You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@samza.apache.org by "Shanthoosh Venkataraman (JIRA)" <ji...@apache.org> on 2017/05/30 23:19:04 UTC
[jira] [Updated] (SAMZA-1316) Request for tryBecomeLeader(time,
TimeUnit) api in LeaderElector
[ https://issues.apache.org/jira/browse/SAMZA-1316?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Shanthoosh Venkataraman updated SAMZA-1316:
-------------------------------------------
Description:
tryBecomeLeader() api in LeaderElector should take in retry mechanism ( (time, TimeUnit) or noOfRetriesOnFailure as parameter) and should not try repeatedly(recursively) and fail on irrecoverable exceptions after fixed retries. Clients should've tunable configuration to control retries.
One more caveat is retries are programmed to happen after fixed amount of time(1000 ms).
Should use ExponentialBackOff like in SystemConsumer.fetchMetadata().
When there're execution environment problems(zookeeper death/network partitioning), a stream app can never become leader. We should handle these extreme failure scenarios gracefully.
was:
tryBecomeLeader() api in LeaderElector should take in retry mechanism ( (time, TimeUnit) or noOfRetriesOnFailure as parameter) and should not try repeatedly(recursively) and fail on irrecoverable exceptions after fixed retries. Clients should've tunable configuration to control retries.
One more caveat is retries are programmed to happen after fixed amount of time(1000 ms).
Should use ExponentialBackOff like in SystemConsumer.fetchMetadata().
When there're execution environmental problems(zookeeper death/network partitioning), a stream app can never become leader. We should handle these extreme failure scenarios gracefully.
> Request for tryBecomeLeader(time, TimeUnit) api in LeaderElector
> -----------------------------------------------------------------
>
> Key: SAMZA-1316
> URL: https://issues.apache.org/jira/browse/SAMZA-1316
> Project: Samza
> Issue Type: Improvement
> Reporter: Shanthoosh Venkataraman
> Priority: Minor
>
> tryBecomeLeader() api in LeaderElector should take in retry mechanism ( (time, TimeUnit) or noOfRetriesOnFailure as parameter) and should not try repeatedly(recursively) and fail on irrecoverable exceptions after fixed retries. Clients should've tunable configuration to control retries.
> One more caveat is retries are programmed to happen after fixed amount of time(1000 ms).
> Should use ExponentialBackOff like in SystemConsumer.fetchMetadata().
> When there're execution environment problems(zookeeper death/network partitioning), a stream app can never become leader. We should handle these extreme failure scenarios gracefully.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)