does hyperdex support multi-master replica

[asif uddin Ahmad]

yes i now that the hyperdex support the replica but i am confuse which reply it will support. i didnot get any link so if possible it will help for me 

[Robert Escriva]

On Tue, Jun 23, 2015 at 07:52:30AM -0700, asif uddin Ahmad wrote:
> yes i now that the hyperdex support the replica but i am confuse which reply it
> will support. i didnot get any link so if possible it will help for me

Hi,

The short answer to your question is that this information is largely
hidden from the user. Different keys will have different replicas
respond to the user, and it's entirely outside the domain of things the
user needs to care about. The user is guaranteed linearizability, and
doesn't need to worry about replica selection at all to get that
guarantee.

If you're more curious about the internals, they are described in our
papers and slide decks: http://hyperdex.org/papers/.

-Robert

[Emin Gün Sirer]

Hi Asif,

Let us know what precise scenario you are concerned about, and we can delve into more detail as necessary. In first-generation NoSQL systems, it matters whether or not the request was sent to the primary replica or the secondary, because the secondary may be out of date and provide stale answers. HyperDex guarantees linearizability, do these concerns do not apply. Every GET will return the result of the latest PUT -- the user doesn't need to know anything about the replicas operation behind the scenes. This guarantee holds even when nodes are failing and new replicas are being added to the cluster. 

But once again, if you have a specific scenario you're concerned about, do let us know and we can go through it in detail.

- egs

On Tue, Jun 23, 2015 at 10:52 AM, asif uddin Ahmad <hias...@gmail.com> wrote:

yes i now that the hyperdex support the replica but i am confuse which reply it will support. i didnot get any link so if possible it will help for me 

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

[why...@gmail.com]

On Tuesday, June 23, 2015 at 4:29:39 PM UTC+1, egs wrote:

Hi Asif,

Let us know what precise scenario you are concerned about, and we can delve into more detail as necessary.

 

Hi! apologies for reviving old threads... I'm also specifically interested in this question.

I have a situation where I need to run a cluster with geographically separated nodes, in a situation where network partition is not only likely, it's guaranteed.

Therefore, I need a kind of multi-master replication setup where every node can still answers queries, (albeit with stale data) when it is isolated from the rest of the cluster.
When network connectivity is restored, all nodes should synchronize any updates, with a most recent update wins strategy.
It seems from reading some documentation that it might be possible to configure a hyperdex cluster to behave like this, although I would need to adjust the partitions and tolerate failures as I add nodes, and I'm not sure how inconvenient that might be.
I would appreciate your opinion.

Thanks!