Merge pull request #24606 from vespa-engine/jonmv/sleep-5-mins-on-parent-host-not-ready-first-timev8.75.16

Sleep 5 minutes on first parent host not ready

2 files changed, 5 insertions, 1 deletions

diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunner.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunner.java
index 7a459e4f29f..e8c92d3e3f6 100644
--- a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunner.java
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunner.java
@@ -179,6 +179,7 @@ public class InternalStepRunner implements StepRunner {
                       controller.jobController().run(id)
                                 .stepInfo(setTheStage ? deployInitialReal : deployReal).get()
                                 .startTime().get(),
+                      id,
                       logger)
                 .filter(result -> {
                     // If no tester cert, or deployment failed, propagate original result.
@@ -200,10 +201,11 @@ public class InternalStepRunner implements StepRunner {
                       controller.jobController().run(id)
                                 .stepInfo(deployTester).get()
                                 .startTime().get(),
+                      id,
                       logger);
     }
 
-    private Optional<RunStatus> deploy(Supplier<DeploymentResult> deployment, Instant startTime, DualLogger logger) {
+    private Optional<RunStatus> deploy(Supplier<DeploymentResult> deployment, Instant startTime, RunId id, DualLogger logger) {
         try {
             DeploymentResult result = deployment.get();
                 logger.logAll(result.log().stream()
@@ -246,6 +248,7 @@ public class InternalStepRunner implements StepRunner {
                 }
                 case LOAD_BALANCER_NOT_READY, PARENT_HOST_NOT_READY -> {
                     logger.log(e.message()); // Consider splitting these messages in summary and details, on config server.
+                    controller.jobController().locked(id, run -> run.sleepingUntil(startTime.plusSeconds(300)));
                     return result;
                 }
                 case NODE_ALLOCATION_FAILURE -> {
diff --git a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunnerTest.java b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunnerTest.java
index 7687a1561a4..9bf762d2f99 100644
--- a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunnerTest.java
+++ b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunnerTest.java
@@ -516,6 +516,7 @@ public class InternalStepRunnerTest {
         assertEquals(oldTrusted, tester.configServer().application(app.instanceId(), id.type().zone()).get().applicationPackage().trustedCertificates());
 
         tester.configServer().throwOnNextPrepare(null);
+        tester.clock().advance(Duration.ofSeconds(300));
         tester.runner().run();
         assertEquals(succeeded, tester.jobs().run(id).stepStatuses().get(Step.deployTester));
         assertEquals(succeeded, tester.jobs().run(id).stepStatuses().get(Step.deployReal));