On Fri, Oct 05, 2018 at 11:34:10AM -0500, Ken Gaillot wrote: > The next big challenge is that high availability is becoming a subset > of the "orchestration" space in terms of how we fit into IT > departments. Systemd and Kubernetes are the clear leaders in service > orchestration today and likely will be for a long while. Other forms of > orchestration such as Ansible are also highly relevant. Tighter > integration with these would go a long way toward establishing > longevity.
Kubernetes seems to be mostly about stateless services and it kind of expects you have some external highly available data store. You can either buy that data store from your cloud provider or try to make it yourself, for example using Pacemaker and Galera. Maybe some agents for registering Pacemaker resources in the Kubernetes etcd or consul would be useful to connect the two worlds. On the other side, the big players seem to have settled on using distributed consensus protocols (Paxos, Raft, Zab) for building replicated state machines across multiple data centers (or even continents). It would be an interesting experiment if we could easily hook up Pacemaker with Zookeeper - selecting a DC node would require just creating a znode file. Master node can than store the cluster configuration as another znode and also ask other nodes to do some work through a job queue. On a global scale this would be quite slow (5-10 changes per second) and node fencing is not available but for some types of resources it might be useful. Some more reading material on the topic for the weekend :) https://landing.google.com/sre/book/chapters/managing-critical-state.html https://www.infoq.com/articles/cap-twelve-years-later-how-the-rules-have-changed -- Valentin _______________________________________________ Users mailing list: [email protected] https://lists.clusterlabs.org/mailman/listinfo/users Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf Bugs: http://bugs.clusterlabs.org
