[ 
https://issues.apache.org/jira/browse/SENTRY-664?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Arun Suresh updated SENTRY-664:
-------------------------------
    Attachment: SENTRY-664.1.patch

Attaching trivial patch.

I have also included some log message modifications that helped me debug the 
issue..

*NOTE*
The fix is pretty trivial (It was caused due the fact that the lastSentSeqNum 
in the MetaStorePlugin was initialized to -1, while it should have been equal 
to the current value of seqNum). Unfortunately writing a testcase is hard.. 
more so because the "miniDfs" tools used in the testCase has a bug in the 
version of hadoop sentry depends on. I have put in the additional tests but 
have commented them out and have placed a TODO to remove the comment once we 
have a working miniDFS

> After Namenode is restarted, Path updates remain unsynched
> ----------------------------------------------------------
>
>                 Key: SENTRY-664
>                 URL: https://issues.apache.org/jira/browse/SENTRY-664
>             Project: Sentry
>          Issue Type: Bug
>    Affects Versions: 1.4.0
>            Reporter: Arun Suresh
>            Assignee: Arun Suresh
>              Labels: sentry-hdfs
>         Attachments: SENTRY-664.1.patch
>
>
> Path updates remain un-synced after a Namenode restart.
> Ideally, after a restart, the Namenode receives both the Path and Permission 
> updates from sentry service. These updates are ordered by a sequence Id. Also 
> the first Path/Permission update is supposed to be a FULL update. 
> From observing the logs, It looks like after namenode restart, the sequence 
> of PathsUpdate objects do not start with a FULL update (The 
> PermissionsUpdates are fine though). Thus the namenode remains unsynched with 
> respect to path information.
> A workaround is to restart the metastore after the namenode has restarted. 
> This will ensure HMS will send a FULL update to Sentry. This will cause 
> Sentry service will flush its existing update queue, so the next time the 
> namenode asks for an update, this FULL update will be sent. 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to