[improve][broker] Add topic_load_failed metric (apache#19236)

Co-authored-by: daojun <[email protected]>
(cherry picked from commit 43b3622)

Author: 2 people

pulsar-broker/src/main/java/org/apache/pulsar/broker/service/BrokerService.java

@@ -1225,6 +1225,10 @@ public void deleteTopicAuthenticationWithRetry(String topic, CompletableFuture<V
 
     private CompletableFuture<Optional<Topic>> createNonPersistentTopic(String topic) {
         CompletableFuture<Optional<Topic>> topicFuture = new CompletableFuture<>();
+        topicFuture.exceptionally(t -> {
+            pulsarStats.recordTopicLoadFailed();
+            return null;
+        });
         if (!pulsar.getConfiguration().isEnableNonPersistentTopics()) {
             if (log.isDebugEnabled()) {
                 log.debug("Broker is unable to load non-persistent topic {}", topic);
@@ -1539,6 +1543,11 @@ private void createPersistentTopic(final String topic, boolean createIfMissing,
         TopicName topicName = TopicName.get(topic);
         final long topicCreateTimeMs = TimeUnit.NANOSECONDS.toMillis(System.nanoTime());
 
+        topicFuture.exceptionally(t -> {
+            pulsarStats.recordTopicLoadFailed();
+            return null;
+        });
+
         if (isTransactionSystemTopic(topicName)) {
             String msg = String.format("Can not create transaction system topic %s", topic);
             log.warn(msg);

pulsar-broker/src/main/java/org/apache/pulsar/broker/service/PulsarStats.java

@@ -240,6 +240,10 @@ public void recordTopicLoadTimeValue(String topic, long topicLoadLatencyMs) {
         }
     }
 
+    public void recordTopicLoadFailed() {
+        brokerOperabilityMetrics.recordTopicLoadFailed();
+    }
+
     public void recordConnectionCreate() {
         brokerOperabilityMetrics.recordConnectionCreate();
     }

pulsar-broker/src/main/java/org/apache/pulsar/broker/stats/BrokerOperabilityMetrics.java

@@ -19,6 +19,7 @@
 package org.apache.pulsar.broker.stats;
 
 import com.google.common.collect.Maps;
+import io.prometheus.client.Counter;
 import java.util.ArrayList;
 import java.util.List;
 import java.util.Map;
@@ -29,6 +30,7 @@
 /**
  */
 public class BrokerOperabilityMetrics {
+    private static final Counter TOPIC_LOAD_FAILED = Counter.build("topic_load_failed", "-").register();
     private final List<Metrics> metricsList;
     private final String localCluster;
     private final DimensionStats topicLoadStats;
@@ -90,7 +92,9 @@ Map<String, String> getDimensionMap(String metricsName) {
     }
 
     Metrics getTopicLoadMetrics() {
-        return getDimensionMetrics("topic_load_times", "topic_load", topicLoadStats);
+        Metrics metrics = getDimensionMetrics("topic_load_times", "topic_load", topicLoadStats);
+        metrics.put("brk_topic_load_failed_count", TOPIC_LOAD_FAILED.get());
+        return metrics;
     }
 
     Metrics getZkWriteLatencyMetrics() {
@@ -136,6 +140,10 @@ public void recordZkReadLatencyTimeValue(long topicLoadLatencyMs) {
         zkReadLatencyStats.recordDimensionTimeValue(topicLoadLatencyMs, TimeUnit.MILLISECONDS);
     }
 
+    public void recordTopicLoadFailed() {
+        this.TOPIC_LOAD_FAILED.inc();
+    }
+
     public void recordConnectionCreate() {
         this.connectionTotalCreatedCount.increment();
         this.connectionActive.increment();

pulsar-broker/src/test/java/org/apache/pulsar/broker/stats/PrometheusMetricsTest.java

@@ -316,6 +316,10 @@ public void testPerTopicStats() throws Exception {
         assertEquals(cm.size(), 1);
         assertEquals(cm.get(0).tags.get("cluster"), "test");
 
+        cm = (List<Metric>) metrics.get("topic_load_failed_total");
+        assertEquals(cm.size(), 1);
+        assertEquals(cm.get(0).tags.get("cluster"), "test");
+
         cm = (List<Metric>) metrics.get("pulsar_in_bytes_total");
         assertEquals(cm.size(), 2);
         assertEquals(cm.get(0).tags.get("topic"), "persistent://my-property/use/my-ns/my-topic2");