Re: MySQL Cluster HA Setup - Ted Schundler - com.mysql.lists.cluster

20 messages in com.mysql.lists.clusterRe: MySQL Cluster HA Setup

From	Sent On	Attachments
Marco Wertejuk	19 Jan 2005 16:38
Ted Schundler	21 Jan 2005 11:54
Marco Wertejuk	21 Jan 2005 13:43
Mikael Ronström	21 Jan 2005 14:46
Marco Wertejuk	21 Jan 2005 15:55
Casto, Robert	21 Jan 2005 16:33
Alex Davies	22 Jan 2005 00:06
Stewart Smith	23 Jan 2005 16:47
Jesse Erlbaum	23 Jan 2005 18:00
Casto, Robert	24 Jan 2005 08:23
Stewart Smith	24 Jan 2005 16:59
pek...@mysql.com	25 Jan 2005 11:45
Jesse Erlbaum	25 Jan 2005 13:59
Stewart Smith	26 Jan 2005 15:37
pek...@mysql.com	27 Jan 2005 00:53
Per Andreas Buer	02 Feb 2005 12:43
Stewart Smith	02 Feb 2005 18:07
Per Buer	03 Feb 2005 00:31
Eugen Leitl	03 Feb 2005 02:40
Robert Heinzmann	03 Feb 2005 10:52

Subject:	Re: MySQL Cluster HA Setup
From:	Ted Schundler (tsch...@gmail.com)
Date:	01/21/2005 11:54:12 AM
List:	com.mysql.lists.cluster

Well, I may be doing things incorrectly, but I setup two mgm servers. It seems all they do is serve up the configuration files. The ndbs then coordinate among themselves. It seems that the arbitrator is really whichever ndbd is "master", not the ndb_mgmd. So on each server, I have a ndb_mgmd running, and the local ndbd's and apis are given the local ndb_mgmd in their connectstring. That way, even if the other server is down, and the one that is up has some hicup, it can still load the config.

Also, its important that the ndbds are shutdown properly with "ndb_mgm ---execute="<id> stop"" (replacing <id> appropriately) to ensure continuous service. If non-master node is improperly shutdown, then it seems most of the time it will keep running. If the master node is improperly shutdown, then the remaining ndbd will have to restart before service resumes.

I'm not sure all that is completely correct, but its based on what seems to be working for me for a two node setup. Doing a "show" command in ndb_mgm sometimes gives weird results in that configugration.

Ted

On Thu, 20 Jan 2005 01:38:35 +0100, Marco Wertejuk <wert...@mwcis.com> wrote:

I'd like to setup two servers to have a clustered data source, I've seen different documentation telling me that I'll need at least three servers because I would also need a management server. I've also read that the management server is not neccessary after the cluster has been started or even after initialization.

My test setup includes two vmware virtual machines running on a private lan. I've also setup one of them to be the management server and of course I get the warning that running the arbitrator and one cluster node on the same host is not a good idea.

Now when I reboot the vm with the ndb_mgmd the other node gets an error and shuts down due to a cluster failure. This is not cleanly reproducable but happens at least in 75% of all reboots.

The logfile tells: Date/Time: Tuesday 18 January 2005 - 07:37:58 Type of error: error Message: Arbitrator shutdown Fault ID: 2305 Problem data: Arbitrator decided to shutdown this node Object of reference: QMGR (Line: 3796) 0x0000000a ProgramName: /usr/local/libexec/ndbd ProcessID: 4220 TraceFile: /var/cluster/db1/ndb_2_trace.log.4 ***EOM***

The tracefile is available here: http://www.hayai.de/misc/ndb_2_trace.log.4.bz2

Any help would be appreciated, regards, Marco