Re: drbd resource script patch

This looks like a really good proposal. I would propose that you can test 
the generality of this system by applying this to another distributed 
program: MySQL. I am looking at trying to build a distributed 
single-master, multiple-slave MySQL database system. I'd like to have it 
all under failsafe control and have failsafe automatically choose a new 
slave to promote to master when the primary master dies.
--
Michael

On Thu, 9 May 2002, Lars Marowsky-Bree wrote:

> On 2002-05-08T16:55:52,
>    Martin Bene <[email protected]> said:
> 
> Hi Martin,
> 
> thanks for the patch! I have folded it into CVS; as I couldn't test drbd
> locally, I will trust your code.
> 
> I have thought about how to handle drbd resources cleanly with FailSafe; the
> scheme is actually pretty similiar for any active/replicated resource, like
> databases etc. If you are using drbd, you might find this helpful - I haven't
> tried it, but I think it makes sense; input and patches appreciated ;-)
> 
> - FailSafe should not only control the primary, but also the secondary via a
>   resource.
> 
>   (ie, the secondary is "online pending" until it is fully sync'ed etc)
>   
> - The primary resource _knows_ about this and can control the secondary
>   resource.
> 
> - A special failover policy script will automatically pick the right node for
>   the primary.
>   
> > drbd is a nasty resource to handle while resync is in progress: it can not
> > be stopped, and for successfull startup, it MUST be started on the node
> > where it's already primary. The patch tries to accomodate this by moving
> > some of the startup checks from exclusive to start: for exclusive check, the
> > node where the resource will be started is still unknown, so it's not
> > possible to handle drbds requirements. Correct check can only be made at
> > start time. Fortunately, failsafe is persistent enough to get the drbd
> > resource correctly online for all reasonable device states. 
> 
> The failover script could take this into account.
> 
> A very minor nit pick - could you try to set your line-length in e-mail to 78
> characters or so?
> 
> 
> Sincerely,
>     Lars Marowsky-Brée <[email protected]>
> 
>