Pid}) ->
+                    throw(
-                        case is_pid(Pid) andalso erlang:is_process_alive(Pid) of
+                        error_summary(
-                            true -> {true, Pid};
+                            #{
-                            _ -> false
+                                cause => "no_connected_partition_leader",
-                        end
+                                kafka_client => ClientId,
-                    end,
+                                kafka_topic => KafkaTopic
-                    LeadersToCheck
+                            },
-                );
+                            Errors
-            {error, _} ->
+                        )
-                []
+                    );
-        end,
+                {_, []} ->
-    case Leaders of
+                    ok;
-        [] ->
+                {_, Errors} ->
+                    ?SLOG(
+                        warning,
+                        "not_all_kafka_partitions_connected",
+                        error_summary(
+                            #{
+                                kafka_client => ClientId,
+                                kafka_topic => KafkaTopic
+                            },
+                            Errors
+                        )
+                    ),
+                    ok
+            end;
+        {error, Reason} ->
+            %% If failed to fetch metadata, wolff_client logs a warning level message
+            %% which includes the reason for each seed host
             throw(#{
-                error => no_connected_partition_leader,
+                cause => Reason,
                 kafka_client => ClientId,
-                kafka_topic => KafkaTopic,
+                kafka_topic => KafkaTopic
-                partitions_limit => MaxPartitions
+            })
-            });
-        _ ->
-            ok
     end.
 
 check_topic_status(ClientId, WolffClientPid, KafkaTopic) ->

apps/emqx_license/test/emqx_license_http_api_SUITE.erl

@@ -245,7 +245,7 @@ t_license_setting_bc(_Config) ->
     ?assertMatch(#{<<"max_connections">> := 25}, request_dump()),
     %% get
     GetRes = request(get, uri(["license", "setting"]), []),
-    %% aslo check that the settings return correctly
+    %% also check that the settings return correctly
     validate_setting(GetRes, <<"75%">>, <<"80%">>, 25),
     %% update
     Low = <<"50%">>,

changes/ee/fix-13070.en.md

@@ -0,0 +1,5 @@
+Improve Kafka connector error logs.
+
+Previously, specific error details, such as unreachable advertised listeners, were not logged.
+Now, error details are captured in the logs to provide more diagnostic information.
+To manage log verbosity, only the first occurrence of an error is logged, accompanied by the total count of similar errors.