[ 
https://issues.apache.org/jira/browse/YARN-2035?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14104587#comment-14104587
 ] 

Zhijie Shen commented on YARN-2035:
-----------------------------------

[~jeagles], is the problematic scenario that NN and TimelineServer (TS) start 
around the same time? Therefore, while NN still stays in the safe mode, TS is 
trying to create a directory on it, result in SafeModeException.

In the patch, checking whether dir exists seems to be necessary. Moreover, 
shall we do something similar to that we did for MR job history server? See 
HistoryFileManager#serviceInit.
{code}
    long maxFSWaitTime = conf.getLong(
        JHAdminConfig.MR_HISTORY_MAX_START_WAIT_TIME,
        JHAdminConfig.DEFAULT_MR_HISTORY_MAX_START_WAIT_TIME);
    createHistoryDirs(new SystemClock(), 10 * 1000, maxFSWaitTime);
{code}
createHistoryDirs is going to retry dir creation until using up waiting time.

> FileSystemApplicationHistoryStore blocks RM and AHS while NN is in safemode
> ---------------------------------------------------------------------------
>
>                 Key: YARN-2035
>                 URL: https://issues.apache.org/jira/browse/YARN-2035
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>    Affects Versions: 2.4.1
>            Reporter: Jonathan Eagles
>            Assignee: Jonathan Eagles
>         Attachments: YARN-2035.patch
>
>
> Small bug that prevents ResourceManager and ApplicationHistoryService from 
> coming up while Namenode is in safemode.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Reply via email to