feat(new_metircs): collect the number of primary and secondary replicas

apache · Dec 2, 2024 · 831d358 · 831d358
1 parent d9f2600
commit 831d358
Show file tree

Hide file tree

Showing 2 changed files with 30 additions and 6 deletions.
diff --git a/src/replica/replica_stub.cpp b/src/replica/replica_stub.cpp
@@ -115,6 +115,16 @@ METRIC_DEFINE_gauge_int64(server,
                           dsn::metric_unit::kReplicas,
                           "The number of closing replicas");
 
+METRIC_DEFINE_gauge_int64(server,
+                          primary_replicas,
+                          dsn::metric_unit::kReplicas,
+                          "The number of primary replicas");
+
+METRIC_DEFINE_gauge_int64(server,
+                          secondary_replicas,
+                          dsn::metric_unit::kReplicas,
+                          "The number of secondary replicas");
+
 METRIC_DEFINE_gauge_int64(server,
                           learning_replicas,
                           dsn::metric_unit::kReplicas,
@@ -368,6 +378,8 @@ replica_stub::replica_stub(replica_state_subscriber subscriber /*= nullptr*/,
       METRIC_VAR_INIT_server(total_replicas),
       METRIC_VAR_INIT_server(opening_replicas),
       METRIC_VAR_INIT_server(closing_replicas),
+      METRIC_VAR_INIT_server(primary_replicas),
+      METRIC_VAR_INIT_server(secondary_replicas),
       METRIC_VAR_INIT_server(learning_replicas),
       METRIC_VAR_INIT_server(learning_replicas_max_duration_ms),
       METRIC_VAR_INIT_server(learning_replicas_max_copy_file_bytes),
@@ -1625,26 +1637,30 @@ void replica_stub::on_replicas_stat()
     LOG_INFO("start replicas statistics, replica_count = {}", rep_stat_info_by_gpid.size());
 
     // statistic learning info
-    uint64_t learning_count = 0;
     uint64_t learning_max_duration_time_ms = 0;
     uint64_t learning_max_copy_file_size = 0;
     uint64_t bulk_load_running_count = 0;
     uint64_t bulk_load_max_ingestion_time_ms = 0;
     uint64_t bulk_load_max_duration_time_ms = 0;
-    uint64_t splitting_count = 0;
     uint64_t splitting_max_duration_time_ms = 0;
     uint64_t splitting_max_async_learn_time_ms = 0;
     uint64_t splitting_max_copy_file_size = 0;
+
+    std::map<partition_status::type, size_t> status_counts;
     for (const auto &[_, rep_stat_info] : rep_stat_info_by_gpid) {
         const auto &rep = rep_stat_info.rep;
+        ++status_counts[rep->status()];
+
         if (rep->status() == partition_status::PS_POTENTIAL_SECONDARY) {
-            learning_count++;
             learning_max_duration_time_ms = std::max(
                 learning_max_duration_time_ms, rep->_potential_secondary_states.duration_ms());
             learning_max_copy_file_size =
                 std::max(learning_max_copy_file_size,
                          rep->_potential_secondary_states.learning_copy_file_size);
+
+            continue;
         }
+
         if (rep->status() == partition_status::PS_PRIMARY ||
             rep->status() == partition_status::PS_SECONDARY) {
             if (rep->get_bulk_loader()->get_bulk_load_status() != bulk_load_status::BLS_INVALID) {
@@ -1654,26 +1670,32 @@ void replica_stub::on_replicas_stat()
                 bulk_load_max_duration_time_ms =
                     std::max(bulk_load_max_duration_time_ms, rep->get_bulk_loader()->duration_ms());
             }
+
+            continue;
         }
+
         // splitting_max_copy_file_size, rep->_split_states.copy_file_size
         if (rep->status() == partition_status::PS_PARTITION_SPLIT) {
-            splitting_count++;
             splitting_max_duration_time_ms =
                 std::max(splitting_max_duration_time_ms, rep->_split_states.total_ms());
             splitting_max_async_learn_time_ms =
                 std::max(splitting_max_async_learn_time_ms, rep->_split_states.async_learn_ms());
             splitting_max_copy_file_size =
                 std::max(splitting_max_copy_file_size, rep->_split_states.splitting_copy_file_size);
+
+            continue;
         }
     }
 
-    METRIC_VAR_SET(learning_replicas, learning_count);
+    METRIC_VAR_SET(primary_replicas, status_counts[partition_status::PS_PRIMARY]);
+    METRIC_VAR_SET(secondary_replicas, status_counts[partition_status::PS_SECONDARY]);
+    METRIC_VAR_SET(learning_replicas, status_counts[partition_status::PS_POTENTIAL_SECONDARY]);
     METRIC_VAR_SET(learning_replicas_max_duration_ms, learning_max_duration_time_ms);
     METRIC_VAR_SET(learning_replicas_max_copy_file_bytes, learning_max_copy_file_size);
     METRIC_VAR_SET(bulk_load_running_count, bulk_load_running_count);
     METRIC_VAR_SET(bulk_load_ingestion_max_duration_ms, bulk_load_max_ingestion_time_ms);
     METRIC_VAR_SET(bulk_load_max_duration_ms, bulk_load_max_duration_time_ms);
-    METRIC_VAR_SET(splitting_replicas, splitting_count);
+    METRIC_VAR_SET(splitting_replicas, status_counts[partition_status::PS_PARTITION_SPLIT]);
     METRIC_VAR_SET(splitting_replicas_max_duration_ms, splitting_max_duration_time_ms);
     METRIC_VAR_SET(splitting_replicas_async_learn_max_duration_ms,
                    splitting_max_async_learn_time_ms);

diff --git a/src/replica/replica_stub.h b/src/replica/replica_stub.h
@@ -528,6 +528,8 @@ class replica_stub : public serverlet<replica_stub>, public ref_counter
     METRIC_VAR_DECLARE_gauge_int64(opening_replicas);
     METRIC_VAR_DECLARE_gauge_int64(closing_replicas);
 
+    METRIC_VAR_DECLARE_gauge_int64(primary_replicas);
+    METRIC_VAR_DECLARE_gauge_int64(secondary_replicas);
     METRIC_VAR_DECLARE_gauge_int64(learning_replicas);
     METRIC_VAR_DECLARE_gauge_int64(learning_replicas_max_duration_ms);
     METRIC_VAR_DECLARE_gauge_int64(learning_replicas_max_copy_file_bytes);