anmolnar commented on PR #5785:
URL: https://github.com/apache/hbase/pull/5785#issuecomment-2070253681

   > Mind taking a look at 
[HBASE-28521](https://issues.apache.org/jira/browse/HBASE-28521)? In general, 
we want to avoid leaking the internal zookeeper, so for replication, we want to 
use the standard connection registry API to connect to peer cluster, for 
compatibilitu, we will use ZKConnectionRegistry by default.
   > 
   > But here after this change, we force hbase replication endpoint to must 
use zookeeper, which is not very good. I suppose we should find another way to 
customize the ZKWatcher creation. IIRC @anmolnar has done something related to 
this area, i.e, how to set zookeeper configurations in hbase configuration, 
with a special prefix?
   > 
   > Thanks.
   
   The big difference between using `ZKClientConfig` or not using it is that 
passing the client config enables using different ZooKeeper client 
configurations in the same JVM process. This is especially useful in a 
replication scenario where you need to maintain two separate ZK connection 
potentially with different TLS settings for instance. I believe this is the 
improvement that @BukrosSzabolcs implemented here.
   
   Using the Connection Registry is also a very good improvement, but at the 
same time we should focus the above as well.
   
   My patch added the ability to set ZK system properties via `hbase-site.xml`, 
but that doesn't solve the problem of multiple ZK connections. I haven't looked 
into `ZKConnectionRegistry` yet, but we might want to introduce using custom 
`ZKClientConfig` in there instead of here.


-- 
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]

Reply via email to