split brain cluster from leader handling failure on older jujus

[JuanJo Ciarlante]

* Deployment:
- openstack HA with 1507 charms release
- percona-cluster r63 (service name: mysql )
- juju 1.20.14

This HA deployment reliably fails to properly
cluster percona units:

- wsrep_cluster_size = 1 on each unit
-> http://paste.ubuntu.com/12124888/

- openstack DBs only present on mysql/0
-> http://paste.ubuntu.com/12124891/

- log excerpts showing all units logging
 "Leader unit - bootstrap required=True"
-> http://paste.ubuntu.com/12124893/

Afaics looks like r62 changes, confirming same deployment
clusters ok with r61.

[Liam Young (gnuoy)]

Please could try setting min-cluster-size when doing the deployment.

[JuanJo Ciarlante (jjo)]

Confirming setting min-cluster-size=3 fixes the split brain issue, ie finding Ok:
wsrep_cluster_size 3
, but I'm now seeing a never-ending loop on the 3 units calling
cluster-relation-changed hook, filed lp#1488140 - feel free to close
this one.

[David Ames (thedac)]

JuanJo,

Have you had a chance to try min-cluster-size. The charm relies on this.

For the record I just tested the amulet test with and without min-cluster-size set. When it is not set the results are as you describe. With it set the cluster comes up.

Note you can run the test directly (without dealing with juju test):

 python AMULET_VIP=$IP1 AMULET_OS_VIP=$IP2 python tests/42-test-bootstrap-multi-min.py

[Launchpad Janitor (janitor)]

[Expired for percona-cluster (Juju Charms Collection) because there has been no activity for 60 days.]