[
https://issues.apache.org/jira/browse/TAJO-1143?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14235235#comment-14235235
]
ASF GitHub Bot commented on TAJO-1143:
--------------------------------------
Github user hyunsik commented on the pull request:
https://github.com/apache/tajo/pull/280#issuecomment-65757536
I think that around ```checkAndInitializeSystemDirectories()``` method
would be the best place for the startup diagnosis of TajoMaster. Also,
serviceInit() would be the best place for TajoWorker. The constructor of
TajoClient would best.
Your patch already includes two rules: Hadoop runtime version check and
TajoConf verification. Both are necessary in TajoMaster and TajoWorker
components. Since Hadoop dependency will be removed from TajoClient (see
https://issues.apache.org/jira/browse/TAJO-1160), only TajoConf verification
would be necessary in TajoClient.
Cheers,
Hyunsik
> TajoMaster, TajoWorker, and TajoClient should have diagnosis phase at startup
> -----------------------------------------------------------------------------
>
> Key: TAJO-1143
> URL: https://issues.apache.org/jira/browse/TAJO-1143
> Project: Tajo
> Issue Type: Improvement
> Components: client, query master, tajo master
> Reporter: Hyunsik Choi
> Assignee: Jihun Kang
> Fix For: 0.9.1
>
>
> I propose that all cluster components (TajoMaster, TajoWorker, and
> TajoClient) in a Tajo cluster should have a diagnosis phase to eliminate or
> detect invalid situations prior to runtime query errors.
> For example, your query can cause some runtime exception due to wrong config
> after a query takes 2 hours. This situation is definitely not acceptable in
> production.
> I think that the diagnosis phase should check all configs, connectivities
> among cluster components, and status of workers.
> In detail, we need a diagnosis executor, extensible diagnosis rule interface,
> and its rules. Also, one of diagnosis rules would be TAJO-1114.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
