GossipRouter: view shows mix of logical address and UUID for nodes
------------------------------------------------------------------
Key: JGRP-1340
URL:
https://issues.jboss.org/browse/JGRP-1340
Project: JGroups
Issue Type: Bug
Reporter: Bela Ban
Assignee: Bela Ban
Fix For: 2.12.2, 3.0
[Vivek Sar]
On Sat, Jun 18, 2011 at 4:56 AM, Bela Ban <belaban(a)yahoo.com> wrote:
> #1 When does this happen (after a restart of a node, or on
starting a new
> node) ?
[vivek]: this usually happens after a node has gone out of a group
(due to some intermittent network outage or communication issues) and
comes back to merge into the group. We have also seen this if we
restart the node without much time between stop and start (this was
the example above). So, it's kind of random - it happens both for
merge views and a new views.
> #2 Do the UUIDs disappear after some time, and the logical names
are shown ?
[vivek]: No, in my experience the UUID remain if it's the same
view,
but if the view changes (if any node goes in and out of a group) then
the logical name come back, otherwise the UUID remains.
> #3 Does this cause any issues, other than a not-so-nice view
with mixed
> logical and UUID addresses ?
[vivek]: It's both - in some cases we have
seen this causes node
isolation - where we are not able to get the merges (see related
https://issues.jboss.org/browse/JGRP-1326) - even after a node come
back it's not able to merge back. In some cases, the communication is
ok and it's just not-so-nice view.
> #4 What's you config ?
We are using Tunnel with PING
with two GRs (this is with JGroups 2.12.1),
(see attached config)
Thanks,
-vivek
>
> On 6/17/11 10:27 PM, vivek sar wrote:
> >>
> >> Hi,
> >>
> >> We are currently using JGroups 2.12.1 with Tunnel protocol (using
> >> PING for discovery). Once in a while we see node represented using
> >> UUID rather than their logical address. I think this issue is not
> >> specific to this version of JGroups as we have seen similar problem
> >> with 2.10 as well. The problem for us is not the view logging, but
> >> different view information on different nodes - some get the logical
> >> name and others just the UUID for the same view id. In past this kind
> >> of view (showing UUIDs instead of logical address) also indicated
> >> nodes and GRs out-of-sync on the node list, which in turn was causing
> >> merge issues.
> >>
> >> Questions,
> >>
> >> 1) When does this happen - getting both logical address and UUID in
> >> the same view?
> >> 2) Why different nodes with same view id get different node
> >> information (some get UUID, while others get the right logical
> >> address)?
> >> 3) How to fix it?
> >>
> >> For ex.,
> >>
> >> On Coordinator (collector_192.168.50.23):
> >>
> >> 2011-06-17 17:25:56,233 INFO
> >> [Incoming-4,192.168.50.22_group,collector_192.168.50.23:4576]
> >> RpcServiceManager - viewAccepted()-> New View:
> >> [collector_192.168.50.23:4576|51] [collector_192.168.50.23:4576,
> >> probe_192.168.50.32:4576, probe_192.168.50.24:4576,
> >> manager_192.168.50.22:4576, probe_192.168.50.56:4576]
> >>
> >> Node 1 (manager_192.168.50.22):
> >>
> >> 2011-06-17 17:25:56,429 INFO
> >> [Incoming-5,192.168.50.22_group,manager_192.168.50.22:4576]
> >> RpcServiceManager - viewAccepted()-> New View:
> >> [collector_192.168.50.23:4576|51] [collector_192.168.50.23:4576,
> >> 110fcd0e-6f80-7cf0-d76b-0813b913daf1,
> >> dd068d2f-ef21-1b66-1430-e8bfc862e6f9, manager_192.168.50.22:4576,
> >> probe_192.168.50.56:4576]
> >>
> >> Node 2 (probe_192.168.50.56):
> >>
> >> 2011-06-17 12:02:01,695 INFO
> >> [Incoming-10,192.168.50.22_group,probe_192.168.50.56:4576]
> >> RpcServiceManager - viewAccepted()-> New View:
> >> [collector_192.168.50.23:4576|51] [collector_192.168.50.23:4576,
> >> probe_192.168.50.32:4576, probe_192.168.50.24:4576,
> >> manager_192.168.50.22:4576, probe_192.168.50.56:4576]
> >>
--
This message is automatically generated by JIRA.
For more information on JIRA, see:
http://www.atlassian.com/software/jira