Merge pull request #24775 from vespa-engine/jonmv/less-agressive-retries-when-waiting-for-hosts

Retry less aggressively while waiting for host provisioning

2 files changed, 4 insertions, 2 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 e8c92d3e3f6..e6d793affff 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
@@ -248,7 +248,9 @@ 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)));
+                    Instant someTimeAfterStart = startTime.plusSeconds(450);
+                    Instant inALittleWhile = controller.clock().instant().plusSeconds(90);
+                    controller.jobController().locked(id, run -> run.sleepingUntil(someTimeAfterStart.isAfter(inALittleWhile) ? someTimeAfterStart : inALittleWhile));
                     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 9bf762d2f99..2f245ab9736 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
@@ -516,7 +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.clock().advance(Duration.ofSeconds(450));
         tester.runner().run();
         assertEquals(succeeded, tester.jobs().run(id).stepStatuses().get(Step.deployTester));
         assertEquals(succeeded, tester.jobs().run(id).stepStatuses().get(Step.deployReal));