Merge pull request #18321 from vespa-engine/mpolden/less-lock-contention

Reduce lock contention for infra deployment

2 files changed, 5 insertions, 3 deletions

diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MaintenanceDeployment.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MaintenanceDeployment.java
index 3b16ecbcaa9..892372f27e7 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MaintenanceDeployment.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MaintenanceDeployment.java
@@ -102,11 +102,13 @@ class MaintenanceDeployment implements Closeable {
     }
 
     private Optional<Mutex> tryLock(ApplicationId application, NodeRepository nodeRepository) {
+        Duration timeout = Duration.ofSeconds(3);
         try {
             // Use a short lock to avoid interfering with change deployments
-            return Optional.of(nodeRepository.nodes().lock(application, Duration.ofSeconds(1)));
+            return Optional.of(nodeRepository.nodes().lock(application, timeout));
         }
         catch (ApplicationLockException e) {
+            log.log(Level.WARNING, () -> "Could not lock " + application + " for maintenance deployment within " + timeout);
             return Optional.empty();
         }
     }
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java
index 96373bd764f..79d6fbfbdcd 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java
@@ -127,7 +127,7 @@ public class NodeRepositoryMaintenance extends AbstractComponent {
             dynamicProvisionerInterval = Duration.ofMinutes(5);
             failedExpirerInterval = Duration.ofMinutes(10);
             failGrace = Duration.ofMinutes(30);
-            infrastructureProvisionInterval = Duration.ofMinutes(1);
+            infrastructureProvisionInterval = Duration.ofMinutes(3);
             loadBalancerExpirerInterval = Duration.ofMinutes(5);
             metricsInterval = Duration.ofMinutes(1);
             nodeFailerInterval = Duration.ofMinutes(15);
@@ -152,7 +152,7 @@ public class NodeRepositoryMaintenance extends AbstractComponent {
 
             if (zone.environment().isProduction() && ! zone.system().isCd()) {
                 inactiveExpiry = Duration.ofHours(4); // enough time for the application owner to discover and redeploy
-                retiredInterval = Duration.ofMinutes(30);
+                retiredInterval = Duration.ofMinutes(15);
                 dirtyExpiry = Duration.ofHours(2); // enough time to clean the node
                 retiredExpiry = Duration.ofDays(4); // give up migrating data after 4 days
             } else {