Distributed Replicated Block Device (DRBD) development
 help / color / mirror / Atom feed
* [Drbd-dev] About the UUID
@ 2006-12-20  7:06 home_king
  2006-12-20  8:40 ` Lars Ellenberg
  0 siblings, 1 reply; 2+ messages in thread
From: home_king @ 2006-12-20  7:06 UTC (permalink / raw)
  To: drbd-dev

hi, Philipp.

Now drbd-0.8.x uses UUID instead of GC as the data-relationship-detection
-tool, right? Its logic seems clearer than GC. However, I am confused
about the UUID in the two-primaries-healedpartition scenario.

Given the original uuids below (both nodes are connected):
P <5,0,0,0>      S <5,0,0,0>
then, I plug out the wire of the primary node (disconnected):
P <6,5,0,0>      S <5,0,0,0>
I let the downgraded secondary become new primary, UUID should become:
P <6,5,0,0>      S <6,5,0,0>
Finally, I recover the wire of primary. And Now, there comes my problem.
Accordding to the UUID algorithm, when Cs == Cp, no resynchronisation
necessary. However it is obviously split-brain, and drbd ignores it.

Why?

BTW, I did not test such scenario, because I can't compile drbd-0.8.x
in my RedHat AS4 distro. I just browsed the doc & code. :-)

Regards,
luo.jinhua



^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [Drbd-dev] About the UUID
  2006-12-20  7:06 [Drbd-dev] About the UUID home_king
@ 2006-12-20  8:40 ` Lars Ellenberg
  0 siblings, 0 replies; 2+ messages in thread
From: Lars Ellenberg @ 2006-12-20  8:40 UTC (permalink / raw)
  To: drbd-dev

/ 2006-12-20 15:06:57 +0800
\ home_king:
> hi, Philipp.
> 
> Now drbd-0.8.x uses UUID instead of GC as the data-relationship-detection
> -tool, right? Its logic seems clearer than GC. However, I am confused
> about the UUID in the two-primaries-healedpartition scenario.
> 
> Given the original uuids below (both nodes are connected):
> P <5,0,0,0>      S <5,0,0,0>
> then, I plug out the wire of the primary node (disconnected):
> P <6,5,0,0>      S <5,0,0,0>
> I let the downgraded secondary become new primary, UUID should become:
> P <6,5,0,0>      S <6,5,0,0>

no. these small digits are only more readable examples.
our UUIDs are 63bit random numbers.
they may clash, and we have to fix that somewhen.
but that is not very likely.

we probably should change the example to read more like
head -c8 /dev/urandom | hd
<3cecd7aed66ff5ad,0,0,0>		<3cecd7aed66ff5ad,0,0,0>
<dc855810336641b6,3cecd7aed66ff5ad,0,0>	<3cecd7aed66ff5ad,0,0,0>
<dc855810336641b6,3cecd7aed66ff5ad,0,0>	<44dc24617bd63967,3cecd7aed66ff5ad,0,0>

to make this more obvious.

> BTW, I did not test such scenario, because I can't compile drbd-0.8.x
> in my RedHat AS4 distro.

works for me.

-- 
: Lars Ellenberg                            Tel +43-1-8178292-55 :
: LINBIT Information Technologies GmbH      Fax +43-1-8178292-82 :
: Vivenotgasse 48, A-1120 Vienna/Europe    http://www.linbit.com :

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2006-12-20  8:40 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2006-12-20  7:06 [Drbd-dev] About the UUID home_king
2006-12-20  8:40 ` Lars Ellenberg

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox