To be done one week before the switchover:
- Create calendar events: Stop DB maintenance (6 days before the switch), enable circular replication (3 days before the switch), disable circular replication (1 day after the switch), resume DB maintenance (after the disabling circular replication) done by @Marostegui
- Ensure the datapoints on https://zarcillo.wikimedia.org/ui/weights are fresh
- compare weights in s* and x3 (they should all be 300 and one replica per section in each DC should be 100)
- compare weights in es* (they should all be 100 on replicas and 0 on the master for writable sections and 100 for all hosts in RO sections)
- compare weights x1 (they should be all 300 and one 100 in each DC)
- check all pc* and ms* sections are pooled
- ensure that 1 master 1 candidate master per section declared and usual topology constraints are OK.
- monitoring notifications enabled on all relevant hosts
- clean orchestrator output
- Write a dns patch for the new masters and compare DNS, puppet, dbctl and real replica topology (e.g. orchestrator/dbtree) agrees using https://switchmaster.toolforge.org/dc-switch
- [skipped] Run compare table among eqiad and codfw masters to ensure they all have the same data (see missing checks at T375507)
- Run diff between live configuration among eqiad and codfw masters (binlog retention, gtid enabled/disabled, server version, etc.)