[Drbd-dev] Why is invalidate command rejected if current node is
Simon.Graham at stratus.com
Sat Jul 15 17:15:39 CEST 2006
Currently, any attempt to invalidate the local disk is rejected if the
local node is Primary (and similarly, invalidate-remote is rejected if
we think the partner is Primary) - given the fact that DRBD is quite
able to handle the case where the Primary does not have good data, why
is this test there? This actually makes it hard, for example, to write
automated tests that invalidate the disk since you have to set the node
secondary which can cause running programs to fail because they can no
longer access the disk.
It's even worse running with dual-primaries in DRBD-8 - I plan on
running with both sides Primary all the time so I shouldn't have to
worry about Secondary at all, but if I want to do a test that
invalidates the disk, I have to set a node secondary first.
So - is there any really good reason for rejecting invalidate in this
PS: Another [albeit minor] problem with the current code is that the
error is reported incorrectly - the error message is 'Only in cstate
'Connected' possible' which is confusing when the invalidate is rejected
due to being primary!
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the drbd-dev