[DRBD-user] Suboptimal behavior in abnormal startup?

Lars Ellenberg Lars.Ellenberg at linbit.com
Tue Jun 8 16:24:00 CEST 2004


/ 2004-06-08 17:42:37 +0400
\ Eugene Crosser:
> Gents,
> 
> I have a feeling (but not 100% certain) that 0.7 does not behave quite
> right when a fresher note is forcibly made secondary.  Consider this
> scenario:
> 
> - node A: primary&syncsource goes down
> - node B: secondary&synctarget becomes primary&wfconnection
> - Oops!  You decide to continue using node A, take down B and boot A.
>   A becomes primary&wfconnection
> - You boot B.
> 
> At this moment B thinks that it sould have become syncsource, but
> because it sees that A is already primary it would not interfere and
> becomes synctarget.  So far so good.  *BUT* it says that "no sync
> needed" and so, supposedly, the changes that where made on node A after
> boot are not propagated to B, and so the data on B is in fact
> inconsistent despite it says that it's consistent.
> 
> Am I missing something?
> 
> If not, I think it would be better to forcibly invalidate B, to prevent
> the chance of using bad data...

there was still a FIXME in the code,
where I happen to work on right now...

  :)

	Lars Ellenberg



More information about the drbd-user mailing list