Page MenuHomePhabricator

WMCS hardware services: 3-node HA redundancy model
Closed, InvalidPublic

Description

It has been discussed multiple times over the years that we should move our hardware services from a 2-node HA redundancy model into a 3-node model.

This ticket is to track the work to implement this:

  • cloudlb eqiad
  • cloudlb codfw
  • cloudgw eqiad
  • cloudgw codfw -- T377568
  • cloudnet eqiad
  • cloudnet codfw -- T377568
  • cloudservices eqiad
  • cloudservices codfw -- T377568

Event Timeline

cloudlb in codfw contains 3 nodes at the moment:

  • cloudlb2002-dev
  • cloudlb2003-dev
  • cloudlb2004-dev (replacing cloudlb2001-dev T377126)

Closed as these hosts:

  • are non-stateful
  • have low-enough loads that a single server can handle usual traffic levels

So a third node would have very little benefits.

ftr I am also good with sticking with 2-node setups.