Out of capacity retrying through triggerReadyJobs

2 files changed, 10 insertions, 6 deletions

diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/DeploymentTrigger.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/DeploymentTrigger.java
index cf4071b9035..6c954191075 100644
--- a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/DeploymentTrigger.java
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/DeploymentTrigger.java
@@ -113,8 +113,10 @@ public class DeploymentTrigger {
                 for (JobType job : jobs)
                      application = trigger(new Triggering(application, job, false, report.jobType().jobName() + " completed"), jobs, false);
             }
-            else if (retryBecauseOutOfCapacity(application, report.jobType()))
-                application = trigger(new Triggering(application, report.jobType(), true, "Retrying on out of capacity"), Collections.emptySet(), false);
+            else if (retryBecauseOutOfCapacity(application, report.jobType())) {
+                triggerReadyJobs(application);
+                return; // Don't overwrite below.
+            }
             else if (retryBecauseNewFailure(application, report.jobType())) {
                 triggerReadyJobs(application);
                 return; // Don't overwrite below.
@@ -166,8 +168,10 @@ public class DeploymentTrigger {
             List<JobType> nextJobs = order.nextAfter(jobType, application);
             for (JobType nextJobType : nextJobs) {
                 JobStatus nextStatus = application.deploymentJobs().jobStatus().get(nextJobType);
-                if (changesAvailable(application, jobStatus, nextStatus) || nextStatus.isHanging(jobTimeoutLimit()))
-                    application = trigger(new Triggering(application, nextJobType, false, "Available change in " + jobType.jobName()), nextJobs, false);
+                if (changesAvailable(application, jobStatus, nextStatus) || nextStatus.isHanging(jobTimeoutLimit())) {
+                    boolean isRetry = nextStatus != null && nextStatus.jobError().filter(JobError.outOfCapacity::equals).isPresent();
+                    application = trigger(new Triggering(application, nextJobType, isRetry, isRetry ? "Retrying on out of capacity" : "Available change in " + jobType.jobName()), nextJobs, false);
+                }
             }
             applications().store(application);
         }
diff --git a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/ControllerTest.java b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/ControllerTest.java
index 594a9c0da44..50bf20c4405 100644
--- a/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/ControllerTest.java
+++ b/controller-server/src/test/java/com/yahoo/vespa/hosted/controller/ControllerTest.java
@@ -470,14 +470,14 @@ public class ControllerTest {
         assertEquals(2, deploymentQueue.jobs().size());
 
         // app1: 4 hours pass in total, staging-test job for app1 is re-queued by periodic trigger mechanism and added at the
-        // back of the queue
+        // front of the queue
         tester.clock().advance(Duration.ofHours(3));
         tester.clock().advance(Duration.ofMinutes(50));
         tester.readyJobTrigger().maintain();
 
+        assertEquals(Collections.singletonList(new BuildService.BuildJob(project1, stagingTest.jobName())), deploymentQueue.takeJobsToRun());
         assertEquals(Collections.singletonList(new BuildService.BuildJob(project2, stagingTest.jobName())), deploymentQueue.takeJobsToRun());
         assertEquals(Collections.singletonList(new BuildService.BuildJob(project3, stagingTest.jobName())), deploymentQueue.takeJobsToRun());
-        assertEquals(Collections.singletonList(new BuildService.BuildJob(project1, stagingTest.jobName())), deploymentQueue.takeJobsToRun());
         assertEquals(Collections.emptyList(), deploymentQueue.takeJobsToRun());
     }