Page MenuHomePhabricator

Rethink kubernetes etcd storage
Closed, DuplicatePublic

Description

There may be an issue with running etcd on raid SSDs with regard to latency. Investigate other storage and/or hosting possibilities.

There is some "guidance" about hardware sizing at: https://etcd.io/docs/v3.6/op-guide/hardware/

Options we could look into:

  • move etcd out of ganeti onto hardware (probably colocate with control-planes to not waste resources that much)
  • create I/O "optimized" ganeti clusters in codfw/eqiad with RAID1 instead of RAID5 to run etcd nodes on