These clusters are *mostly* unused, and need to be reorganized.
EQUIAD (rdb1*):
Our end goal is to have two master/slave couples, rdb1009/1010 and rdb1005/1006, and decommission the remaining servers.
- Check which clients exist, and where do they connect. We are aware of change-propagation connecting to rdb1001/1003 via nutcracker
- Create a new puppet role for redis::misc::master, which should use profile::redis::master
- Create a corresponding role for redis::misc::slave
- Install rdb1009 as master, 1010 as slave; Define their puppet cluster variable as redis_main (also do the same for all servers below.
- Reimage rdb1005 with stretch as master, rdb1006 as slave
- Migrate existing clients to rdb1005/1009
- decommission rdb1001-1004 and rdb1007-1008
CODFW (rdb2*)
Our goal here is to reconfigure rdb2003-2006 as two master/slave couples, and return 2001/2002 to the spares list
- Check which clients exist, and where do they connect. We are aware of change-propagation connecting to rdb2001/2003 via nutcracker
- Reimage rdb2005/2006 with the same configuration as 1005/1006
- Migrate all clients to use those
- Reimage rdb2003/2004 in a similar fashion
- Redistribute clients across the two groups
- Reimage rdb2001/2002 as spares (were not re-imaged since this will be done by dc ops)