38 messages in com.mysql.lists.clusterRE: Is MySQL Cluster stable and matur...
FromSent OnAttachments
Konstantin Rozinov24 Jul 2008 00:01 
Krishna Chandra Prajapati24 Jul 2008 00:19 
Mark Callaghan24 Jul 2008 00:31 
Serge Kozlov24 Jul 2008 03:46 
Josh Miller24 Jul 2008 07:15 
Konstantin Rozinov24 Jul 2008 12:49 
Konstantin Rozinov24 Jul 2008 13:00 
Konstantin Rozinov24 Jul 2008 13:04 
Josh Miller25 Jul 2008 07:58 
Ben Wiechman25 Jul 2008 08:43 
Serge Kozlov25 Jul 2008 09:44 
Pascal Charest25 Jul 2008 12:48 
Konstantin Rozinov25 Jul 2008 17:40 
Konstantin Rozinov25 Jul 2008 17:54 
Konstantin Rozinov25 Jul 2008 17:56 
Mark Callaghan25 Jul 2008 22:58 
Serge Kozlov25 Jul 2008 23:19 
Massimo26 Jul 2008 00:26 
Mark Callaghan26 Jul 2008 07:38 
Jeff Sturm26 Jul 2008 20:49 
Andrew Garner26 Jul 2008 22:51 
Mark Callaghan27 Jul 2008 06:23 
Jeff Sturm27 Jul 2008 12:14 
Jeff Sturm27 Jul 2008 12:49 
Serge Fonville28 Jul 2008 01:04 
Konstantin Rozinov28 Jul 2008 23:53 
living liquid | Christian Meisinger29 Jul 2008 00:07 
Konstantin Rozinov29 Jul 2008 00:28 
living liquid | Christian Meisinger29 Jul 2008 00:56 
Massimo29 Jul 2008 05:19 
Matthew Montgomery29 Jul 2008 05:51 
Hartmut Holzgraefe29 Jul 2008 05:54 
Mikael Ronström29 Jul 2008 06:56 
Mikael Ronström29 Jul 2008 07:07 
Serge Fonville29 Jul 2008 07:29 
Mikael Ronström29 Jul 2008 08:04 
Burhan Khalid29 Jul 2008 14:37 
Konstantin Rozinov29 Jul 2008 17:28 
Subject:RE: Is MySQL Cluster stable and mature enough to run a social network?
From:Jeff Sturm (jeff@eprize.com)
Date:07/27/2008 12:49:05 PM
List:com.mysql.lists.cluster

-----Original Message----- From: muza@gmail.com [mailto:muza@gmail.com] On Behalf Of Andrew Garner Sent: Sunday, July 27, 2008 1:52 AM To: clus@lists.mysql.com Subject: Re: Is MySQL Cluster stable and mature enough to run a social network?

On Sat, Jul 26, 2008 at 10:49 PM, Jeff Sturm <jeff@eprize.com> wrote:

One of the myths I hear about MySQL Cluster is that schemas and/or software must specifically be adapted to it. I haven't found this to be universally true. Although our engineering staff specifically tests our releases with NDB tables, the characteristics of NDB do not generally influence the design of our software (with a few notable exceptions).

This is hardly a myth. There are, of course, applications that fit very well into the cluster paradigm, but in my experience, it's much more often the case that common web applications have queries and schema which need non-trivial work to obtain acceptable performance on the cluster.

Hard to say without discussing specifics. If you want to talk about how some queries are hard to optimize on NDB due to lack of index statistics, or some joins that simply aren't suitable for the cluster architecture, I can agre with that. But then you're getting away from the product's sweet spot, and the advantages of NDB won't be evident anyway.

We've rearchitected our internal applications around a message bus architecture, so that we can now use a completely different DB architecture for transaction processing and reporting systems. I guess that does constitute "non-trival work" but it's a good practice anyway, and we would've done so with or without cluster, if for no other reason than to give us the freedom to spec and configure each db host to optimize it to its task.

This is assuming reasonable size datasets where one doesn't have to entertain the idea of disk based tables and the large performance hit those bring.

I'm not saying it's right for everyone. Just that the rumors NDB is hard to use for common applications may be overstated. If you have terabytes of data, you probably shouldn't bother with it.

The high availability features are great when they work, and extremely frustrating when all you get is a cascading node failure with "Temporary error, restart node". Caveat emptor.

No doubt about that. Does anyone have good/bad reports to tell on 6.2 stability? It may be too early to tell.

Jeff