[Enhancement] add metrics for statistics collect jobs (backport StarRocks#56693) (StarRocks#56798)

Signed-off-by: silverbullet233 <[email protected]>
Co-authored-by: eyes_on_me <[email protected]>
Co-authored-by: silverbullet233 <[email protected]>

Author: mergify[bot]

fe/fe-core/src/main/java/com/starrocks/metric/MetricRepo.java

@@ -147,6 +147,21 @@ public final class MetricRepo {
                     () -> new LongCounterMetric("query_queue_v2_category_state", MetricUnit.REQUESTS,
                             "the current state of each category"));
 
+
+    public static final MetricWithLabelGroup<LongCounterMetric> COUNTER_RUNNING_STATS_COLLECT_JOB =
+            new MetricWithLabelGroup<>("type",
+                    () -> new LongCounterMetric("running_stats_collect_job", MetricUnit.REQUESTS,
+                            "the number of running statistics collect jobs"));
+    public static final MetricWithLabelGroup<LongCounterMetric> COUNTER_TOTAL_STATS_COLLECT_JOB =
+            new MetricWithLabelGroup<>("type",
+                    () -> new LongCounterMetric("total_stats_collect_job", MetricUnit.REQUESTS,
+                            "the number of total statistics collect jobs"));
+    public static final MetricWithLabelGroup<LongCounterMetric> COUNTER_FAILED_STATS_COLLECT_JOB =
+            new MetricWithLabelGroup<>("type",
+                    () -> new LongCounterMetric("failed_stats_collect_job", MetricUnit.REQUESTS,
+                            "the number of failed statistics collect jobs"));
+
+
     public static LongCounterMetric COUNTER_UNFINISHED_BACKUP_JOB;
     public static LongCounterMetric COUNTER_UNFINISHED_RESTORE_JOB;
 

fe/fe-core/src/main/java/com/starrocks/statistic/ExternalFullStatisticsCollectJob.java

@@ -98,6 +98,11 @@ public List<String> getPartitionNames() {
         return partitionNames;
     }
 
+    @Override
+    public String getName() {
+        return "ExternalFull";
+    }
+
     @Override
     public void collect(ConnectContext context, AnalyzeStatus analyzeStatus) throws Exception {
         long finishedSQLNum = 0;

fe/fe-core/src/main/java/com/starrocks/statistic/ExternalHistogramStatisticsCollectJob.java

@@ -69,6 +69,11 @@ public String getCatalogName() {
         return catalogName;
     }
 
+    @Override
+    public String getName() {
+        return "ExternalHistogram";
+    }
+
     @Override
     public void collect(ConnectContext context, AnalyzeStatus analyzeStatus) throws Exception {
         context.getSessionVariable().setNewPlanerAggStage(1);

fe/fe-core/src/main/java/com/starrocks/statistic/FullStatisticsCollectJob.java

@@ -382,4 +382,9 @@ public String toString() {
         sb.append('}');
         return sb.toString();
     }
+
+    @Override
+    public String getName() {
+        return "Full";
+    }
 }

fe/fe-core/src/main/java/com/starrocks/statistic/HistogramStatisticsCollectJob.java

@@ -157,4 +157,9 @@ private String buildCollectHistogram(Database database, Table table, double samp
         builder.append(build(context, COLLECT_HISTOGRAM_STATISTIC_TEMPLATE));
         return builder.toString();
     }
+
+    @Override
+    public String getName() {
+        return "Histogram";
+    }
 }

fe/fe-core/src/main/java/com/starrocks/statistic/SampleStatisticsCollectJob.java

@@ -87,4 +87,9 @@ public void collect(ConnectContext context, AnalyzeStatus analyzeStatus) throws
         }
     }
 
+    @Override
+    public String getName() {
+        return "Sample";
+    }
+
 }

fe/fe-core/src/main/java/com/starrocks/statistic/StatisticExecutor.java

@@ -31,6 +31,8 @@
 import com.starrocks.common.Status;
 import com.starrocks.common.util.DebugUtil;
 import com.starrocks.common.util.UUIDUtil;
+import com.starrocks.metric.LongCounterMetric;
+import com.starrocks.metric.MetricRepo;
 import com.starrocks.qe.ConnectContext;
 import com.starrocks.qe.StmtExecutor;
 import com.starrocks.server.GlobalStateMgr;
@@ -339,6 +341,11 @@ public AnalyzeStatus collectStatistics(ConnectContext statsConnectCtx,
         Database db = statsJob.getDb();
         Table table = statsJob.getTable();
 
+        LongCounterMetric runningJobs = MetricRepo.COUNTER_RUNNING_STATS_COLLECT_JOB.getMetric(statsJob.getName());
+        LongCounterMetric totalJobs = MetricRepo.COUNTER_TOTAL_STATS_COLLECT_JOB.getMetric(statsJob.getName());
+        runningJobs.increase(1L);
+        totalJobs.increase(1L);
+
         try {
             Stopwatch watch = Stopwatch.createStarted();
             statsConnectCtx.getSessionVariable().setEnableProfile(Config.enable_statistics_collect_profile);
@@ -356,8 +363,10 @@ public AnalyzeStatus collectStatistics(ConnectContext statsConnectCtx,
             analyzeStatus.setEndTime(LocalDateTime.now());
             analyzeStatus.setReason(e.getMessage());
             GlobalStateMgr.getCurrentState().getAnalyzeMgr().addAnalyzeStatus(analyzeStatus);
+            MetricRepo.COUNTER_FAILED_STATS_COLLECT_JOB.getMetric(statsJob.getName()).increase(1L);
             return analyzeStatus;
         } finally {
+            runningJobs.increase(-1L);
             GlobalStateMgr.getCurrentState().getAnalyzeMgr().unregisterConnection(analyzeStatus.getId(), false);
         }
 

fe/fe-core/src/main/java/com/starrocks/statistic/StatisticsCollectJob.java

@@ -144,6 +144,8 @@ public Priority getPriority() {
         return this.priority;
     }
 
+    abstract String getName();
+
     protected void setDefaultSessionVariable(ConnectContext context) {
         SessionVariable sessionVariable = context.getSessionVariable();
         // Statistics collecting is not user-specific, which means response latency is not that important.