Amazon RDS - thoughts?

[Jared Hirsch]

Hey Sheeri,

Gene and I are starting to enumerate migration paths from Mozilla physical
infrastructure to AWS. Amazon RDS is a natural option to consider, given
that we're already on MySQL.

You've poked around at our topology a bit; any thoughts on RDS pros and
cons, and whether it seems like a good option for us to improve robustness
as we migrate?

I think the leading alternative, in a minimal-change scenario, would be to
reproduce our current topology on provisioned IOPS instances spread across
2 regions (analogous to 2 datacenters), and manage the mysql instances
ourselves.

Copying the public list in the interest of greater openness.

Cheers,

Jared

[Sheeri Cabral]

What are the robustness problems you're having right now, that moving to AWS will hopefully fix? We have multiple data centers for redundancy, so I'm not sure what you're gaining by going to AWS (whether you go to RDS or not).

RDS is good for robustness, but if you need to change a global parameter you have to make a new parameter group and restart the RDS database. Also, you have no access to logs, so if anything goes wrong, you're stuck because you have no idea what happened. There are things you can do without restarting, like starting a slave if it stops...but in general I'd stay away from something where you have to administer it (like configuration variables with RDS) but don't have access to logs (binary logs, slow query logs, error logs) to debug problems when they come up.

-Sheeri Cabral
Senior DB Admin/Architect
Mozilla

[Mike G]

I like xeround myself but very expensive.


Thanks,
Mike

> _______________________________________________
> dev-identity mailing list
> dev-id...@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-identity
>

[Gene Wood]

Sheeri,
Our current DB architecture has a single point of failure with the write
master. When that server goes down, writes to the system stop working.

We are moving to AWS for reasons unrelated to database architecture.
Question is, does moving to RDS over running our own AWS instances of MySQL
buy us anything? Specifically does it buy us anything in regards to
addressing our current single point of failure?
-Gene

[Sheeri Cabral]

Hi Gene,

You can solve the single point of failure on the write master by using load balancing, whether with a traditional load balancer or a service like ScaleBase (we use both these) or tools like MHA (http://code.google.com/p/mysql-master-ha/). You can also write to more than one master, but that depends on if the app can handle it (mostly, can it handle auto-increments that increment by more than 1). This doesn't *scale* writes, but it allows 2 masters to be used at the same time in the case that network routing breaks for certain locations.

But since the move to AWS is unrelated, that's not necessarily an issue. Running RDS over AWS doesn't gain you anything over running your own MySQL with regards to having a single point of failure. This might help - see #5 in particular - http://cloud.dzone.com/articles/10-tops-deciding-between - the article is from 6 months ago and as far as I know, it's still up to date.