Expose netsplit/partition info via Prometheus

cloudamqp · Sep 12, 2023 · 00f2451 · 00f2451
1 parent d50e318
commit 00f2451
Show file tree

Hide file tree

Showing 3 changed files with 52 additions and 0 deletions.
diff --git a/deps/rabbitmq_prometheus/app.bzl b/deps/rabbitmq_prometheus/app.bzl
@@ -12,6 +12,7 @@ def all_beam_files(name = "all_beam_files"):
             "src/collectors/prometheus_process_collector.erl",
             "src/collectors/prometheus_rabbitmq_alarm_metrics_collector.erl",
             "src/collectors/prometheus_rabbitmq_core_metrics_collector.erl",
+            "src/collectors/prometheus_rabbitmq_dynamic_collector.erl",
             "src/collectors/prometheus_rabbitmq_global_metrics_collector.erl",
             "src/rabbit_prometheus_app.erl",
             "src/rabbit_prometheus_dispatcher.erl",
@@ -42,6 +43,7 @@ def all_test_beam_files(name = "all_test_beam_files"):
             "src/collectors/prometheus_process_collector.erl",
             "src/collectors/prometheus_rabbitmq_alarm_metrics_collector.erl",
             "src/collectors/prometheus_rabbitmq_core_metrics_collector.erl",
+            "src/collectors/prometheus_rabbitmq_dynamic_collector.erl",
             "src/collectors/prometheus_rabbitmq_global_metrics_collector.erl",
             "src/rabbit_prometheus_app.erl",
             "src/rabbit_prometheus_dispatcher.erl",
@@ -83,6 +85,7 @@ def all_srcs(name = "all_srcs"):
             "src/collectors/prometheus_process_collector.erl",
             "src/collectors/prometheus_rabbitmq_alarm_metrics_collector.erl",
             "src/collectors/prometheus_rabbitmq_core_metrics_collector.erl",
+            "src/collectors/prometheus_rabbitmq_dynamic_collector.erl",
             "src/collectors/prometheus_rabbitmq_global_metrics_collector.erl",
             "src/rabbit_prometheus_app.erl",
             "src/rabbit_prometheus_dispatcher.erl",

diff --git a/deps/rabbitmq_prometheus/src/collectors/prometheus_rabbitmq_dynamic_collector.erl b/deps/rabbitmq_prometheus/src/collectors/prometheus_rabbitmq_dynamic_collector.erl
@@ -0,0 +1,48 @@
+%%% Collector for dynamic metrics that are calculated at collection time
+-module(prometheus_rabbitmq_dynamic_collector).
+
+-behaviour(prometheus_collector).
+-include_lib("prometheus/include/prometheus.hrl").
+
+-export([deregister_cleanup/1,
+         collect_mf/2]).
+
+-import(prometheus_model_helpers, [create_mf/5,
+                                   gauge_metric/1,
+                                   gauge_metric/2]).
+
+-define(METRIC_NAME_PREFIX, "rabbitmq_").
+
+-define(METRICS, [{partitioned_from, gauge,
+                   "Indicates that the current node is partitioned "
+                   "from the peer node during a network partition."}
+                 ]).
+
+%%====================================================================
+%% Collector API
+%%====================================================================
+
+deregister_cleanup(_) -> ok.
+
+collect_mf(_Registry, Callback) ->
+    _ = lists:foreach(
+          fun({Name, Type, Help}) ->
+                  Callback(
+                    prometheus_model_helpers:create_mf(
+                      ?METRIC_NAME(Name),
+                      Help,
+                      Type,
+                      values(Name))
+                   )
+          end,
+          ?METRICS
+         ),
+    ok.
+
+%%====================================================================
+%% Private Parts
+%%====================================================================
+
+values(partitioned_from) ->
+    [{[{peer, Node}], 1}
+     || Node <- rabbit_node_monitor:partitions()].
diff --git a/deps/rabbitmq_prometheus/src/rabbit_prometheus_dispatcher.erl b/deps/rabbitmq_prometheus/src/rabbit_prometheus_dispatcher.erl
@@ -17,6 +17,7 @@ build_dispatcher() ->
         prometheus_rabbitmq_core_metrics_collector,
         prometheus_rabbitmq_global_metrics_collector,
         prometheus_rabbitmq_alarm_metrics_collector,
+        prometheus_rabbitmq_dynamic_collector,
         prometheus_process_collector]),
     prometheus_registry:register_collectors('per-object', [
         prometheus_vm_system_info_collector,