Meeting Notes

CEP-2 for C* Kubernetes Operator

https://cwiki.apache.org/confluence/display/CASSANDRA/CEP-2+Kubernetes+Operator

Using Operator Framework's definition of operator maturity

  • Target Level 3 (plus scaling) for MVP = Full Lifecycle, provide pathway to Level 4, Listed on OperatorHub
  • Does ^ make sense?
    • why standardizing on operator framework? can get into those details as we go
    • upgrades can be tricky to do correctly, thoughts
    • node placement is difficult
    • community can define the things that are needed for running in production
    • can define as a project what these levels mean to the C* world
    • level of configuration would also have to be defined
    • do these things extend to components such as encryption?
    • token management is another example
    • SSL between nodes
  • Question on seed provider
    • bleeds into the docker image a bit, not black and white
    • could be a pluggable seed provider somewhere
  • Is operator delegating to sidecars?
    • there will be differing opinions, need to get to a agreed approach

Chat Log

David Capwell: One thing to call out, many k8s teams expect to be able to do a rolling restart of k8s, which kills all pods on the nodes.  They believe this is fine since operators recover; if we don’t bake replacing a node, many users cluster will vanish after a few days.


Patrick McFadin: 
There are a lot of k8s concepts that could destroy a c* cluster for sure


David Capwell: Yep yep


Patrick McFadin: We’ll need to be really clear on enumerating those. Are these things we need to tackle in Kubernetes?
 David, do you think the sidecar and operator projects are really one?


David Capwell: One no, helpful together? yes.  for example we could have a snitch which lives with the sidecar and that can fetch the details from k8s since it could know operator launched it.


Jan Schlicht:  https://kuttl.dev/

Stefan Miklosovic: This is how we are dealing with backup / restores in our operator if anybody is interested, just a food for thought https://github.com/instaclustr/cassandra-operator/wiki/Backup-and-Restore

Andreas Neumann [D2iQ]: I'm going to look at that for sure


Jim Dickinson : 
Nice!
 operator-sdk’s testing story was the worst choice we looked at


Jan Schlicht: 
kuttl.dev is what Zain was talking about

Zoom Video Recordings



Cassandra Kubernetes Operator Meeting 1(APAC/Western US Friendly)

Cassandra Kubernetes Operator Meeting 2(CEST/Eastern US Friendly)

  • No labels