[
https://issues.apache.org/jira/browse/HDFS-14914?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16955310#comment-16955310
]
Ayush Saxena commented on HDFS-14914:
-------------------------------------
{quote}What makes it worse is that when Routers talking to Observers, since
Router distinguishes StandbyException and RetriableException, it will keep
retry (default 3) times and then return to the client an RetriableException
{quote}
Observer is not supported as of now in RBF. It is still open, so I think, so
lets not get in Observer-RBF stuff until and unless we finalize it.
Regarding the 30 call stuff, That is not for just Observer, If ANN also throws
Exception, it would land up being 30. but that is configurable, I guess there
is Jira too regarding this, I don't exactly remember which but something
related that at router it doesn't wait before retry or something like this. I
will try finding it.
{quote}The current ObserverReadProxyProvider doesn't get affected since both
RetriableException and StandbyException will make it failover.
{quote}
Does this conclude, that this is not a problem in the present code? and this
change would not have any benefit in the present setup.
> Observer should throw StandbyException in Safemode
> --------------------------------------------------
>
> Key: HDFS-14914
> URL: https://issues.apache.org/jira/browse/HDFS-14914
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: Fengnan Li
> Assignee: Fengnan Li
> Priority: Minor
> Attachments: HDFS-14914-001.patch
>
>
> When observer is in safemode, calling getBlockLocations will make it throw
> RetriableException as in HDFS-13898. However, during startup the safemode is
> taking a really long time and retry would not help much here.
> What makes it worse is that when Routers talking to Observers, since Router
> distinguishes StandbyException and RetriableException, it will keep retry
> (default 3) times and then return to the client an RetriableException.
> The change is to make it failover so that Router can immediately try another
> Observer or Active namenode (depends on the design). The current
> ObserverReadProxyProvider doesn't get affected since both RetriableException
> and StandbyException will make it failover.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
