HDDS-1783 : Latency metric for applyTransaction in ContainerStateMachine (#1363).

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/common/transport/server/ratis/CSMMetrics.java

@@ -60,6 +60,9 @@ public class CSMMetrics {
   private @Metric MutableCounterLong numStartTransactionVerifyFailures;
   private @Metric MutableCounterLong numContainerNotOpenVerifyFailures;
 
+  private @Metric MutableRate applyTransaction;
+  private @Metric MutableRate writeStateMachineData;
+
   public CSMMetrics() {
     int numCmdTypes = ContainerProtos.Type.values().length;
     this.opsLatency = new MutableRate[numCmdTypes];
@@ -186,6 +189,10 @@ public class CSMMetrics {
     return numBytesCommittedCount.value();
   }
 
+  public MutableRate getApplyTransactionLatency() {
+    return applyTransaction;
+  }
+
   public void incPipelineLatency(ContainerProtos.Type type, long latencyNanos) {
     opsLatency[type.ordinal()].add(latencyNanos);
     transactionLatency.add(latencyNanos);
@@ -199,6 +206,13 @@ public class CSMMetrics {
     numContainerNotOpenVerifyFailures.incr();
   }
 
+  public void recordApplyTransactionCompletion(long latencyNanos) {
+    applyTransaction.add(latencyNanos);
+  }
+
+  public void recordWriteStateMachineCompletion(long latencyNanos) {
+    writeStateMachineData.add(latencyNanos);
+  }
 
   public void unRegister() {
     MetricsSystem ms = DefaultMetricsSystem.instance();

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/common/transport/server/ratis/ContainerStateMachine.java

@@ -411,7 +411,8 @@ public class ContainerStateMachine extends BaseStateMachine {
   }
 
   private CompletableFuture<Message> handleWriteChunk(
-      ContainerCommandRequestProto requestProto, long entryIndex, long term) {
+      ContainerCommandRequestProto requestProto, long entryIndex, long term,
+      long startTime) {
     final WriteChunkRequestProto write = requestProto.getWriteChunk();
     RaftServer server = ratisServer.getServer();
     Preconditions.checkState(server instanceof RaftServerProxy);
@@ -461,6 +462,8 @@ public class ContainerStateMachine extends BaseStateMachine {
             write.getBlockID() + " logIndex " + entryIndex + " chunkName " +
             write.getChunkData().getChunkName());
         raftFuture.complete(r::toByteString);
+        metrics.recordWriteStateMachineCompletion(
+            Time.monotonicNowNanos() - startTime);
       }
 
       writeChunkFutureMap.remove(entryIndex);
@@ -477,6 +480,7 @@ public class ContainerStateMachine extends BaseStateMachine {
   public CompletableFuture<Message> writeStateMachineData(LogEntryProto entry) {
     try {
       metrics.incNumWriteStateMachineOps();
+      long writeStateMachineStartTime = Time.monotonicNowNanos();
       ContainerCommandRequestProto requestProto =
           getContainerCommandRequestProto(
               entry.getStateMachineLogEntry().getLogData());
@@ -493,7 +497,7 @@ public class ContainerStateMachine extends BaseStateMachine {
       switch (cmdType) {
       case WriteChunk:
         return handleWriteChunk(requestProto, entry.getIndex(),
-            entry.getTerm());
+            entry.getTerm(), writeStateMachineStartTime);
       default:
         throw new IllegalStateException("Cmd Type:" + cmdType
             + " should not have state machine data");
@@ -673,6 +677,7 @@ public class ContainerStateMachine extends BaseStateMachine {
             .setTerm(trx.getLogEntry().getTerm())
             .setLogIndex(index);
 
+    long applyTxnStartTime = Time.monotonicNowNanos();
     try {
       applyTransactionSemaphore.acquire();
       metrics.incNumApplyTransactionsOps();
@@ -740,7 +745,11 @@ public class ContainerStateMachine extends BaseStateMachine {
           }
         }
         return applyTransactionFuture;
-      }).whenComplete((r, t) -> applyTransactionSemaphore.release());
+      }).whenComplete((r, t) ->  {
+        applyTransactionSemaphore.release();
+        metrics.recordApplyTransactionCompletion(
+            Time.monotonicNowNanos() - applyTxnStartTime);
+      });
       return applyTransactionFuture;
     } catch (IOException | InterruptedException e) {
       metrics.incNumApplyTransactionsFails();

hadoop-ozone/dist/src/main/compose/ozonesecure-mr/docker-config

@@ -47,7 +47,7 @@ HDFS-SITE.XML_dfs.datanode.address=0.0.0.0:1019
 HDFS-SITE.XML_dfs.datanode.http.address=0.0.0.0:1012
 CORE-SITE.XML_dfs.data.transfer.protection=authentication
 CORE-SITE.XML_hadoop.security.authentication=kerberos
-COER-SITE.XML_hadoop.security.auth_to_local=RULE:[2:$1@$0](.*@EXAMPLE.COM)s/@.*///L
+CORE-SITE.XML_hadoop.security.auth_to_local=RULE:[2:$1@$0](.*@EXAMPLE.COM)s/@.*///L
 CORE-SITE.XML_hadoop.security.key.provider.path=kms://http@kms:9600/kms
 
 #temporary disable authorization as org.apache.hadoop.yarn.server.api.ResourceTrackerPB is not properly annotated to support it

hadoop-ozone/integration-test/src/test/java/org/apache/hadoop/ozone/container/common/transport/server/ratis/TestCSMMetrics.java

@@ -19,6 +19,7 @@
 package org.apache.hadoop.ozone.container.common.transport.server.ratis;
 
 import static org.apache.hadoop.test.MetricsAsserts.assertCounter;
+import static org.apache.hadoop.test.MetricsAsserts.getDoubleGauge;
 import static org.apache.hadoop.test.MetricsAsserts.getMetrics;
 
 import java.io.File;
@@ -49,6 +50,8 @@ import org.apache.hadoop.test.GenericTestUtils;
 import org.apache.hadoop.hdds.conf.OzoneConfiguration;
 
 import static org.apache.ratis.rpc.SupportedRpcType.GRPC;
+import static org.junit.Assert.assertTrue;
+
 import org.apache.ratis.protocol.RaftGroupId;
 import org.apache.ratis.util.function.CheckedBiConsumer;
 
@@ -118,6 +121,12 @@ public class TestCSMMetrics {
       assertCounter("NumStartTransactionVerifyFailures", 0L, metric);
       assertCounter("NumContainerNotOpenVerifyFailures", 0L, metric);
       assertCounter("WriteChunkNumOps", 0L, metric);
+      double applyTransactionLatency = getDoubleGauge(
+          "ApplyTransactionAvgTime", metric);
+      assertTrue(applyTransactionLatency == 0.0);
+      double writeStateMachineLatency = getDoubleGauge(
+          "WriteStateMachineDataAvgTime", metric);
+      assertTrue(writeStateMachineLatency == 0.0);
 
       // Write Chunk
       BlockID blockID = ContainerTestHelper.getTestBlockID(ContainerTestHelper.
@@ -152,6 +161,13 @@ public class TestCSMMetrics {
           RaftGroupId.valueOf(pipeline.getId().getId()).toString());
       assertCounter("NumQueryStateMachineOps", 1L, metric);
       assertCounter("NumApplyTransactionOps", 1L, metric);
+      applyTransactionLatency = getDoubleGauge(
+          "ApplyTransactionAvgTime", metric);
+      assertTrue(applyTransactionLatency > 0.0);
+      writeStateMachineLatency = getDoubleGauge(
+          "WriteStateMachineDataAvgTime", metric);
+      assertTrue(writeStateMachineLatency > 0.0);
+
     } finally {
       if (client != null) {
         client.close();