Support configurable two-phase state transitions in cluster controller

1 files changed, 16 insertions, 3 deletions

diff --git a/clustercontroller-core/src/test/java/com/yahoo/vespa/clustercontroller/core/StateChangeTest.java b/clustercontroller-core/src/test/java/com/yahoo/vespa/clustercontroller/core/StateChangeTest.java
index 32de3591f2d..85106ce7e3c 100644
--- a/clustercontroller-core/src/test/java/com/yahoo/vespa/clustercontroller/core/StateChangeTest.java
+++ b/clustercontroller-core/src/test/java/com/yahoo/vespa/clustercontroller/core/StateChangeTest.java
@@ -171,6 +171,9 @@ public class StateChangeTest extends FleetControllerTest {
         options.nodeStateRequestTimeoutMS = 60 * 60 * 1000;
         options.minTimeBetweenNewSystemStates = 0;
         options.maxInitProgressTime = 50000;
+        // This test makes very specific assumptions about the amount of work done in a single tick.
+        // Two-phase cluster state activation changes this quite a bit, so disable it. At least for now.
+        options.enableTwoPhaseClusterStateActivation = false;
 
         initialize(options);
 
@@ -1035,7 +1038,7 @@ public class StateChangeTest extends FleetControllerTest {
         // Assert that the failed node has not acknowledged the latest version.
         // (The version may still be larger than versionBeforeChange if the fleet controller sends a
         // "stable system" update without timestamps in the meantime
-        assertTrue(fleetController.getCluster().getNodeInfo(nodes.get(1).getNode()).getSystemStateVersionAcknowledged() < versionAfterChange);
+        assertTrue(fleetController.getCluster().getNodeInfo(nodes.get(1).getNode()).getClusterStateVersionBundleAcknowledged() < versionAfterChange);
 
         // Ensure non-concurrent access to getNewestSystemStateVersionSent
         synchronized(timer) {
@@ -1343,11 +1346,19 @@ public class StateChangeTest extends FleetControllerTest {
 
         void sendAllDeferredDistributorClusterStateAcks() throws Exception {
             communicator.sendAllDeferredDistributorClusterStateAcks();
-            ctrl.tick();
+            ctrl.tick(); // Process cluster state bundle ACKs
+            if (ctrl.getOptions().enableTwoPhaseClusterStateActivation) {
+                ctrl.tick(); // Send activations
+                ctrl.tick(); // Process activation ACKs
+            }
         }
 
         void processScheduledTask() throws Exception {
             ctrl.tick(); // Cluster state recompute iteration and send
+            if (ctrl.getOptions().enableTwoPhaseClusterStateActivation) {
+                ctrl.tick(); // Send activations
+                ctrl.tick(); // Process activation ACKs
+            }
             ctrl.tick(); // Iff ACKs were received, process version dependent task(s)
         }
 
@@ -1440,7 +1451,7 @@ public class StateChangeTest extends FleetControllerTest {
 
     @Test
     public void no_op_synchronous_remote_task_waits_until_current_state_is_acked() throws Exception {
-         RemoteTaskFixture fixture = createFixtureWith(optionsWithZeroTransitionTime());
+        RemoteTaskFixture fixture = createFixtureWith(optionsWithZeroTransitionTime());
 
         communicator.setShouldDeferDistributorClusterStateAcks(true);
         fixture.markStorageNodeDown(0);
@@ -1523,7 +1534,9 @@ public class StateChangeTest extends FleetControllerTest {
     public void cluster_state_ack_is_not_dependent_on_state_send_grace_period() throws Exception {
         FleetControllerOptions options = defaultOptions();
         options.minTimeBetweenNewSystemStates = 10_000;
+        //options.enableTwoPhaseClusterStateActivation = false; ////
         RemoteTaskFixture fixture = createFixtureWith(options);
+
         // Have to increment timer here to be able to send state generated by the scheduled task
         timer.advanceTime(10_000);