Rejoining galera cluster fails with default wsrep_sst_method setting
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack-Ansible |
Fix Released
|
Critical
|
David Wilde | ||
Juno |
Fix Released
|
Critical
|
David Wilde | ||
Kilo |
Fix Released
|
Critical
|
David Wilde | ||
Trunk |
Fix Released
|
Critical
|
David Wilde |
Bug Description
wsrep_sst_method is set to xtrabackup, but this seems to be causing a problem when a node attempts to re-join a cluster by requesting a state transfer.
This issue was run into when an infrastructure host was offline for around a week for hardware replacements, changing the setting to wsrep_stt_
Reproduced on a Juno deployment:
From within single infra host's galera container,
service mysql stop
rm /var/lib/mysql/*
service mysql start
Attaching logs for the joining node and the donor node, which was chosen to provide the state transfer.
The actual failure looks to be coming from the /usr/bin/
# dpkg -l | grep xtrabackup
ii percona-xtrabackup 2.1.8-1 amd64 Open source backup tool for InnoDB and XtraDB
ii xtrabackup 2.1.8-1 all Transitional package for percona-xtrabackup
# dpkg -l | grep mariadb-galera
ii mariadb-
In Kilo we're working on getting the stack to use MariaDB10 w/ Xtrabackup-v2 which is a spec/PR awaiting reviews here: https:/ /review. openstack. org/#/c/ 178259/ . As for Juno we're looking into the problems and will about getting a fix in soonish.