[ 
https://issues.apache.org/jira/browse/SENTRY-1477?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15511283#comment-15511283
 ] 

Li Li commented on SENTRY-1477:
-------------------------------

There are some retry logic implemented in 
[SENTRY-1411|https://issues.apache.org/jira/browse/SENTRY-1411]. Basically, 
PoolClientInvocationHandler maintains a freshestEndpointIdx var shared by all 
threads for a single client. All threads will connect to the same endpoint, and 
increment the freshestEndpointIdx and retry with max retry times if failed to 
connect. But it always ends up with all threads (maybe clients as well) 
connecting to the same sentry server.
In this jira, I am gonna apply the same retry logic as in SENTRY-1411. If we 
think it is better do some load balance to avoid all nodes connecting to the 
same sentry server, we can create a new jira for it.


> Sentry clients should retry with another server when they get connection 
> errors
> -------------------------------------------------------------------------------
>
>                 Key: SENTRY-1477
>                 URL: https://issues.apache.org/jira/browse/SENTRY-1477
>             Project: Sentry
>          Issue Type: Sub-task
>          Components: Core
>    Affects Versions: sentry-ha-redesign
>            Reporter: Alexander Kolbasov
>            Assignee: Li Li
>             Fix For: sentry-ha-redesign
>
>
> When Sentry clients get a connection error when talking to a server they 
> should re-try connection with another server. This should be normal client 
> behavior when one of the servers goes down.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to