Is voldemort db obsolete ?

1214 views
Skip to first unread message

abbas...@tokopedia.com

unread,
Aug 16, 2018, 6:51:50 AM8/16/18
to project-voldemort
I dont see much activity on this group nor i see any commit in the github repo. Is the Voldemort DB is dead ?

Félix GV

unread,
Aug 16, 2018, 12:00:19 PM8/16/18
to project-...@googlegroups.com
Hello,

Good question.

LinkedIn is the originator of Voldemort, and until now has been the most significant contributor and maintainer.

As of a few months ago, LinkedIn finished migrating off of Voldemort and into a new system, Venice, which is still proprietary as of now.

Therefore, LinkedIn is not using Voldemort anymore and will not be providing any significant support to the Voldemort open-source community going forward. Even if we wanted to, most of the Voldemort engineers have since moved on to other companies and projects, so we cannot realistically rely on them. We (LinkedIn) may still merge in patches if they are simple to review and in good shape, but if this project is to continue thriving, it would need a proper community behind it.

I know that there are still other Voldemort users out there, including some with their own fork (whether in open-source or proprietary). Perhaps one of these users might want to be the project’s new shepherd.

As for you, if you are starting a new project and considering Voldemort for it, my advice to you would be two-fold:

1. If you are looking for an online read-write datastore with tunable consistency, Voldemort can achieve this pretty well, but so can other systems. In particular, Cassandra should have similar characteristics to Voldemort Read-Write in terms of scalability, latency and tunable consistency. Cassandra has a much stronger open-source community behind it, and also has ScyllaDB, which is an API-compatible C++ rewrite for even better performance. You may want to consider these projects, or others, if you have a read-write use case.

2. If you are looking at Voldemort Read-Only, which can bulkload data from Hadoop, then I’m afraid that there is no strong contender in that space. Cassandra has some sort of hadoop-to-cass project, and HBase allows side loading of HFiles created on Hadoop, but, as far as I know, these solutions are not as mature and easy to use as Voldemort Read-Only. Don’t take my word for it though, maybe these projects have evolved more than I know since I last looked at them, or perhaps there are new contenders in the space. Do your own research. But if you need Hadoop to online serving capabilities, it may still be a decent choice to use V-RO, especially if the other V-RO users decide to band together and shepherd the project going forward.

Good luck. Cheers!

-F
On Thu, Aug 16, 2018 at 03:51 abbas.zaidi via project-voldemort <project-...@googlegroups.com> wrote:
I dont see much activity on this group nor i see any commit in the github repo. Is the Voldemort DB is dead ?

--
You received this message because you are subscribed to the Google Groups "project-voldemort" group.
To unsubscribe from this group and stop receiving emails from it, send an email to project-voldem...@googlegroups.com.
Visit this group at https://groups.google.com/group/project-voldemort.
For more options, visit https://groups.google.com/d/optout.

Canute

unread,
Aug 17, 2018, 8:08:51 AM8/17/18
to project-voldemort
In that case, it may be a good time to revitalise it, because the model that it is based on is the future.
It is the main contender to replace cloud based storage in that the cloud can be "anywhere". It also support meta-data and that enables tons of extensions. The problem is that it was not invented by Microsoft, and not even Google, but is perceived as a threat to the business model of the great US companies.
Lets get cracking again!
-k

David Ongaro

unread,
Nov 4, 2018, 3:46:52 PM11/4/18
to project-voldemort
That sounds a little hyperbolic for me, but if that provides motivation for someone to bring this project forward then go on!

In any case I think the above point 2 of Félix GV is true. If you need a very low latency read-only lookup of values by keys on edge nodes and are fine with preparing the stores by batch processing on hadoop and updating them over the network there might not be a better contender!

The problem is that for LinkedIn this project is obsolete, so I think it needs another maintainer. Maybe LinkedIn could actively help with transitioning this project, but probably some volunteers need to show up first.
Reply all
Reply to author
Forward
0 new messages