Voldemort 0.70 is released

[Alex Feinberg]

Hello,

Voldemort 0.70 is finally out of the candidate stage. While the plan
was to release this on Monday 1/25, testing has uncovered a few
non-trivial bugs that have since been fixed. You can get the release
at its usual place: < http://github.com/voldemort/voldemort/downloads >.
Additionally, you can browse the source code in the release-070
branch (which has since been merged up the master branch) on github:
< http://github.com/voldemort/voldemort >. Javadocs have been updated
and are viewable at: < http://project-voldemort.com/javadoc/client/ >.

This is the first official release to include rebalancing, the
long-awaited feature which enables online cluster expansion. We have
tested rebalancing extensively: on our own workstations, on EC2 and
finally in our QA lab (where we successfully rebalanced 300+ gb of data
under simulated load, going from a six to eight nodes). Despite this,
rebalancing should still be considered a *beta* feature: further
issues may be found, further documentation (e.g. recommended settings and
procedures) will follow. We'd like to thank all those who have
started experimenting with this feature, the feedback had been very
useful and more is always welcome.

Below are the release notes for this release (viewable at
< http://github.com/voldemort/voldemort/blob/master/release_notes.txt >):

Changes made since 0.60.1:

* A beta of rebalancing (dynamic cluster expansion) support merged
into the main branch. See the project's wiki for more information:
http://wiki.github.com/voldemort/voldemort/voldemort-rebalancing
* New failure detector merged into the main branch:
http://wiki.github.com/voldemort/voldemort/failure-detection
* Beta mechanism for restoring all of node's data from replicas on
demand. This is an alternative to a more gradual mechanism provided
by read-repair: useful when a machine is down for a prolonged period
and is then re-inserted into the cluster.
Invoked via JMX: the operation is restoreDataFromReplication in the
voldemort.server.VoldemortServer MBean, with a mandatory parameter
(integer >= 1) indicating the number of transfers to do in parallel.
* Simple gossip protocol (for cluster metadata) merged
into the main branch. Disabled by default: use "enable.gossip=true"
to enable, use "gossip.interval.ms" to set an interval at which gossip
occurs (default: 30000 i.e., 30 seconds).
* Fixed issue 190: add a way of aggregating performance data over
all stores
* Fixed issue 181: stack traces shouldn't be filled for Obsolete
version exception

Thanks,
- Alex

[B. Todd Burruss]

thx alex, could someone expand on what the gossip protocol gives
voldemort? do we setup our config files differently?

> --
> You received this message because you are subscribed to the Google Groups "project-voldemort" group.
> To post to this group, send email to project-...@googlegroups.com.
> To unsubscribe from this group, send email to project-voldem...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/project-voldemort?hl=en.
>
>

[Alex Feinberg]

When rebalancing happens, metadata is synchronously updated on all the
nodes in the cluster *at that time*. However, if some nodes are down
at the time (or simply experience a transient failure when the
metadata update is used), the gossip protocol will allow them to pick
up data about the new machines entering the cluster.

I'd enable gossip on clusters that you expect to rebalance on a
regular basis. Simply setting enable.gossip=true in server.properties
should suffice, additionally you may want to set gossip.interval.ms to
a value that's more reasonable for your cluster.

Thanks,
- Alex