Fix container restart after node-admin restart

2 files changed, 23 insertions, 2 deletions

diff --git a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImpl.java b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImpl.java
index fcd0cb8b8af..49375836a1c 100644
--- a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImpl.java
+++ b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImpl.java
@@ -82,7 +82,7 @@ public class NodeAgentImpl implements NodeAgent {
         RUNNING_HOWEVER_RESUME_SCRIPT_NOT_RUN,
         RUNNING
     }
-    private ContainerState containerState = ABSENT;
+    private ContainerState containerState = RUNNING_HOWEVER_RESUME_SCRIPT_NOT_RUN;
 
     // The attributes of the last successful node repo attribute update for this node. Used to avoid redundant calls.
     private NodeAttributes lastAttributesSet = null;
@@ -119,7 +119,6 @@ public class NodeAgentImpl implements NodeAgent {
                 .ifPresent(container -> {
                     vespaVersion = dockerOperations.getVespaVersion(container.name);
                     lastCpuMetric = new CpuUsageReporter(container.created);
-                    containerState = RUNNING_HOWEVER_RESUME_SCRIPT_NOT_RUN;
                 });
     }
 
diff --git a/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImplTest.java b/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImplTest.java
index f59ec0d1250..1855c2d96c2 100644
--- a/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImplTest.java
+++ b/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/nodeagent/NodeAgentImplTest.java
@@ -24,6 +24,7 @@ import org.junit.Test;
 import org.mockito.InOrder;
 
 import java.io.File;
+import java.io.IOException;
 import java.nio.file.Files;
 import java.time.Duration;
 import java.util.Collections;
@@ -360,6 +361,27 @@ public class NodeAgentImplTest {
     }
 
     @Test
+    public void testRestartDeadContainerAfterNodeAdminRestart() throws IOException {
+        final ContainerNodeSpec nodeSpec = nodeSpecBuilder
+                .wantedDockerImage(dockerImage)
+                .nodeState(Node.State.active)
+                .vespaVersion(vespaVersion)
+                .build();
+
+        NodeAgentImpl nodeAgent = makeNodeAgent(dockerImage, false);
+
+        when(nodeRepository.getContainerNodeSpec(eq(hostName))).thenReturn(Optional.of(nodeSpec));
+        when(dockerOperations.shouldScheduleDownloadOfImage(eq(dockerImage))).thenReturn(false);
+        when(pathResolver.getApplicationStoragePathForNodeAdmin()).thenReturn(Files.createTempDirectory("foo"));
+
+        nodeAgent.tick();
+
+        verify(dockerOperations, times(1)).removeContainer(any());
+        verify(dockerOperations, times(1)).startContainer(eq(containerName), eq(nodeSpec));
+
+    }
+
+    @Test
     public void resumeProgramRunsUntilSuccess() throws Exception {
         final long restartGeneration = 1;
         final ContainerNodeSpec nodeSpec = nodeSpecBuilder