[ClusterLabs] Corosync ifdown / ifup crash with Assertion `token_memb_entries >= 1' failed.
jfriesse at redhat.com
Wed Jun 3 08:51:39 EDT 2015
Dennis Jacobfeuerborn napsal(a):
> On 02.06.2015 12:57, Jan Friesse wrote:
>> Alexander T napsal(a):
>>> Thank you for the explanation. What do you mean by "knet migration" in
>>> Corosync 3.0?
>> We are planning to replace current network related code with knet.
>> Basically currently we have multicast, udpu and IBA (infiniband)
>> transport. Problem with multicast is mainly in badly configured
>> switches, because many network admins believe multicast is evil + some
>> switches are really bad in multicast. IBA code is almost unmaintained.
>> So last is UDPU. UDPU works quite well, but still, it's just 1:1 rewrite
>> of multicast where multicast is simulated by sending messages to all
>> configured members. Still all weird requirements and weird code base
> I don't think it's switches that are the problem these days but the fact
> that in many environments multicast isn't available at all (e.g. clouds)
I'm still forgoing this reason, thanks for comment. Yes, all the
cloud/VM/... environments are simply ether not capable of multicast at
all, or special configuration is needed.
> and even if it is it might still be a pain to get working. If you want
> to create a cluster of virtual machines for example you need to know the
> magic incantation for /proc to make multicast work.
> Given that a lot of people only set up 2-node clusters anyway unicast is
> the easiest way to get things working.
>> Last but not least is RRP. RRP itself works very well sadly it works in
>> totally different way then most of people expects.
>> Solution is knet (http://www.kronosnet.org/). Knet should give corosync
>> proper support for multiple NICs (like software bonding), very fast
>> reaction time to link failure, proper handling of MTU, ...
> That has no documentation whatsoever and what little communication there
> was on the mailing lists was in 2010. Are you sure it's a good idea to
> rely on a project that seems to be dead?
It's because it's not yet in wide spread use (+ knet itself is not that
interesting. what we are interesting in is libknet). But take a look to
github commits and you will find out that knet is very alive. So yes,
I'm sure it's good idea to rely on knet :)
> Users mailing list: Users at clusterlabs.org
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org
More information about the Users