notes.txt

1) POds recreate and move faster:
    Node down detection
    Read write once:
        Safe lun mapping snapshot revokation

2) Active-passive:
    * corosync/pacemaker:
    * leader election:
      * ucarp: small
      * vrrpd: big 34mb in mariner
      * etcd: promotion
        * convergenge under <10s

3)

k8s things:
1) run one pod on node foo.
2) node foo goes down
3) after k8s detects its been offline (180 seconds say), it marks node offline
4) k8s scheduler starts clock ticking on how long the pod can be on a down node (i think this is currently around 240 seconds by default)
5) time has expired, and so it starts to find a new node to run the pod on
6) schedules pod to run on node bar
6a) creates pod sandbox on bar
6b) undercloud csi starts to attach pv to node
6c) pod starts nfs container (may need to pull it if not already there)
6d) mounts filesystem
6e) starts nfs server





NFS - Active-Passive HA:

2 or more Pods:
 *) Each pod has the volume from the pure mapped to it (ie there is a lun attached to the host)
 *) Each contains a leader election mechanim (pacemaker/ucarp/vrrpd) that on deciding its the:
    leader(the active node):
      a) takes snapshot of volume via pure api
      b) mounts filesystem of volume
      c) starts nfs server
      d) moves VIP to it
    followers(the passive nodes):
      a) stops nfs server if running
      b) unmouts fileystem if mounted

Questions:
1) how do we do the VIP:
        a) k8s `native`?
        b) keepalived or similar?
2) how to monitor
3) need to look back throuh kusto of deployments to look for potential 'false' triggers etc like network drops and feed this into how we do the leader election
4) how to validate/ensure clients are happy