[
https://issues.apache.org/jira/browse/BOOKKEEPER-704?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14064955#comment-14064955
]
Ivan Kelly commented on BOOKKEEPER-704:
---------------------------------------
{quote}
I think we basically just need be able to retry zookeeper calls
{quote}
Actually, I don't think this is what we need. Retrying should be left up to the
user of the zookeeper connection, as it's the client who can make the decision
whether an operation should be retried or not. What we need is a _reconnecting_
zookeeper client. The problem, as I see it is that the zookeeper client is tied
to a single zookeeper session. This is needed for the things that depend on
sessions, like ephemerals and stuff. There's a lot of stuff that doesn't
require sessions though. setData, getData, create, delete don't need session
stuff.
What I've often seen happen is that there is one zookeeper client opened at
process start and this passed all around the place. This binds the process
lifetime to the zookeeper session lifetime. So when the zookeeper session goes
away, the process has to, which is bad. However, if we had a zookeeper client
which didn't have ephemerals we could just reconnect as needed, like we do with
PerChannelBookieClient. If a session goes away, what was pending on that
session would fail, but the client itself could be reused. We should be able to
deal with these transient errors in any case.
> reconnectable zookeeper client wrapper
> --------------------------------------
>
> Key: BOOKKEEPER-704
> URL: https://issues.apache.org/jira/browse/BOOKKEEPER-704
> Project: Bookkeeper
> Issue Type: Sub-task
> Components: bookkeeper-client, bookkeeper-server
> Reporter: Sijie Guo
> Assignee: Sijie Guo
> Fix For: 4.4.0
>
> Attachments: BOOKKEEPER-704.diff
>
>
> create a reconnectable zookkeeper client wrapper to handle session expire
> event.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
