Re: [Pacemaker] Need an idea for dynamic configuration dependng on resource distribution
-Ursprüngliche Nachricht- Von: Florian Haas Gesendet: 29.04.2010 09:06:44 An: The Pacemaker cluster resource manager Betreff: Re: [Pacemaker] Need an idea for dynamic configuration dependng on resource distribution Hi Florian, >Are you talking about innodb_buffer_pool_size? > >If so, and regardless of the fact that this is beside the point of what >you are asking about, but if you use a buffer pool that's "as much as >possible", you'll potentially suffer pretty dramatic failover times due >to InnoDB recovery after a node crash. If you're on shared storage or >DRBD, that is. Your data would be safe but your MySQL might take minutes >or even hours to recover on startup. The webinar we did about the new >replication features in the MySQL RA covers this in some more detail. Very interesting. Where can I find it. > >Just my €.02. Good two cents. ;-) Best regards Andreas Mock signature.asc Description: OpenPGP digital signature ___ Pacemaker mailing list: Pacemaker@oss.clusterlabs.org http://oss.clusterlabs.org/mailman/listinfo/pacemaker Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf___ Pacemaker mailing list: Pacemaker@oss.clusterlabs.org http://oss.clusterlabs.org/mailman/listinfo/pacemaker Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Re: [Pacemaker] Need an idea for dynamic configuration dependng on resource distribution
On Thu, Apr 29, 2010 at 12:15 AM, Andreas Mock wrote: > -Ursprüngliche Nachricht- > Von: Andrew Beekhof > Gesendet: 27.04.2010 20:56:48 > An: The Pacemaker cluster resource manager > Betreff: Re: [Pacemaker] Need an idea for dynamic configuration dependng on > resource distribution > >>> configuration in the default case (everything o.k.). In this scenario the >>> resources >>> shall take as much "server estate" as possible. >> >>"server estate" == RAM/CPU etc? >>And you can tell the app how much to use? > > > a) Yes > b) More or less. Think about mysql with innodb. One central variable is > buffer_cache > which should be "as much as possible". When only one instance is running on a > server > you can make a feasible guess, e.g. 80% of physical mem. As soon as two > instances > are running on on server each instance can only have 80% / 2 of memory for > buffer_cache. > As soon as you "overbook" real memory things go dramatically worse. > > Any ideas? Perhaps add something to the start and stop actions which checks how many instances are running and fudges the settings accordingly. Or, maybe you could use a clone with globally-unique=true and pick which database to start based on the instance number - that way each instance would get notifications when the other started/stopped. ___ Pacemaker mailing list: Pacemaker@oss.clusterlabs.org http://oss.clusterlabs.org/mailman/listinfo/pacemaker Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Re: [Pacemaker] Need an idea for dynamic configuration dependng on resource distribution
On 04/29/2010 12:15 AM, Andreas Mock wrote: > -Ursprüngliche Nachricht- > Von: Andrew Beekhof > Gesendet: 27.04.2010 20:56:48 > An: The Pacemaker cluster resource manager > Betreff: Re: [Pacemaker] Need an idea for dynamic configuration dependng on > resource distribution > >>> configuration in the default case (everything o.k.). In this scenario the >>> resources >>> shall take as much "server estate" as possible. >> >> "server estate" == RAM/CPU etc? >> And you can tell the app how much to use? > > > a) Yes > b) More or less. Think about mysql with innodb. One central variable is > buffer_cache > which should be "as much as possible". Are you talking about innodb_buffer_pool_size? If so, and regardless of the fact that this is beside the point of what you are asking about, but if you use a buffer pool that's "as much as possible", you'll potentially suffer pretty dramatic failover times due to InnoDB recovery after a node crash. If you're on shared storage or DRBD, that is. Your data would be safe but your MySQL might take minutes or even hours to recover on startup. The webinar we did about the new replication features in the MySQL RA covers this in some more detail. Just my €.02. Florian signature.asc Description: OpenPGP digital signature ___ Pacemaker mailing list: Pacemaker@oss.clusterlabs.org http://oss.clusterlabs.org/mailman/listinfo/pacemaker Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Re: [Pacemaker] Need an idea for dynamic configuration dependng on resource distribution
-Ursprüngliche Nachricht- Von: Andrew Beekhof Gesendet: 27.04.2010 20:56:48 An: The Pacemaker cluster resource manager Betreff: Re: [Pacemaker] Need an idea for dynamic configuration dependng on resource distribution >> configuration in the default case (everything o.k.). In this scenario the >> resources >> shall take as much "server estate" as possible. > >"server estate" == RAM/CPU etc? >And you can tell the app how much to use? a) Yes b) More or less. Think about mysql with innodb. One central variable is buffer_cache which should be "as much as possible". When only one instance is running on a server you can make a feasible guess, e.g. 80% of physical mem. As soon as two instances are running on on server each instance can only have 80% / 2 of memory for buffer_cache. As soon as you "overbook" real memory things go dramatically worse. Any ideas? Best regards Andreas ___ Pacemaker mailing list: Pacemaker@oss.clusterlabs.org http://oss.clusterlabs.org/mailman/listinfo/pacemaker Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Re: [Pacemaker] Need an idea for dynamic configuration dependng on resource distribution
On Tue, Apr 27, 2010 at 2:45 PM, Andreas Mock wrote: > Hi all, > > I need an idea how I could achieve the following with corosync/pacemaker. > 2 servers in a cluster. Each server is running a resource on its own by > pacemaker > configuration in the default case (everything o.k.). In this scenario the > resources > shall take as much "server estate" as possible. "server estate" == RAM/CPU etc? And you can tell the app how much to use? > As soon as one server fails and one of the resources has to move so that both > resources run on one server I would like to change the configuration so that > both resources share the "server estate". > > Has anyone an idea how to achieve this? > Or probably done before successfully? > > Best regards > Andreas Mock > > ___ > Pacemaker mailing list: Pacemaker@oss.clusterlabs.org > http://oss.clusterlabs.org/mailman/listinfo/pacemaker > > Project Home: http://www.clusterlabs.org > Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf > ___ Pacemaker mailing list: Pacemaker@oss.clusterlabs.org http://oss.clusterlabs.org/mailman/listinfo/pacemaker Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Re: [Pacemaker] Need an idea for dynamic configuration dependng on resource distribution
-Ursprüngliche Nachricht- Von: Michael Schwartzkopff Gesendet: 27.04.2010 14:51:58 An: The Pacemaker cluster resource manager Betreff: Re: [Pacemaker] Need an idea for dynamic configuration dependng on resource distribution >Am Dienstag, 27. April 2010 14:45:21 schrieb Andreas Mock: >> Hi all, >> >> I need an idea how I could achieve the following with corosync/pacemaker. >> 2 servers in a cluster. Each server is running a resource on its own by >> pacemaker configuration in the default case (everything o.k.). In this >> scenario the resources shall take as much "server estate" as possible. >> As soon as one server fails and one of the resources has to move so that >> both resources run on one server I would like to change the configuration >> so that both resources share the "server estate". >> >> Has anyone an idea how to achieve this? >> Or probably done before successfully? > >First solution: >Write external scripts that check the coexistance of the services on a node >and reconfigure them accordingly. Hi Michael, That's too late. Both services have been started already probably tearing down the server. I don't know if there is a way to get the result of the newly calculated transition. The RA needs to know that it shall start the resource knowing that the other resource will be started on the same node. Is there a api for that kind of "knowledge" in the cluster? A way to transport the result of a calculated transition to the RA? Probably ptest at RA runtime? Best regards Andreas Mock ___ Pacemaker mailing list: Pacemaker@oss.clusterlabs.org http://oss.clusterlabs.org/mailman/listinfo/pacemaker Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Re: [Pacemaker] Need an idea for dynamic configuration dependng on resource distribution
Am Dienstag, 27. April 2010 14:45:21 schrieb Andreas Mock: > Hi all, > > I need an idea how I could achieve the following with corosync/pacemaker. > 2 servers in a cluster. Each server is running a resource on its own by > pacemaker configuration in the default case (everything o.k.). In this > scenario the resources shall take as much "server estate" as possible. > As soon as one server fails and one of the resources has to move so that > both resources run on one server I would like to change the configuration > so that both resources share the "server estate". > > Has anyone an idea how to achieve this? > Or probably done before successfully? > > Best regards > Andreas Mock > > ___ > Pacemaker mailing list: Pacemaker@oss.clusterlabs.org > http://oss.clusterlabs.org/mailman/listinfo/pacemaker > > Project Home: http://www.clusterlabs.org > Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf First solution: Write external scripts that check the coexistance of the services on a node and reconfigure them accordingly. Second solution: Hack the RA scripts and add such a check to the start section. Basically that means to derive your own RAs from the existing ones. Please keep in mind to revert the config changes as soon the services run on different nodes again. Greetings, -- Dr. Michael Schwartzkopff MultiNET Services GmbH Addresse: Bretonischer Ring 7; 85630 Grasbrunn; Germanyt Tel: +49 - 89 - 45 69 11 0 Fax: +49 - 89 - 45 69 11 21 mob: +49 - 174 - 343 28 75 mail: mi...@multinet.de web: www.multinet.de Sitz der Gesellschaft: 85630 Grasbrunn Registergericht: Amtsgericht München HRB 114375 Geschäftsführer: Günter Jurgeneit, Hubert Martens --- PGP Fingerprint: F919 3919 FF12 ED5A 2801 DEA6 AA77 57A4 EDD8 979B Skype: misch42 ___ Pacemaker mailing list: Pacemaker@oss.clusterlabs.org http://oss.clusterlabs.org/mailman/listinfo/pacemaker Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
[Pacemaker] Need an idea for dynamic configuration dependng on resource distribution
Hi all, I need an idea how I could achieve the following with corosync/pacemaker. 2 servers in a cluster. Each server is running a resource on its own by pacemaker configuration in the default case (everything o.k.). In this scenario the resources shall take as much "server estate" as possible. As soon as one server fails and one of the resources has to move so that both resources run on one server I would like to change the configuration so that both resources share the "server estate". Has anyone an idea how to achieve this? Or probably done before successfully? Best regards Andreas Mock ___ Pacemaker mailing list: Pacemaker@oss.clusterlabs.org http://oss.clusterlabs.org/mailman/listinfo/pacemaker Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf