[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-ha-dev
Subject:    Re: [Linux-ha-dev] [Bug 1722] First item in a group is not stopped
From:       Lars Marowsky-Bree <lmb () suse ! de>
Date:       2007-10-29 21:42:12
Message-ID: 20071029214212.GH4040 () marowsky-bree ! de
[Download RAW message or body]

On 2007-10-29T21:16:17, Keisuke MORI <kskmori@intellilink.co.jp> wrote:

> In the HA database cluster, the database service is typically provided
> by the group like:
>   Filesystem + MySQL + IP
> 
> If any of the resources failed then the database service is no
> longer available. Running only Filesystem does not mean anything to
> "the service availability."

But the CRM doesn't know that, nor should it. It tries to bring online
as many of the configured resources as possible.

> We usually group up resources because they need to run together to
> provide "the service" (database, web server, or whatever) as a whole,
> therefore running a part of the group does not make sense.

Yes and no.

Remember that groups are internally expanded into regular dependencies.
The cluster will try to bring online as much of the hierarchy /
dependency tree as possible; the goal being 100%, all the time.

Consider the IP/FS/DB/Web example. The DB might be used by other
services as well; why shouldn't it be brought online?

Also, you might have explicitly set the target_role of the last resource
in the group to "stopped", to not bring the resources beyond that point
online - this would then be stopped completely as well.

> > Let me instead ask what you believe you gain by stopping the first
> > resource.
> Because it is just simple and intuitive for users.

But they don't gain anything by stopping the resources which might as
well be running. Instead, they'd gain a special case where the group
behaved differently from regular dependencies, and I'm quite certain
that would not be a good idea.

> And I believe that the most of commercial HA software would also
> behave like this (at least in the typical usage).

IRIS FailSafe wouldn't, and I doubt that IBM's Tivoli System Automation
would; SteelEye LifeKeeper I'm reasonably sure about too.

> Our costomers are considering to migrate from a commercial HA
> software to heartbeat, and all of them expect to behave like
> this so far.
> 
> At least it would be nice if it's able to be customized, I would think.

You could get this behaviour back if you switched from groups to
constraints - and used rsc_colocation with symmetrical="true", I think.
Or make a patch to have a "symmetrical" attribute on the group to
propagate it down.


Regards,
    Lars

-- 
Teamlead Kernel, SuSE Labs, Research and Development
SUSE LINUX Products GmbH, GF: Markus Rex, HRB 16746 (AG Nürnberg)
"Experience is the name everyone gives to their mistakes." -- Oscar Wilde

_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
[prev in list] [next in list] [prev in thread] [next in thread]