Disk Failure

• Monitored via SMART data
• Hades responsible for monitoring
• VM impact:
  • HA Event: NO
  • Failed I/O: NO
  • Latency: NO
• In event of failure, Curator scan occurs immediately
• Scans metadata to find data previously hosted on failed disk
• Re-replicates (distributes) to nodes throughout cluster
• All CVM’s participate

CVM Failure

• Failure = I/O’s redirected to other CVM’s in cluster
• VM impact:
  • HA Event: NO
  • Failed I/O: NO
  • Latency: Potentially higher given I/O’s are over network (not local)
• ESXi/Hyper-V handle via CVM Autopathing = leverages HA.py (happy) where routes are modified to forward traffic from internal address (192.168.5.2) to external IP of another CVM.
  • Keeps datastore intact
  • Once local CVM is back online, route is removed and local CVM takes back I/O
• KVM = iSCSI multipathing leveraged
  • Primary path = local CVM, other two paths = remote CVM

Node Failure

• VM impact:
  • HA Event: Yes
  • Failed I/O: NO
  • Latency: NO
• VM HA event will occur, restarting VMs on other nodes
• Curator will find data previously hosted on node and replicate
• In event node is down for prolonged period of time, downed CVM will be removed from metadata ring.
  • Will re-join after up and stable