When: Anytime, writes will be disabled
Prerequisites: https://wikitech.wikimedia.org/wiki/MariaDB/Primary_switchover
Checklist:
NEW primary: es1038
OLD primary: es1037
- Check configuration differences between new and old primary:
$ sudo pt-config-diff --defaults-file /root/.my.cnf h=es1037.eqiad.wmnet h=es1038.eqiad.wmnet 15 config differences Variable es1037 es1038 ========================= ========================= ========================= general_log_file es1037.log es1038.log gtid_domain_id 172006403 172007428 gtid_slave_pos 0-171966521-9791330871... 0-171966521-9791330871... hostname es1037 es1038 log_bin_basename /srv/sqldata/es1037-bin /srv/sqldata/es1038-bin log_bin_index /srv/sqldata/es1037-bi... /srv/sqldata/es1038-bi... pid_file /srv/sqldata/es1037.pid /srv/sqldata/es1038.pid read_only OFF ON report_host es1037.eqiad.wmnet es1038.eqiad.wmnet rpl_semi_sync_master_e... ON OFF rpl_semi_sync_slave_en... OFF ON server_id 172006403 172007428 slow_query_log_file es1037-slow.log es1038-slow.log thread_pool_stall_limit 100 10 wsrep_node_name es1037 es1038
- Disable writes in es6 merging https://gerrit.wikimedia.org/r/c/operations/mediawiki-config/+/1049555 and deploying it:
(in deployment.eqiad.org):
scap backport 1049555
- Check es6 is indeed read-only
Failover prep:
- Silence alerts on all hosts:
sudo cookbook sre.hosts.downtime --hours 1 -r "Primary switchover es6 T368401" 'A:db-section-es6'
- Set NEW primary with weight 0 (and depool it from API or vslow/dump groups if it is present).
sudo dbctl instance es1038 set-weight 0 sudo dbctl config commit -m "Set es1038 with weight 0 T368401"
- Topology changes, move all replicas under NEW primary
sudo db-switchover --timeout=25 --only-slave-move es1037 es1038
- Disable puppet on both nodes
sudo cumin 'es1037* or es1038*' 'disable-puppet "primary switchover T368401"'
- Merge gerrit puppet change to promote NEW primary: https://gerrit.wikimedia.org/r/1049550
Failover:
- Log the failover:
!log Starting es6 eqiad failover from es1037 to es1038 - T368401
- Switch primaries:
sudo db-switchover --skip-slave-move es1037 es1038 echo "===== es1037 (OLD)"; sudo db-mysql es1037 -e 'show slave status\G' echo "===== es1038 (NEW)"; sudo db-mysql es1038 -e 'show slave status\G'
- Promote NEW primary in dbctl, and remove read-only
sudo dbctl --scope eqiad section es6 set-master es1038 sudo dbctl config commit -m "Promote es1038 to es6 primary T368401"
- Restart puppet on both hosts:
sudo cumin 'es1037* or es1038*' 'run-puppet-agent -e "primary switchover T368401"'
Clean up tasks:
- Clean up heartbeat table(s).
sudo db-mysql es1038 heartbeat -e "delete from heartbeat where file like 'es1037%';"
- change events for query killer:
curl -sS 'https://gerrit.wikimedia.org/r/plugins/gitiles/operations/software/+/refs/heads/master/dbtools/events_coredb_master.sql?format=TEXT' | base64 -d | sudo db-mysql es1038 curl -sS 'https://gerrit.wikimedia.org/r/plugins/gitiles/operations/software/+/refs/heads/master/dbtools/events_coredb_slave.sql?format=TEXT' | base64 -d | sudo db-mysql es1037
- Update DNS: https://gerrit.wikimedia.org/r/1049551
- Update candidate primary dbctl and orchestrator notes
sudo dbctl instance es1037 set-candidate-master --section es6 true sudo dbctl instance es1038 set-candidate-master --section es6 false (dborch1001): sudo orchestrator-client -c untag -i es1038 --tag name=candidate (dborch1001): sudo orchestrator-client -c tag -i es1037 --tag name=candidate
- Check zarcillo was updated
- db-switchover should do this. If it fails, do it manually: https://phabricator.wikimedia.org/P13956
sudo db-mysql db1215 zarcillo -e "select * from masters where section = 'es6';"
- (If needed): Depool es1037 for maintenance.
sudo dbctl instance es1037 depool sudo dbctl config commit -m "Depool es1037 T368401"
- Change es1037 weight to mimic the previous weight es1038:
sudo dbctl instance es1037 edit
- Enable writes in es6 by merging and deploying revert of mediawiki config patch:
(in deployment.eqiad.org)
scap backport 1050096
https://gerrit.wikimedia.org/r/c/operations/dns/+/1050096
- Update/resolve this ticket.