[Pacemaker] time synchronisation

[Lars Marowsky-Bree]

On 2012-12-19T10:06:25, James Harper <james.harper at bendigoit.com.au> wrote:

> What is the behaviour of a cluster when the nodes are up to 10 minutes
> out of sync with each other, because they've just been booted up after
> a crash and the hwclocks are out of date and there is no ntp time
> source reachable? Could it cause lots of sig11's and constant
> re-elections because that's what I was seeing and it all seemed to go
> away once I artificially synchronised the clocks... I haven't tried to
> reproduce the problem so I can't be sure.

The cluster stack itself should not show such behaviour. Of course, it's
going to make reading and lining up logfiles hell, but it shouldn't
crash. If you can reproduce this, please file bugs.

Services (especially active/active ones) may exhibit strange behaviour
if they use the wallclock timestamps for anything relevant, and some
don't like failing over into the past. But that would be bugs in the
services.

They should all read Lamport ;-)


Regards,
    Lars

-- 
Architect Storage/HA
SUSE LINUX Products GmbH, GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer, HRB 21284 (AG Nürnberg)
"Experience is the name everyone gives to their mistakes." -- Oscar Wilde