TisonKun commented on a change in pull request #11338: [FLINK-10052][ha]
Tolerate temporarily suspended ZooKeeper connections
URL: https://github.com/apache/flink/pull/11338#discussion_r390914444
##########
File path: docs/_includes/generated/expert_high_availability_zk_section.html
##########
@@ -14,6 +14,12 @@
<td>String</td>
<td>Defines the ACL (open|creator) to be configured on ZK node.
The configuration value can be set to “creator” if the ZooKeeper server
configuration has the “authProvider” property mapped to use
SASLAuthenticationProvider and the cluster is configured to run in secure mode
(Kerberos).</td>
</tr>
+ <tr>
+
<td><h5>high-availability.zookeeper.client.connection-loss-tolerate</h5></td>
+ <td style="word-wrap: break-word;">true</td>
+ <td>Boolean</td>
+ <td>Defines whether or not leader election tolerates connection
loss exception. Enable this option will improve the stability when ZK ensemble
in an unstable network environment. For technical details, see FLINK-10052.</td>
Review comment:
Curator beyond 3.0 will send a session expire event when it believes the
expire timeout elapsed. So briefly, it doesn't wait forever because it will
receive a `ConnectionState.LOST` event and the `LeaderLatch` revokes
leadership and re-contends. Anyway, it is a configuration in Curator scope.
Yes I'd prefer always turn on this feature. React to
`ConnectionState.SUSPENDED` often turns out to be the cause of instability. If
there is anything we gain from disable this feature, it will be
- Give up leadership fast, so that we don't request modification on
JobGraphStore(etc.) which will be later retried by Curator and successfully
modify without leadership. But even we disable it we don't solve the problem.
See also FLINK-10333. The general case is, user suffers from unstable zk
connection while leadership doesn't switch.
----------------------------------------------------------------
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:
[email protected]
With regards,
Apache Git Services
