IGNITE-29061 Add a metric for queries producing big result sets to HeavyQueriesTracker

docs/_docs/monitoring-metrics/new-metrics.adoc

@@ -522,4 +522,5 @@ Register name: `sql.queries.user`
 |canceled| long | The number of canceled SQL queries.
 |resultSetSizeHistogram| histogram | Histogram of fetched result set sizes for SQL queries.
 |maxResultSetSize| max value | Maximum fetched result set size for SQL queries.
+|bigResultSetEventsCount| long | Number of events when a SQL query produced a big result set.
 |===

modules/calcite/src/test/java/org/apache/ignite/internal/processors/query/calcite/integration/SqlDiagnosticIntegrationTest.java

@@ -97,6 +97,7 @@
 import static org.apache.ignite.internal.processors.performancestatistics.AbstractPerformanceStatisticsTest.stopCollectStatisticsAndRead;
 import static org.apache.ignite.internal.processors.query.QueryParserMetricsHolder.QUERY_PARSER_METRIC_GROUP_NAME;
 import static org.apache.ignite.internal.processors.query.calcite.CalciteQueryProcessor.IGNITE_CALCITE_USE_QUERY_BLOCKING_TASK_EXECUTOR;
+import static org.apache.ignite.internal.processors.query.running.HeavyQueriesTracker.BIG_RESULT_SET_EVENTS_CNT;
 import static org.apache.ignite.internal.processors.query.running.HeavyQueriesTracker.BIG_RESULT_SET_MSG;
 import static org.apache.ignite.internal.processors.query.running.HeavyQueriesTracker.LONG_QUERY_ERROR_MSG;
 import static org.apache.ignite.internal.processors.query.running.HeavyQueriesTracker.LONG_QUERY_EXEC_MSG;
@@ -874,6 +875,20 @@ public void testBigResultSet() throws Exception {
         assertTrue(logLsnr2.check(1000L));
     }
 
+    /** */
+    @Test
+    public void testBigResultSetMetric() {
+        grid(0).context().query().runningQueryManager().heavyQueriesTracker()
+            .setResultSetSizeThreshold(BIG_RESULT_SET_THRESHOLD);
+
+        sql(grid(0), "SELECT * FROM TABLE(SYSTEM_RANGE(1, ?))", BIG_RESULT_SET_THRESHOLD * 2 + 1);
+
+        LongMetric metric = grid(0).context().metric().registry(SQL_USER_QUERIES_REG_NAME)
+            .findMetric(BIG_RESULT_SET_EVENTS_CNT);
+
+        assertNotNull("Metric is not registered", metric);
+        assertTrue("Metric is not incremented", metric.value() > 0);
+    }
 
     /** */
     @Test

modules/core/src/main/java/org/apache/ignite/internal/processors/query/running/HeavyQueriesTracker.java

@@ -23,6 +23,7 @@
 import org.apache.ignite.IgniteLogger;
 import org.apache.ignite.internal.GridKernalContext;
 import org.apache.ignite.internal.IgniteInterruptedCheckedException;
+import org.apache.ignite.internal.processors.metric.impl.AtomicLongMetric;
 import org.apache.ignite.internal.util.typedef.internal.LT;
 import org.apache.ignite.internal.util.typedef.internal.U;
 import org.apache.ignite.internal.util.worker.GridWorker;
@@ -54,6 +55,12 @@ public final class HeavyQueriesTracker {
     /** */
     public static final String BIG_RESULT_SET_MSG = "Query produced big result set.";
 
+    /** Big result set events count metric name. */
+    public static final String BIG_RESULT_SET_EVENTS_CNT = "bigResultSetEventsCount";
+
+    /** Big result set events count metric description. */
+    private static final String BIG_RESULT_SET_EVENTS_CNT_DESC = "Number of events when a SQL query produced a big result set.";
+
     /** Queries collection. Sorted collection isn't used to reduce 'put' time. */
     private final ConcurrentHashMap<TrackableQuery, TimeoutChecker> qrys = new ConcurrentHashMap<>();
 
@@ -63,6 +70,9 @@ public final class HeavyQueriesTracker {
     /** Logger. */
     private final IgniteLogger log;
 
+    /** Big result set events count metric. */
+    private final AtomicLongMetric bigResultSetEvtsCnt;
+
     /** Long query timeout milliseconds. */
     private volatile long timeout;
 
@@ -97,6 +107,9 @@ public final class HeavyQueriesTracker {
     public HeavyQueriesTracker(GridKernalContext ctx) {
         log = ctx.log(HeavyQueriesTracker.class);
 
+        bigResultSetEvtsCnt = ctx.metric().registry(RunningQueryManager.SQL_USER_QUERIES_REG_NAME)
+            .longMetric(BIG_RESULT_SET_EVENTS_CNT, BIG_RESULT_SET_EVENTS_CNT_DESC);
+
         checkWorker = new GridWorker(ctx.igniteInstanceName(), "long-qry", log) {
             @Override protected void body() throws InterruptedException, IgniteInterruptedCheckedException {
                 while (true) {
@@ -156,7 +169,7 @@ public void stopTracking(TrackableQuery qryInfo, @Nullable Throwable err) {
      * @param qryInfo Query info.
      */
     public ResultSetChecker resultSetChecker(TrackableQuery qryInfo) {
-        return new ResultSetChecker(log, qryInfo, rsSizeThreshold, rsSizeThresholdMult);
+        return new ResultSetChecker(log, qryInfo, rsSizeThreshold, rsSizeThresholdMult, bigResultSetEvtsCnt);
     }
 
     /**
@@ -314,12 +327,22 @@ public static class ResultSetChecker {
         /** Big results flag. */
         private boolean bigResults;
 
+        /** Big result set events count metric. */
+        private final AtomicLongMetric bigResultSetEvtsCnt;
+
         /** Ctor. */
-        private ResultSetChecker(IgniteLogger log, TrackableQuery qryInfo, long threshold, int thresholdMult) {
+        private ResultSetChecker(
+            IgniteLogger log,
+            TrackableQuery qryInfo,
+            long threshold,
+            int thresholdMult,
+            AtomicLongMetric bigResultSetEvtsCnt
+        ) {
             this.log = log;
             this.qryInfo = qryInfo;
             this.threshold = threshold;
             this.thresholdMult = thresholdMult;
+            this.bigResultSetEvtsCnt = bigResultSetEvtsCnt;
         }
 
         /**
@@ -332,6 +355,9 @@ public void checkOnFetchNext() {
             if (threshold > 0 && fetchedSize >= threshold) {
                 LT.warn(log, BIG_RESULT_SET_MSG + qryInfo.queryInfo("fetched=" + fetchedSize));
 
+                if (bigResultSetEvtsCnt != null)
+                    bigResultSetEvtsCnt.increment();
+
                 if (thresholdMult > 1)
                     threshold *= thresholdMult;
                 else