Am I right that the configuration that is actually being used will come from the leader? Well, except the exception like the bootstrap list...
Asking because I'd like to have a clear process to change the configuration in a way that at the end of the process I can be sure that all nodes use the new configuration. So assuming that I have a 2+ node MM2 cluster, all nodes has the same configuration and I'd like to change the configuration for the cluster, I plan to restart each node after each other with the new configuration. When I change configuration, I change it to the same on all nodes. Am I right that at the end I can be sure that the cluster uses the new configuration, because on the process I will bounce the leader for sure with the new config. I actually plan to run it on Kubernetes, so I am using a rolling restart on Kubernetes. I only have one mirroring path, but I guess this would work if I have more paths as well. Peter On Thu, 16 Jan 2020 at 16:36, Ryanne Dolan <ryannedo...@gmail.com> wrote: > MM2 nodes only communicate via Kafka -- no connection is required between > them. > > To reconfigure, a rolling restart probably won't do what you expect, since > the configuration is always dictated by a single leader. Once the leader is > bounced, it will broadcast the new configuration via Kafka. If you bounce a > follower, it will still use the old configuration. > > It's slightly more complicated than that actually, since there are multiple > connectors in each cluster->cluster "herder". When there are multiple > clusters being replicated, there may be dozens of leaders. So a rolling > restart might be a good idea for larger deployments. > > And technically there are some properties that do affect followers, e.g. a > follower will read connection info like bootstrap.servers directly from > mm2.properties, not from the leader via Kafka. Obviously that would be a > chicken-egg problem otherwise! So a rolling restart would be prudent when > making such changes. > > So I guess, generally speaking, rolling restarts are a good idea -- just be > advised that it won't necessarily behave as you expect. > > Ryanne > > On Thu, Jan 16, 2020, 7:55 AM Péter Sinóros-Szabó > <peter.sinoros-sz...@transferwise.com.invalid> wrote: > > > Hi, > > > > I run two instances of MM2 with the command connect-mirror-maker.sh > > > > Q1., Is there any requirement to cluster MM2? Like a network connection > > between the nodes? How do MM2 coordinate the work between nodes? > > > > Q2., Assuming I run two instances and want to update the configuration, > > should it work if I rolling restart the nodes after each other? It seems > > that after the restart both nodes still run with the old configuration. > So > > what should be the correct way of reconfiguring a MM2 cluster? > > > > Cheers, > > Peter > > >
