When: During a pre-defined DBA maintenance windows
Affected wikis:: https://noc.wikimedia.org/conf/highlight.php?file=dblists/s7.dblist
Checklist:
NEW primary: db1181
OLD primary: db1236
- Check configuration differences between new and old primary:
$ sudo pt-config-diff --defaults-file /root/.my.cnf h=db1236.eqiad.wmnet h=db1181.eqiad.wmnet 17 config differences Variable db1236 db1181 ========================= ========================= ========================= general_log_file db1236.log db1181.log gtid_binlog_pos 171966638-171966638-73... 171966638-171966638-73... gtid_binlog_state 171966638-171966638-73... 171966638-171966638-73... gtid_current_pos 0-180359185-3359637071... 0-180359185-3359637071... gtid_domain_id 171966638 171974795 gtid_slave_pos 0-180359185-3359637071... 0-180359185-3359637071... hostname db1236 db1181 innodb_buffer_pool_size 405874409472 404800667648 log_bin_basename /srv/sqldata/db1236-bin /srv/sqldata/db1181-bin log_bin_index /srv/sqldata/db1236-bi... /srv/sqldata/db1181-bi... pid_file /srv/sqldata/db1236.pid /srv/sqldata/db1181.pid report_host db1236.eqiad.wmnet db1181.eqiad.wmnet rpl_semi_sync_master_e... ON OFF rpl_semi_sync_slave_en... OFF ON server_id 171966638 171974795 slow_query_log_file db1236-slow.log db1181-slow.log wsrep_node_name db1236 db1181
Failover prep:
- Silence alerts on all hosts:
sudo cookbook sre.hosts.downtime --hours 1 -r "Primary switchover s7 T376868" 'A:db-section-s7'
- Set NEW primary with weight 0
sudo dbctl instance db1181 set-weight 0 sudo dbctl config commit -m "Set db1181 with weight 0 T376868"
- Depool NEW from any specific group (API, vslow, dump) if present.
sudo dbctl instance db1181 edit # If some changes were made: sudo dbctl config commit -m "Remove db1181 from API/vslow/dump T376868"
- Topology changes, move all replicas under NEW primary
sudo db-switchover --timeout=25 --replicating-master --read-only-master --only-slave-move db1236 db1181
- Disable puppet on both nodes
sudo cumin 'db1236* or db1181*' 'disable-puppet "primary switchover T376868"'
- Merge gerrit puppet change to promote NEW primary: FIXME
Failover:
- Log the failover:
!log Starting s7 eqiad failover from db1236 to db1181 - T376868
- Switch primaries:
sudo db-switchover --replicating-master --read-only-master --skip-slave-move db1236 db1181 echo "===== db1236 (OLD)"; sudo db-mysql db1236 -e 'show slave status\G' echo "===== db1181 (NEW)"; sudo db-mysql db1181 -e 'show slave status\G'
- Promote NEW primary in dbctl
sudo dbctl --scope eqiad section s7 set-master db1181 sudo dbctl config commit -m "Promote db1181 to s7 primary T376868"
- Restart puppet on both hosts:
sudo cumin 'db1236* or db1181*' 'run-puppet-agent -e "primary switchover T376868"'
Clean up tasks:
- Clean up heartbeat table(s).
sudo db-mysql db1181 heartbeat -e "delete from heartbeat where file like 'db1236%';"
- 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 db1181 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 db1236
- Update candidate primary dbctl and orchestrator notes
sudo dbctl instance db1236 set-candidate-master --section s7 true sudo dbctl instance db1181 set-candidate-master --section s7 false sudo cumin 'dborch*' 'orchestrator-client -c untag -i db1181 --tag name=candidate' sudo cumin 'dborch*' 'orchestrator-client -c tag -i db1236 --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 = 's7';"
- (If needed): Depool db1236 for maintenance.
sudo dbctl instance db1236 depool sudo dbctl config commit -m "Depool db1236 T376868"
- Change db1236 weight to mimic the previous weight db1181:
sudo dbctl instance db1236 edit
- Apply outstanding schema changes to db1236 (if any)
- Update/resolve this ticket.