diff --git a/deps/rabbit/test/metrics_SUITE.erl b/deps/rabbit/test/metrics_SUITE.erl
index 4f83f0959f5a..fb7fe75a3d28 100644
--- a/deps/rabbit/test/metrics_SUITE.erl
+++ b/deps/rabbit/test/metrics_SUITE.erl
@@ -46,7 +46,8 @@ merge_app_env(Config) ->
     rabbit_ct_helpers:merge_app_env(Config,
                                     {rabbit, [
                                               {collect_statistics, fine},
-                                              {collect_statistics_interval, 500}
+                                              {collect_statistics_interval, 500},
+                                              {core_metrics_gc_interval, 5000}
                                              ]}).
 init_per_suite(Config) ->
     rabbit_ct_helpers:log_environment(),
@@ -186,6 +187,10 @@ queue_metric_idemp(Config, {N, R}) ->
               Queue
           end || _ <- lists:seq(1, N)],
 
+    ?awaitMatch(N, length(read_table_rpc(Config, queue_metrics)),
+                30000),
+    ?awaitMatch(N, length(read_table_rpc(Config, queue_coarse_metrics)),
+                30000),
     Table = [ Pid || {Pid, _, _} <- read_table_rpc(Config, queue_metrics)],
     Table2 = [ Pid || {Pid, _, _} <- read_table_rpc(Config, queue_coarse_metrics)],
     % refresh stats 'R' times
@@ -195,12 +200,16 @@ queue_metric_idemp(Config, {N, R}) ->
          gen_server2:call(Pid, flush)
       end|| {Pid, _, _} <- ChanTable ] || _ <- lists:seq(1, R)],
     force_metric_gc(Config),
+    ?awaitMatch(N, length(read_table_rpc(Config, queue_metrics)),
+                30000),
+    ?awaitMatch(N, length(read_table_rpc(Config, queue_coarse_metrics)),
+                30000),
     TableAfter = [ Pid || {Pid, _, _} <- read_table_rpc(Config,  queue_metrics)],
     TableAfter2 = [ Pid || {Pid, _, _} <- read_table_rpc(Config, queue_coarse_metrics)],
     [ delete_queue(Chan, Q) || Q <- Queues],
     rabbit_ct_client_helpers:close_connection(Conn),
-    (Table2 == TableAfter2) and (Table == TableAfter) and
-    (N == length(Table)) and (N == length(TableAfter)).
+    (lists:sort(Table2) == lists:sort(TableAfter2))
+        and (lists:sort(Table) == lists:sort(TableAfter)).
 
 connection_metric_count(Config, Ops) ->
     add_rem_counter(Config, Ops,
@@ -293,9 +302,12 @@ connection(Config) ->
     [_] = read_table_rpc(Config, connection_coarse_metrics),
     ok = rabbit_ct_client_helpers:close_connection(Conn),
     force_metric_gc(Config),
-    [] = read_table_rpc(Config, connection_created),
-    [] = read_table_rpc(Config, connection_metrics),
-    [] = read_table_rpc(Config, connection_coarse_metrics),
+    ?awaitMatch([], read_table_rpc(Config, connection_created),
+                30000),
+    ?awaitMatch([], read_table_rpc(Config, connection_metrics),
+                30000),
+    ?awaitMatch([], read_table_rpc(Config, connection_coarse_metrics),
+                30000),
     ok.
 
 channel(Config) ->
diff --git a/deps/rabbitmq_prometheus/test/prometheus_rabbitmq_federation_collector_SUITE.erl b/deps/rabbitmq_prometheus/test/prometheus_rabbitmq_federation_collector_SUITE.erl
index 0e83efc8cbce..b54a77f80e04 100644
--- a/deps/rabbitmq_prometheus/test/prometheus_rabbitmq_federation_collector_SUITE.erl
+++ b/deps/rabbitmq_prometheus/test/prometheus_rabbitmq_federation_collector_SUITE.erl
@@ -101,8 +101,10 @@ single_link_then_second_added(Config) ->
               timer:sleep(3000),
               [_L1] = rabbit_ct_broker_helpers:rpc(Config, 0,
                                                    rabbit_federation_status, status, []),
-              MFs = get_metrics(Config),
-              [?ONE_RUNNING_METRIC] = MFs,
+              rabbit_ct_helpers:eventually(?_assertEqual([?ONE_RUNNING_METRIC],
+                                                         get_metrics(Config)),
+                                           500,
+                                           5),
               maybe_declare_queue(Config, Ch, q(<<"fed.downstream2">>, [{<<"x-queue-type">>, longstr, <<"classic">>}])),
               %% here we race against queue.declare... most of the times there is going to be
               %% new status=starting metric. In this case we wait a bit more for running=2.