[
https://issues.apache.org/jira/browse/HDFS-2713?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13174980#comment-13174980
]
Aaron T. Myers commented on HDFS-2713:
--------------------------------------
{quote}
After any DFSClient operation fails due to Namenode unavailability, the most
important thing to do is to detect when the Active Namenode becomes available
again.
So the background thread is not doing any unnecessary work, it is doing the
high priority work.
{quote}
That's not necessarily true. It's only important *if* the DFSClient will indeed
be used later for another client operation after some client operation has
timed out. If it's not reused, then any work the background thread has done
will in fact have been unnecessary. I would guess (pure conjecture) that most
client programs will not be structured this way. That is, once a single call
times out for a given DFSClient, that DFSClient will not be reused.
{quote}
My intention is that, when one client call finds failover is required and not
able to complete the failover within the wait time, then why do I need to wait
till next call comes to try again and failover after mindealy wait?
Even though the first call fails, this background thread will ensure to find
the active proxy instance. If next call comes now(this is user thread), it need
not wait to connect and failover again. Immediately it can make use of that
proxy instance and goahead.
{quote}
But the delay on the subsequent call should be minimal (less than a second),
assuming the NN is indeed back up. Given that failover of the NN will likely
take tens of seconds, this particular aspect of client failover doesn't seem to
me like it's something worth optimizing aggressively.
Two things:
# This isn't so much "an alternative approach to clients handling NN failover"
as it is a potential performance improvement for subsequent calls after a
failed call due to the active NN being down. Would you mind changing the JIRA
summary and description to better reflect this?
# Given that it's a performance improvement, could you provide some benchmarks?
Do you have an actual workload which benefits from this change?
> HA : An alternative approach to clients handling Namenode failover.
> --------------------------------------------------------------------
>
> Key: HDFS-2713
> URL: https://issues.apache.org/jira/browse/HDFS-2713
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: ha, hdfs client
> Affects Versions: HA branch (HDFS-1623)
> Reporter: Uma Maheswara Rao G
> Assignee: Uma Maheswara Rao G
>
> This is the approach for client failover which we adopted when we developed
> HA for Hadoop. I would like to propose thia approach for others to review &
> include in the HA implementation, if found useful.
> This is similar to the ConfiguredProxyProvider in the sense that the it takes
> the address of both the Namenodes as the input. The major differences I can
> see from the current implementation are
> 1) During failover, user threads can be controlled very accurately about *the
> time they wait for active namenode* to be available, awaiting the retry.
> Beyond this, the threads will not be made to wait; DFS Client will throw an
> Exception indicating that the operation has failed.
> 2) Failover happens in a seperate thread, not in the client application
> threads. The thread will keep trying to find the Active Namenode until it
> succeeds.
> 3) This also means that irrespective of whether the operation's RetryAction
> is RETRY_FAILOVER or FAIL, the user thread can trigger the client's failover.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
