Page MenuHomePhabricator

Switchover s8 master (db1109 -> db1126)
Closed, InvalidPublic

Description

When: During a pre-defined DBA maintenance windows

Prerequisites: https://wikitech.wikimedia.org/wiki/MariaDB/Primary_switchover

  • Team calendar invite

Affected wikis:: https://noc.wikimedia.org/conf/highlight.php?file=dblists/s8.dblist

Checklist:

NEW primary: db1126
OLD primary: db1109

  • Check configuration differences between new and old primary:
sudo pt-config-diff --defaults-file /root/.my.cnf h=db1109.eqiad.wmnet h=db1126.eqiad.wmnet

Failover prep:

  • Silence alerts on all hosts:
sudo cookbook sre.hosts.downtime --hours 1 -r "Primary switchover s8 T330988" 'A:db-section-s8'
  • Set NEW primary with weight 0 (and depool it from API or vslow/dump groups if it is present).
sudo dbctl instance db1126 set-weight 0
sudo dbctl config commit -m "Set db1126 with weight 0 T330988"
  • Topology changes, move all replicas under NEW primary
sudo db-switchover --timeout=25 --only-slave-move db1109 db1126
  • Disable puppet on both nodes
sudo cumin 'db1109* or db1126*' 'disable-puppet "primary switchover T330988"'

Failover:

  • Log the failover:
!log Starting s8 eqiad failover from db1109 to db1126 - T330988
  • Set section read-only:
sudo dbctl --scope eqiad section s8 ro "Maintenance until 06:15 UTC - T330988"
sudo dbctl config commit -m "Set s8 eqiad as read-only for maintenance - T330988"
  • Check s8 is indeed read-only
  • Switch primaries:
sudo db-switchover --skip-slave-move db1109 db1126
echo "===== db1109 (OLD)"; sudo db-mysql db1109 -e 'show slave status\G'
echo "===== db1126 (NEW)"; sudo db-mysql db1126 -e 'show slave status\G'
  • Promote NEW primary in dbctl, and remove read-only
sudo dbctl --scope eqiad section s8 set-master db1126
sudo dbctl --scope eqiad section s8 rw
sudo dbctl config commit -m "Promote db1126 to s8 primary and set section read-write T330988"
  • Restart puppet on both hosts:
sudo cumin 'db1109* or db1126*' 'run-puppet-agent -e "primary switchover T330988"'

Clean up tasks:

  • Clean up heartbeat table(s).
sudo db-mysql db1126 heartbeat -e "delete from heartbeat where file like 'db1109%';"
  • change events for query killer:
events_coredb_master.sql on the new primary db1126
events_coredb_slave.sql on the new slave db1109
  • Update candidate primary dbctl and orchestrator notes
sudo dbctl instance db1109 set-candidate-master --section s8 true
sudo dbctl instance db1126 set-candidate-master --section s8 false
(dborch1001): sudo orchestrator-client -c untag -i db1126 --tag name=candidate
(dborch1001): sudo orchestrator-client -c tag -i db1109 --tag name=candidate
sudo db-mysql db1115 zarcillo -e "select * from masters where section = 's8';"
  • (If needed): Depool db1109 for maintenance.
sudo dbctl instance db1109 depool
sudo dbctl config commit -m "Depool db1109 T330988"
  • Change db1109 weight to mimic the previous weight db1126:
sudo dbctl instance db1109 edit
  • Apply outstanding schema changes to db1109 (if any)
  • Update/resolve this ticket.

Event Timeline

Change 893430 had a related patch set uploaded (by Gerrit maintenance bot; author: Gerrit maintenance bot):

[operations/puppet@production] mariadb: Promote db1126 to s8 master

https://gerrit.wikimedia.org/r/893430

Marostegui triaged this task as Medium priority.
Marostegui moved this task from Triage to Blocked on the DBA board.
Marostegui subscribed.

To be done after the eqiad row A switch maintenance is done (T329073) as the candidate master is on row A

The script needed to be updated after the dc switchover now (it has change for dns and such while it's not needed), I fixed it so you can fix this ticket or just abandon the patch and call switchmaster again.

Yeah but it has the check option for it and the stuff for --read-only-master in switchover, the usual difference between primary and non-primary dc.

Going to regenerate it

Change 893430 abandoned by Marostegui:

[operations/puppet@production] mariadb: Promote db1126 to s8 master

Reason:

https://gerrit.wikimedia.org/r/893430