Andrei Dobre

Depends on ability to add meta fields on the pod. Liked to #59 .

The process of recovering a node or updating the cluster can be long and with many steps. We have to introduce as many types of failures as possible to see...

- [x] All K8s resources are up (operator deployment, pods, configmap, rediscluster, rbac, clusterrole, serviceaccount, services) - [x] The operator can handle multiple cycles of creation-destruction of the RedisCluster resource...

The current setup assumes the controller pods are deployed independent of the CR. The controllers can't handle the case where a RedisCluster is recreate and the case with multiple RedisClusters...

Before the operator is production ready we need to measure the load put on the Kubernetes master and etcd by the operator and optimise the API calls if needed. We...

Currently flag passing via make -> docker -> go binary is messy and does not work well. - [ ] The operator should also be configurable so that it is...

We have extensive E2E coverage but the tests take a long time to run and slows down development. We must have them run as part of the CI pipeline.

It would be a good idea to run each E2E test in a new namespace. This helps with testing the RBAC setup better, resource cleanup and allows for parallel E2E...

The controller only reacts to changes in Kubernetes resources, mainly Pods and the containers inside. If Redis fails without causing a change to the Kubernetes objects the reconccile loop will...