[DRBD-user] Retrieving the inconsistent + local disk state

Lars Marowsky-Bree lmb at suse.de
Thu Sep 22 15:36:28 CEST 2005

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On 2005-09-22T15:24:22, Philipp Reisner <philipp.reisner at linbit.com> wrote:

> Could you describe in more detail what you going to do there?

So, the current heartbeat CVS has the ability to model resources which
have more than one state. After start, they are in "demoted" state, and
then they might be promoted later. (And, vice-versa, demoted from
promoted state.)

That's an extension of the clone support in heartbeat 2.0.1.

During start (to demoted state), we can pass a hint (integer) to the
cluster manager about the ability / preference to be promoted on this
particular node, or say that we can't possibly be promoted. (Like, for
example, after we have timed out to connect to the peer and don't have
consistent local storage.) 

OTOH, one might argue that we'll find out early enough when we in fact
do try to promote ;-) So it's just a hint where to place resources.

There's also notifications about what the CRM is doing to the peer(s)
(plural in principle, but of course, with drbd, there's always just one
peer or none). This could be used together with "freeze_io" to switch a
primary to "standalone" mode (and thus continue) after the peer has
been fenced/stopped after failure, which is the other half of the "mark
local copy inconsistent" after an internal comm failure of drbd.

It's not yet fully thought out, Andrew and I am sketching this as we go
along ;-)

> Where can we find out more about the master/slave-support of heartbeat?

Welllll, there's the heartbeat CVS, there's test cases in
crm/pengine/testcases/ in CVS,
http://wiki.linux-ha.org/ClusterResourceManager/DTD1.0/Annotated#complex_resource
has some information too.

Andrew and I also hang out on the freenode.net #linux-ha channel,
despite being at the SUSE Labs conference ;-)

> Should this thread go to the drbd-dev ML ?

Either that or the linux-ha-dev list, as the master/slave support isn't
yet fully worked out yet I assume.


Sincerely,
    Lars Marowsky-Brée <lmb at suse.de>

-- 
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business	 -- Charles Darwin
"Ignorance more frequently begets confidence than does knowledge"




More information about the drbd-user mailing list