zpinto commented on code in PR #2661:
URL: https://github.com/apache/helix/pull/2661#discussion_r1367174009


##########
helix-core/src/main/java/org/apache/helix/manager/zk/ZKHelixAdmin.java:
##########
@@ -410,13 +472,260 @@ public ZNRecord update(ZNRecord currentData) {
 
   @Override
   public boolean isEvacuateFinished(String clusterName, String instanceName) {
-    return !instanceHasCurrentSateOrMessage(clusterName, instanceName) && 
(getInstanceConfig(clusterName,
+    return !instanceHasCurrentStateOrMessage(clusterName, instanceName) && 
(getInstanceConfig(
+        clusterName,
         
instanceName).getInstanceOperation().equals(InstanceConstants.InstanceOperation.EVACUATE.name()));
   }
 
+  /**
+   * Find the instance that this passed instance is swapping with. If the 
passed instance has
+   * SWAP_OUT instanceOperation, then find the corresponding instance that has 
SWAP_IN
+   * instanceOperation. If the passed instance has SWAP_IN instanceOperation, 
then find the
+   * corresponding instance that has SWAP_OUT instanceOperation.
+   *
+   * @param clusterName    The cluster name
+   * @param instanceConfig The instance to find the swap instance for
+   * @return The swap instance if found, null otherwise.
+   */
+  @Nullable
+  private InstanceConfig findMatchingSwapInstance(String clusterName,
+      InstanceConfig instanceConfig) {
+    String logicalIdKey =
+        
ClusterTopologyConfig.createFromClusterConfig(_configAccessor.getClusterConfig(clusterName))
+            .getEndNodeType();
+
+    for (String potentialSwappingInstance : getConfigKeys(
+        new 
HelixConfigScopeBuilder(HelixConfigScope.ConfigScopeProperty.PARTICIPANT,
+            clusterName).build())) {
+      InstanceConfig potentialSwappingInstanceConfig =
+          getInstanceConfig(clusterName, potentialSwappingInstance);
+
+      // Return if there is a matching Instance with the same logicalId and 
opposite InstanceOperation swap operation.
+      if (potentialSwappingInstanceConfig.getLogicalId(logicalIdKey)
+          .equals(instanceConfig.getLogicalId(logicalIdKey)) && (
+          instanceConfig.getInstanceOperation()
+              .equals(InstanceConstants.InstanceOperation.SWAP_IN.name())
+              && potentialSwappingInstanceConfig.getInstanceOperation()
+              .equals(InstanceConstants.InstanceOperation.SWAP_OUT.name())) || 
(
+          instanceConfig.getInstanceOperation()
+              .equals(InstanceConstants.InstanceOperation.SWAP_OUT.name())
+              && potentialSwappingInstanceConfig.getInstanceOperation()
+              .equals(InstanceConstants.InstanceOperation.SWAP_IN.name()))) {
+        return potentialSwappingInstanceConfig;
+      }
+    }
+
+    return null;
+  }
+
+  /**
+   * Check to see if swapping between two instances is ready to be completed. 
Checks: 1. Both
+   * instances must be alive. 2. Both instances must only have one session and 
not be carrying over
+   * from a previous session. 3. Both instances must have no pending messages. 
4. Both instances
+   * cannot have partitions in the ERROR state 4. SwapIn instance must have 
correct state for all
+   * partitions that are currently assigned to the SwapOut instance.
+   * TODO: We may want to make this a public API in the future.

Review Comment:
   I don't think this will work in every case because the ideal state is not 
persisted in ZK without PERSIST_BEST_POSSIBLE_STATE set to true. I think that 
only checking current state and no pending messages is sufficient because if 
the replica from the ideal state is not already being bootstrapped on SWAP_OUT 
node then it will be sent directly to SWAP_IN node right after the swap 
completes.
   
   If load balance is causing the new replica to be assigned to SWAP_OUT node 
then if the ST message never gets to the SWAP_OUT node before completing the 
SWAP it is okay because once we switch to the SWAP_IN node we have not dropped 
the replica from its old location yet.
   
   This means that it will still be N -> N+1 -> N for the newly assigned 
replica.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to