[GitHub] [pulsar] rdhabalia commented on a change in pull request #5604: [pulsar-broker] recover zk-badversion while updating cursor metadata

[GitBox <gi...@apache.org>]

rdhabalia commented on a change in pull request #5604: [pulsar-broker] recover zk-badversion while updating cursor metadata
URL: https://github.com/apache/pulsar/pull/5604#discussion_r346036936
 
 

 ##########
 File path: managed-ledger/src/main/java/org/apache/bookkeeper/mledger/impl/ManagedCursorImpl.java
 ##########
 @@ -2033,6 +2033,30 @@ public void operationComplete(Void result, Stat stat) {
 
                     @Override
                     public void operationFailed(MetaStoreException e) {
+                        if (e instanceof MetaStoreException.BadVersionException) {
+                            log.warn("[{}] Failed to update cursor metadata for {} due to version conflict {}",
+                                    ledger.name, name, e.getMessage());
+                            // it means previous owner of the ml might have updated the version incorrectly. So, check
+                            // the ownership and refresh the version again.
+                            if (ledger.mlOwnershipChecker != null && ledger.mlOwnershipChecker.get()) {
+                                ledger.getStore().asyncGetCursorInfo(ledger.getName(), name,
+                                        new MetaStoreCallback<ManagedCursorInfo>() {
+                                            @Override
+                                            public void operationComplete(ManagedCursorInfo info, Stat stat) {
+                                                cursorLedgerStat = stat;
 
 Review comment:
   Yes, I have thought about it earlier but it will be tricky because of race condition:
    if we retry the same function `persistPositionMetaStore` or any other functions with same metadata then by that time markDelete position could have been changed. also, metadata contains multiple info eg: ledgerId, markDelete-position and its properties, etc.. and cursor only maintains latest markDeletePosition in memory.. so, it's tricky to update metadata with all latest information in a synchronized manner. So, broker can rely on next acked message as this exception occurs rarely. 

----------------------------------------------------------------
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.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services