Duo Zhang commented on HBASE-16835:

Why do we need curator here rather than our own zk stuff?

Duo Zhang
As said above, if we decide to use watcher then the curator-recipes is very 
useful. And if we fetch data from zk every time, then our RecoverableZooKeeper 
does not support async operations. I could add the support for it but I think 
our zk stuff is too heavy to be used at client side, most logic is unnecessary 
for a client implementation as we do not store any data on zk at client, only 
read. And we need to be careful when modify the zk stuff as it will also be 
used at server side which is much more critical. The recovery and retry 
provided by curator is enough to be used at client side.

Michael Stack
Suggest you throw this paragraph and above zk back and forth into the new zk 
discussion issue. Its good stuff.

So Curator has an async API? If so, it is an acceptable justification to bring 
in that dependency.

> Revisit the zookeeper usage at client side
> ------------------------------------------
>                 Key: HBASE-16835
>                 URL: https://issues.apache.org/jira/browse/HBASE-16835
>             Project: HBase
>          Issue Type: Sub-task
>            Reporter: Duo Zhang
> Watcher or not.
> Curator or not.
> Keep connection or not.
> ...

This message was sent by Atlassian JIRA

Reply via email to