Once the remaining use-case is migrated, and restbase has been fully provisioned, the 12 Cassandra nodes of the legacy cluster (6 ea in eqiad and codfw) can be forcibly decommissioned (shutdown and wiped), re-imaged, and bootstrapped into the new cluster.
- restbase1011.eqiad.wmnet
- restbase1016.eqiad.wmnet
- restbase1013.eqiad.wmnet
- restbase1017.eqiad.wmnet
- restbase1015.eqiad.wmnet
- restbase1018.eqiad.wmnet
- restbase2008.codfw.wmnet
- restbase2011.codfw.wmnet
- restbase2007.codfw.wmnet
- restbase2010.codfw.wmnet
- restbase2009.codfw.wmnet
- restbase2012.codfw.wmnet
In the course of T178177: Investigate aberrant Cassandra columnfamily read latency of restbase101{0,2,4}, it was determined that the performance of HP nodes configured for JBOD-like access via the smartarray controllers suffers (in comparison to the Dell hosts). We would like to test a configuration with the controller in HBA mode, to determine if this fares better. I propose the following sequence for re-imaging/provisioning:
- restbase1011.eqiad.wmnet (rack a, HP, to be configured in HBA mode)
- restbase1017.eqiad.wmnet (rack b, Dell)
- restbase1018.eqiad.wmnet (rack d, Dell)
- restbase2007.codfw.wmnet (rack b, HP, to be configured according to the outcome of #1)
- restbase2008.codfw.wmnet (rack c, HP, to be configured according to the outcome of #1)
- restbase2009.codfw.wmnet (rack d, HP, to be configured according to the outcome of #1)
- restbase1016.eqiad.wmnet (rack a, Dell)
- restbase1013.eqiad.wmnet (rack b, HP, to be configured according to the outcome of #1))
- restbase1015.eqiad.wmnet (rack d, HP, to be configured according to the outcome of #1))
- restbase2010.codfw.wmnet (rack b, Dell)
- restbase2011.codfw.wmnet (rack c, Dell)
- restbase2012.codfw.wmnet (rack d, Dell)