Do you wait for hinted handoffs when doing backups?

Jon Meredith jmeredith at basho.com
Wed Oct 30 12:37:35 EDT 2013


This has one of those 'it depends' answers.

While the node you are doing maintenance on is down, other members of the
cluster will temporarily handle the fallback partitions.

As soon as the node is brought up again, it will immediately start handling
requests again with the data it had at the point it went down.  The
fallback nodes stop receiving traffic and after a period of inactivity
(60s) will return it to the primary node that you had just taken down. The
handoff transfer limit controls how much handoff will take place
concurrently.  In the mean time, if stale data from that primary node is
retrieved before the handoff completes it will be read repaired.

If you take down another node before handoff has completed, for Nval=3 you
could be in a state where the original node you took down has stale data,
the second node you took down is now being serviced by a possibly empty
fallback and the third replica is being serviced by a normal primary.
 The answers you get when requesting data depend on the settings you pass
in.  If you just do an R=quorum, PR=0 it will be a race between the empty
fallback, the stale primary and the up to date primary.  You will either
get the stale object or the up to date object depending on which two nodes
responded first.  If you do an R=quorum, PR=quorum read then you'll get the
up to date data.  In both cases, if you retried the read you'll get the up
to date value once read repair completes as it will fix the out of date
primary and the secondary.

As all the nodes come back up, everything will be handed off back to the
correct locations.  The 'it depends' part of the answer comes from how well
your application tolerates eventual consistency.

Jon




On Wed, Oct 30, 2013 at 10:08 AM, morgandev <morgandev at gmail.com> wrote:

> I have the same question.  It seems like you should wait for all the
> handoffs
> before moving to the next node.
>
> Any updates from Basho?
>
>
>
> --
> View this message in context:
> http://riak-users.197444.n3.nabble.com/Do-you-wait-for-hinted-handoffs-when-doing-backups-tp4028966p4029645.html
> Sent from the Riak Users mailing list archive at Nabble.com.
>
> _______________________________________________
> riak-users mailing list
> riak-users at lists.basho.com
> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
>



-- 
Jon Meredith
VP, Engineering
Basho Technologies, Inc.
jmeredith at basho.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.basho.com/pipermail/riak-users_lists.basho.com/attachments/20131030/22a9d8ef/attachment.html>


More information about the riak-users mailing list