I'm pretty sure that is not the case. If you create a minimized
example/reproducer I will take a look. I prefer that you base it on the
SimpleClusterApp example:
scala:
It looks like logic of identifying new incarnation based on ip/port and
downing previous incarnation of actorsystem happens only on seed nodes and
not on members. So if I crash seed node and restart it again, the
membership list will have the new incarnation of the seed in its membership
list,
No, there is no majority decision here. Perhaps you don't join the right
node. You should have both in seed-nodes and in same order. It should be
clear from info level logging what is going on.
tis 4 apr. 2017 kl. 12:26 skrev Unmesh Joshi :
> Curiously, my observation is
Curiously, my observation is that if instead of two, I have four node
cluster and crash/restart a node with same host/port, I do not get this
warning. I get this only on two node cluster, which made me think that
there is majority needed to mark the new incarnation as 'seen' and then
down the
One way of downing is to join again with same host:port. That will trigger
downing of previous incarnation and when removal is done the new
incarnation can join by trying to join again. The seed-nodes joining will
retry the joining automatically.
However, sooner or later you will need a real
Indeed. This is the relevant bit of docs I believe
(http://doc.akka.io/docs/akka/2.4.17/common/cluster.html#Membership):
The node identifier internally also contains a UID that uniquely
identifies this actor system instance at thathostname:port. Akka uses
the UID to be able to reliably
No, it needs to be Downed.
On Tue, Apr 4, 2017 at 9:50 AM, Unmesh Joshi wrote:
> Hi,
>
> If I restart the crashed node on same host and port, it should be
> reachable now and consensus should be reached isnt it?
>
> Thanks,
> Unmesh
>
> On Tuesday, 4 April 2017 13:09:22
Hi,
If I restart the crashed node on same host and port, it should be reachable
now and consensus should be reached isnt it?
Thanks,
Unmesh
On Tuesday, 4 April 2017 13:09:22 UTC+5:30, Michal Borowiecki wrote:
>
> Hi Unmesh,
>
> AFAIK, the crashed node has to be downed (whether manually or
>
Hi Unmesh,
AFAIK, the crashed node has to be downed (whether manually or
automatically) for the cluster to reach convergence.
Only once there are no unreachable nodes observed by any member can the
leader resume it's duties and allow the new member (your re-started
instance) to join.
For
Is it possibly because in a two node cluster, there can never be majority (
> 50%) nodes agreeing on membership to mark a node as 'seen'?
On Tuesday, 4 April 2017 12:46:17 UTC+5:30, Unmesh Joshi wrote:
>
> Hi,
>
> I have a two node cluster in a cluster. If I crash one of the nodes
>
10 matches
Mail list logo