[Enhancement] Add agent task metrics (#29707)

Signed-off-by: gengjun-git <gengjun@starrocks.com> (cherry picked from commit 90f88b9)
StarRocks · Aug 25, 2023 · 8a928e3 · 8a928e3
1 parent 1b5a3ca
commit 8a928e3
Show file tree

Hide file tree

Showing 3 changed files with 82 additions and 62 deletions.
diff --git a/fe/fe-core/src/main/java/com/starrocks/metric/MetricRepo.java b/fe/fe-core/src/main/java/com/starrocks/metric/MetricRepo.java
@@ -74,6 +74,7 @@
 import com.starrocks.service.ExecuteEnv;
 import com.starrocks.system.Backend;
 import com.starrocks.system.SystemInfoService;
+import com.starrocks.task.AgentTaskQueue;
 import com.starrocks.transaction.TransactionState;
 import org.apache.logging.log4j.LogManager;
 import org.apache.logging.log4j.Logger;
@@ -709,6 +710,16 @@ public Long getValue() {
         };
         queryCoordinatorCount.addLabel(new MetricLabel("type", "query_coordinator_count"));
         STARROCKS_METRIC_REGISTER.addMetric(queryCoordinatorCount);
+
+        GaugeMetric<Long> agentTaskCount = new GaugeMetric<Long>("memory", MetricUnit.NOUNIT,
+                "The count of agent task") {
+            @Override
+            public Long getValue() {
+                return (long) AgentTaskQueue.getTaskNum();
+            }
+        };
+        agentTaskCount.addLabel(new MetricLabel("type", "agent_task_count"));
+        STARROCKS_METRIC_REGISTER.addMetric(agentTaskCount);
     }
 
     // to generate the metrics related to tablets of each backends

diff --git a/test/sql/test_metrics/R/test_memory_metrics b/test/sql/test_metrics/R/test_memory_metrics
@@ -1,10 +1,10 @@
--- name: test_memory_metrics_txn_callback_count
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"txn_callback_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+-- name: test_memory_metrics_load_profile_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"load_profile_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_delete_job_info_count
+-- name: test_memory_metrics_txn_bytes
 CREATE TABLE `t1` (
     a int
 ) ENGINE=OLAP
@@ -18,23 +18,20 @@ PROPERTIES (
 insert into t1 values (1);
 -- result:
 -- !result
-delete from t1 where a = 1;
--- result:
--- !result
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"delete_job_info_count\" | awk '{if ($NF >= 1) print "true"; else print "false";}'
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"txn_bytes\" | awk '{if ($NF >= 832) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_compaction_stats_count
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"compaction_stats_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+-- name: test_memory_metrics_agent_task_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"agent_task_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_tablet_count
+-- name: test_memory_metrics_tablet_bytes
 CREATE TABLE `t1` (
-    a int 
+    a int
 ) ENGINE=OLAP
 DUPLICATE KEY(`a`)
 DISTRIBUTED BY HASH(`a`) BUCKETS 10
@@ -43,12 +40,12 @@ PROPERTIES (
 );
 -- result:
 -- !result
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"tablet_count\" | awk '{if ($NF >= 10) print "true"; else print "false";}'
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"tablet_bytes\" | awk '{if ($NF >= 1360) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_query_profile_count
+-- name: test_memory_metrics_tablet_count
 CREATE TABLE `t1` (
     a int 
 ) ENGINE=OLAP
@@ -59,27 +56,26 @@ PROPERTIES (
 );
 -- result:
 -- !result
-set enable_profile=true;
--- result:
--- !result
-select count(*) from t1;
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"tablet_count\" | awk '{if ($NF >= 10) print "true"; else print "false";}'
 -- result:
 0
+true
 -- !result
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"query_profile_count\" | awk '{if ($NF >= 1) print "true"; else print "false";}'
+-- name: test_memory_metrics_running_task_run_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"running_task_run_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_pending_task_run_count
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"pending_task_run_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+-- name: test_memory_metrics_delete_job_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"delete_job_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_replica_bytes
+-- name: test_memory_metrics_query_profile_count
 CREATE TABLE `t1` (
-    a int
+    a int 
 ) ENGINE=OLAP
 DUPLICATE KEY(`a`)
 DISTRIBUTED BY HASH(`a`) BUCKETS 10
@@ -88,24 +84,19 @@ PROPERTIES (
 );
 -- result:
 -- !result
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"replica_bytes\" | awk '{if ($NF >= 1440) print "true"; else print "false";}'
+set enable_profile=true;
 -- result:
-0
-true
 -- !result
--- name: test_memory_metrics_delete_job_count
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"delete_job_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+select count(*) from t1;
 -- result:
 0
-true
 -- !result
--- name: test_memory_metrics_task_count
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"task_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"query_profile_count\" | awk '{if ($NF >= 1) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_txn_count
+-- name: test_memory_metrics_replica_count
 CREATE TABLE `t1` (
     a int
 ) ENGINE=OLAP
@@ -116,27 +107,40 @@ PROPERTIES (
 );
 -- result:
 -- !result
-insert into t1 values (1);
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"replica_count\" | awk '{if ($NF >= 10) print "true"; else print "false";}'
 -- result:
+0
+true
 -- !result
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"txn_count\" | awk '{if ($NF >= 1) print "true"; else print "false";}'
+-- name: test_memory_metrics_query_detail_count
+admin set frontend config ("enable_collect_query_detail_info" = "true");
+-- result:
+-- !result
+[UC]show frontends;
+-- result:
+172.26.92.139_9312_1692690164465	172.26.92.139	9312	8332	9332	9322	LEADER	1640562713	true	true	2292	2023-08-22 17:44:15	true		2023-08-22 17:42:09	add_agent_task_metrics-ba5647e
+-- !result
+admin set frontend config ("enable_collect_query_detail_info" = "false");
+-- result:
+-- !result
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"query_detail_count\" | awk '{if ($NF >= 1) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_catalogs_count
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"catalogs_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+-- name: test_memory_metrics_history_task_run_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"history_task_run_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_stream_load_task_count
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"stream_load_task_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+-- name: test_memory_metrics_compaction_stats_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"compaction_stats_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_txn_bytes
+-- name: test_memory_metrics_delete_job_info_count
 CREATE TABLE `t1` (
     a int
 ) ENGINE=OLAP
@@ -150,12 +154,21 @@ PROPERTIES (
 insert into t1 values (1);
 -- result:
 -- !result
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"txn_bytes\" | awk '{if ($NF >= 832) print "true"; else print "false";}'
+delete from t1 where a = 1;
+-- result:
+-- !result
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"delete_job_info_count\" | awk '{if ($NF >= 1) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_tablet_bytes
+-- name: test_memory_metrics_pending_task_run_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"pending_task_run_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+-- result:
+0
+true
+-- !result
+-- name: test_memory_metrics_replica_bytes
 CREATE TABLE `t1` (
     a int
 ) ENGINE=OLAP
@@ -166,40 +179,36 @@ PROPERTIES (
 );
 -- result:
 -- !result
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"tablet_bytes\" | awk '{if ($NF >= 1360) print "true"; else print "false";}'
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"replica_bytes\" | awk '{if ($NF >= 1440) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_running_task_run_count
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"running_task_run_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+-- name: test_memory_metrics_catalogs_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"catalogs_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_load_profile_count
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"load_profile_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+-- name: test_memory_metrics_stream_load_task_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"stream_load_task_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_query_detail_count
-admin set frontend config ("enable_collect_query_detail_info" = "true");
--- result:
--- !result
-[UC]show frontends;
--- result:
-172.26.92.139_9312_1690292126093	172.26.92.139	9312	8332	9332	9322	LEADER	2056272365	true	true	232533	2023-08-03 19:36:20	true		2023-08-03 16:04:39	add_memory_metrics-40b18af
--- !result
-admin set frontend config ("enable_collect_query_detail_info" = "false");
+-- name: test_memory_metrics_txn_callback_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"txn_callback_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
 -- result:
+0
+true
 -- !result
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"query_detail_count\" | awk '{if ($NF >= 1) print "true"; else print "false";}'
+-- name: test_memory_metrics_task_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"task_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
 -- result:
 0
 true
 -- !result
--- name: test_memory_metrics_replica_count
+-- name: test_memory_metrics_txn_count
 CREATE TABLE `t1` (
     a int
 ) ENGINE=OLAP
@@ -210,13 +219,10 @@ PROPERTIES (
 );
 -- result:
 -- !result
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"replica_count\" | awk '{if ($NF >= 10) print "true"; else print "false";}'
+insert into t1 values (1);
 -- result:
-0
-true
 -- !result
--- name: test_memory_metrics_history_task_run_count
-shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"history_task_run_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"txn_count\" | awk '{if ($NF >= 1) print "true"; else print "false";}'
 -- result:
 0
 true
@@ -232,4 +238,4 @@ shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"query_coordinat
 -- result:
 0
 true
--- !result
+-- !result
diff --git a/test/sql/test_metrics/T/test_memory_metrics b/test/sql/test_metrics/T/test_memory_metrics
@@ -134,3 +134,6 @@ shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"load_profile_co
 -- name: test_memory_metrics_query_coordinator_count
 shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"query_coordinator_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
 
+-- name: test_memory_metrics_agent_task_count
+shell: curl '${url}/metrics' | grep starrocks_fe_memory | grep \"agent_task_count\" | awk '{if ($NF >= 0) print "true"; else print "false";}'
+