Re: Why would restarting a node abort a transaction? - Kevin Burton - com.mysql.lists.cluster

12 messages in com.mysql.lists.clusterRe: Why would restarting a node abort...

From	Sent On	Attachments
Kevin Burton	19 Jan 2007 21:47
Mikael Ronström	20 Jan 2007 09:19
Kevin Burton	20 Jan 2007 12:19
Stewart Smith	20 Jan 2007 23:21
Andy Smith	21 Jan 2007 00:03
Philipp Taprogge	21 Jan 2007 03:50
Geert Vanderkelen	21 Jan 2007 07:16
Kevin Burton	21 Jan 2007 11:01
Hartmut Holzgraefe	21 Jan 2007 15:01
pek...@mysql.com	21 Jan 2007 15:46
Kevin Burton	21 Jan 2007 16:55
Stewart Smith	01 Feb 2007 04:55

Subject:	Re: Why would restarting a node abort a transaction?
From:	Kevin Burton (bur...@tailrank.com)
Date:	01/21/2007 11:01:21 AM
List:	com.mysql.lists.cluster

You have to do it on application level anyway. No matter what DBMS you use which support transactions. If the DBMS is aborting the transaction the application has to consider retrying it, give a failure after 3 attempts, for example.

In a high availability and distributed DBMS cluster not doing this in the DMBS is a cop out.

Is it possible to rerun the part of the transaction that failed on the replica of the data node that failed instead of having to run it in a larger cluster?

I'm just trying to think that on a NDB install with a LOT of node a data node failing would require a LOT of extra work on the nodes that stay online.

Also..... shouldn't there be a section about this in the manual?

There should be a wiki for NDB that's actually used. the HashMySQL wiki doesn't have a cluster section that I saw.