[
https://issues.apache.org/jira/browse/OOZIE-1877?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14041389#comment-14041389
]
Robert Kanter commented on OOZIE-1877:
--------------------------------------
{quote}There is some time difference between when oozie comes up and when safe
mode is enabled. The workflows can fail within this timeframe as well.{quote}
We could change Oozie to startup in safe mode, and if everything checks out, it
would switch to normal mode (I believe the NameNode does something like this).
{quote}Has anyone tried Oozie safe mode?{quote}
I haven't tried safe mode, but I was looking at its code a little while back
for some reason. IIRC, it currently serves no purpose.
An advantage of having Oozie sit in safe mode instead of shutting down is that
it can keep checking automatically for HDFS to come up, or for the sharelib to
be available. When that happens, it can automatically switch to normal mode.
This seems easier and more transparent for the user. We can also have Oozie
print out something to the log about why its in safe mode. In safe mode, we
could have Oozie allow looking at jobs etc, which don't require HDFS, but you
wouldn't be able to run jobs (and we could make it say why).
My concern with having Oozie shut down is that anyone who starts up a cluster
in the "wrong" order will be wondering why Oozie doesn't start up and will have
to fix their script or procedure. With safe mode, it would just appear that
Oozie took a little longer than normal to become operational.
> Setting to fail oozie server startup in case of sharelib misconfiguration
> -------------------------------------------------------------------------
>
> Key: OOZIE-1877
> URL: https://issues.apache.org/jira/browse/OOZIE-1877
> Project: Oozie
> Issue Type: Sub-task
> Reporter: Purshotam Shah
> Assignee: Purshotam Shah
> Attachments: OOZIE-1877-V1.patch
>
>
> "OOZIE-1584 Setup sharelib using script and pickup latest(honor
> ship.launcher) and remove DFS dependency at startup" has removed sharelib
> dependency at startup.
> If DFS is down or sharelib is misconfigured. Server will start without
> loading sharelib, admin can issue sharelibupdate command to load sharelib.
> This is good, may not be acceptable in production. If sharelib is
> misconfigured then oozie server will come up without loading sharelib and all
> submitted hadoop job will fail.
> Better to have a property to shutdown oozie server, if sharelib
> initialization fails.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
