Re: SQL Node clarification

[Harrison Fisk]

Hi,

Shaheeb Roshan wrote:
> Hello,
>
> I realize that this question has been asked and
> answered a few times, but I'm still not
> comfortable. Does the following setup make
> sense:
>
> Server 1: SQL Node + Data Node
> Server 2: SQL Node + Data Node
> Server 3: MGM Node
>
> Servers 1&2 are identical with adequate RAM.
> Server 3 contains my application middleware in
> addition to serving as the managment node.
>
> Does the above setup offer high availability? So
> if Server 1 fails (hardware fails or os crashes),
> will I still have uninterrupted
> functionality/access to my data? If no, then why
> not?

Yes. Since the MGM can act as the arbitrator, if any single server crashes everything else will continue to run. If you lose two servers however, the entire thing will shutdown.

>
> If yes, I don't quite understand how to tell my
> application which mysql server to query.
>
> Is there a way to present a single IP address or
> hostname to my application that will then query
> against any available nodes in the cluster? Do I
> have to use a separate technology like a director
> (maybe ldirectord) to accomplish this?

Yes, you need to use something else. Some API's natively support this (JDBC being the main one), and there are other solutions as well. This could be as simple as something like round-robin DNS, or something like LVS. I've also seen people use hardware load balancers for this as well.

Keep in mind that you can query to both machines since they are both active.

> If so,
> then what is the advantage of Mysql Cluster in the
> HA context? Can I not achieve the same net result
> with replication, a director and heartbeats from
> the servers?

No. Replication is asyncronous one-way. This means that you can't actively write to both nodes at a time (like you can with cluster). It also means that when you failover to the slave, you can be potentially losing data, which is undesirable for many applications.

> And finally I'd just like to note that this
> community is very impressive in its technical
> contribution. I've found answers to a lot of
> (seemingly) unique and complex problems. I hope
> this time is no different!
>
>
> Shaheeb Roshan
> MRS Inc.