7 messages in com.mysql.lists.clusterRe: NDB commit durability
FromSent OnAttachments
James Kingsbery Jr.12 Feb 2008 16:09 
Geert Vanderkelen18 Feb 2008 00:24 
James Kingsbery Jr.18 Feb 2008 08:08 
John David Duncan18 Feb 2008 08:51 
Rayson Ho18 Feb 2008 09:54 
Jeff Sturm18 Feb 2008 10:01 
Frazer Clement18 Feb 2008 10:30 
Subject:Re: NDB commit durability
From:Geert Vanderkelen (gee@mysql.com)
Date:02/18/2008 12:24:59 AM
List:com.mysql.lists.cluster

Hi James,

On Feb 13, 2008, at 1:10 , James Kingsbery Jr. wrote:

I was wondering if anyone had pointers to durability of commits in NDB. The company I work for is exploring using NDB, and someone here raised this issue. What we're concerned about is making sure that things get written to the binary log as soon as possible so data doesn't get lost. I'm looking for documentation at a lower level than the manual - any information would be greatly appreciated.

Well, Cluster is ACId, with small d where commits happen to memory on all data nodes (or none). Small d is nothing official, and it's still durable since everything is check pointed to disk and on different machines. So, if one machine is failing, it's still on the other.

About binary logging.. that's something else. You'll need at least two Master MySQL server, SQL nodes, which log to binary logs, so you can fail over. There will be gaps in the logs, but all should be logged to two machines, so nothing really is lost. What you see in manual, is what we have. You had anything specific?

Cheers,

Geert

-- Geert Vanderkelen, Senior Support Engineer MySQL GmbH, Germany, www.mysql.com