Add artifact type for JFR recording

3 files changed, 87 insertions, 5 deletions

diff --git a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/maintenance/servicedump/JavaFlightRecorder.java b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/maintenance/servicedump/JavaFlightRecorder.java
new file mode 100644
index 00000000000..289f5622aeb
--- /dev/null
+++ b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/maintenance/servicedump/JavaFlightRecorder.java
@@ -0,0 +1,51 @@
+// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+package com.yahoo.vespa.hosted.node.admin.maintenance.servicedump;
+
+import com.yahoo.concurrent.Sleeper;
+import com.yahoo.vespa.hosted.node.admin.container.ContainerOperations;
+import com.yahoo.vespa.hosted.node.admin.nodeagent.NodeAgentContext;
+import com.yahoo.vespa.hosted.node.admin.task.util.file.UnixPath;
+
+import java.io.IOException;
+import java.time.Duration;
+import java.util.List;
+
+/**
+ * Creates a Java Flight Recorder dump.
+ *
+ * @author bjorncs
+ */
+class JavaFlightRecorder extends AbstractProducer {
+
+    private final Sleeper sleeper;
+
+    JavaFlightRecorder(ContainerOperations container, Sleeper sleeper) {
+        super(container);
+        this.sleeper = sleeper;
+    }
+
+    static String NAME = "jfr-recording";
+
+    @Override public String name() { return NAME; }
+
+    @Override
+    public void produceArtifact(NodeAgentContext ctx, String configId, ServiceDumpReport.DumpOptions options,
+                                UnixPath resultDirectoryInNode) throws IOException {
+        int pid = findVespaServicePid(ctx, configId);
+        int seconds = (int) (duration(ctx, options, 30.0));
+        UnixPath outputFile = resultDirectoryInNode.resolve("recording.jfr");
+        List<String> startCommand = List.of("jcmd", Integer.toString(pid), "JFR.start", "name=host-admin",
+                "path-to-gc-roots=true", "settings=profile", "filename=" + outputFile, "duration=" + seconds + "s");
+        executeCommand(ctx, startCommand, true);
+        sleeper.sleep(Duration.ofSeconds(seconds).plusSeconds(1));
+        int maxRetries = 10;
+        List<String> checkCommand = List.of("jcmd", Integer.toString(pid), "JFR.check", "name=host-admin");
+        for (int i = 0; i < maxRetries; i++) {
+            boolean stillRunning = executeCommand(ctx, checkCommand, true).getOutputLines().stream()
+                    .anyMatch(l -> l.contains("name=host-admin") && l.contains("running"));
+            if (!stillRunning) return;
+            sleeper.sleep(Duration.ofSeconds(1));
+        }
+        throw new IOException("Failed to wait for JFR dump to complete after " + maxRetries + " retries");
+    }
+}
diff --git a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/maintenance/servicedump/VespaServiceDumperImpl.java b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/maintenance/servicedump/VespaServiceDumperImpl.java
index 83e661ec0fd..21964473cf1 100644
--- a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/maintenance/servicedump/VespaServiceDumperImpl.java
+++ b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/maintenance/servicedump/VespaServiceDumperImpl.java
@@ -1,6 +1,7 @@
 // Copyright Verizon Media. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 package com.yahoo.vespa.hosted.node.admin.maintenance.servicedump;
 
+import com.yahoo.concurrent.Sleeper;
 import com.yahoo.text.Lowercase;
 import com.yahoo.vespa.hosted.node.admin.configserver.noderepository.NodeAttributes;
 import com.yahoo.vespa.hosted.node.admin.configserver.noderepository.NodeRepository;
@@ -42,19 +43,20 @@ public class VespaServiceDumperImpl implements VespaServiceDumper {
     private final Map<String, ArtifactProducer> artifactProducers;
 
     public VespaServiceDumperImpl(ContainerOperations container, SyncClient syncClient, NodeRepository nodeRepository) {
-        this(container, syncClient, nodeRepository, Clock.systemUTC());
+        this(container, syncClient, nodeRepository, Clock.systemUTC(), Sleeper.DEFAULT);
     }
 
     // For unit testing
     VespaServiceDumperImpl(ContainerOperations container, SyncClient syncClient, NodeRepository nodeRepository,
-                           Clock clock) {
+                           Clock clock, Sleeper sleeper) {
         this.container = container;
         this.syncClient = syncClient;
         this.nodeRepository = nodeRepository;
         this.clock = clock;
         List<AbstractProducer> producers = List.of(
                 new JvmDumpProducer(container),
-                new PerfReportProducer(container));
+                new PerfReportProducer(container),
+                new JavaFlightRecorder(container, sleeper));
         this.artifactProducers = producers.stream()
                 .collect(Collectors.toMap(ArtifactProducer::name, Function.identity()));
     }
diff --git a/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/maintenance/servicedump/VespaServiceDumperImplTest.java b/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/maintenance/servicedump/VespaServiceDumperImplTest.java
index 3b512130608..c42316774c2 100644
--- a/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/maintenance/servicedump/VespaServiceDumperImplTest.java
+++ b/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/maintenance/servicedump/VespaServiceDumperImplTest.java
@@ -1,5 +1,6 @@
 package com.yahoo.vespa.hosted.node.admin.maintenance.servicedump;// Copyright Verizon Media. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 
+import com.yahoo.concurrent.Sleeper;
 import com.yahoo.test.ManualClock;
 import com.yahoo.vespa.hosted.node.admin.configserver.noderepository.NodeSpec;
 import com.yahoo.vespa.hosted.node.admin.configserver.noderepository.NodeState;
@@ -73,7 +74,7 @@ class VespaServiceDumperImplTest {
         NodeSpec initialSpec = createNodeSpecWithDumpRequest(nodeRepository, JvmDumpProducer.NAME, null);
 
         // Create dumper and invoke tested method
-        VespaServiceDumper reporter = new VespaServiceDumperImpl(operations, syncClient, nodeRepository, clock);
+        VespaServiceDumper reporter = new VespaServiceDumperImpl(operations, syncClient, nodeRepository, clock, Sleeper.NOOP);
         NodeAgentContextImpl context = new NodeAgentContextImpl.Builder(initialSpec)
                 .fileSystem(fileSystem)
                 .build();
@@ -106,7 +107,7 @@ class VespaServiceDumperImplTest {
         ManualClock clock = new ManualClock(Instant.ofEpochMilli(1600001000000L));
         NodeSpec nodeSpec = createNodeSpecWithDumpRequest(nodeRepository, PerfReportProducer.NAME, new ServiceDumpReport.DumpOptions(true, 45.0));
 
-        VespaServiceDumper reporter = new VespaServiceDumperImpl(operations, syncClient, nodeRepository, clock);
+        VespaServiceDumper reporter = new VespaServiceDumperImpl(operations, syncClient, nodeRepository, clock, Sleeper.NOOP);
         NodeAgentContextImpl context = new NodeAgentContextImpl.Builder(nodeSpec)
                 .fileSystem(fileSystem)
                 .build();
@@ -122,6 +123,34 @@ class VespaServiceDumperImplTest {
                         " > /opt/vespa/tmp/vespa-service-dump/perf-report/perf-report.txt");
     }
 
+    @Test
+    void invokes_jcmd_commands_when_creating_jfr_recording() {
+        // Setup mocks
+        ContainerOperations operations = mock(ContainerOperations.class);
+        when(operations.executeCommandInContainerAsRoot(any(), any()))
+                .thenReturn(new CommandResult(null, 0, "12345"))
+                .thenReturn(new CommandResult(null, 0, "ok"))
+                .thenReturn(new CommandResult(null, 0, "name=host-admin success"));
+        SyncClient syncClient = createSyncClientMock();
+        NodeRepoMock nodeRepository = new NodeRepoMock();
+        ManualClock clock = new ManualClock(Instant.ofEpochMilli(1600001000000L));
+        NodeSpec nodeSpec = createNodeSpecWithDumpRequest(
+                nodeRepository, JavaFlightRecorder.NAME, new ServiceDumpReport.DumpOptions(null, null));
+
+        VespaServiceDumper reporter = new VespaServiceDumperImpl(operations, syncClient, nodeRepository, clock, Sleeper.NOOP);
+        NodeAgentContextImpl context = new NodeAgentContextImpl.Builder(nodeSpec)
+                .fileSystem(fileSystem)
+                .build();
+        reporter.processServiceDumpRequest(context);
+
+        verify(operations).executeCommandInContainerAsRoot(
+                context, "/opt/vespa/libexec/vespa/find-pid", "default/container.1");
+        verify(operations).executeCommandInContainerAsRoot(
+                context, "jcmd", "12345", "JFR.start", "name=host-admin", "path-to-gc-roots=true", "settings=profile",
+                "filename=/opt/vespa/tmp/vespa-service-dump/jfr-recording/recording.jfr", "duration=30s");
+        verify(operations).executeCommandInContainerAsRoot(context, "jcmd", "12345", "JFR.check", "name=host-admin");
+    }
+
     private static NodeSpec createNodeSpecWithDumpRequest(NodeRepoMock repository, String artifactName, ServiceDumpReport.DumpOptions options) {
         ServiceDumpReport request = ServiceDumpReport.createRequestReport(
                 Instant.ofEpochMilli(1600000000000L), null, "default/container.1", List.of(artifactName), options);