We've forked this project because the previous devops engineers were running this deployment and we don't want to do all the customization to get this working with mariadb native container for newer versions.

SecurtyESys Updates will be streamed in here for anything that we add to the codebase to make this more agreeable to what we are using it for.

There are a rather large number of variables that are required for multi-master WAN setups; planning on streaming these in the build process natively so we don't have to have a 20 foot mysqld deployment line.

We hope that this image will not be required in the future pending the integration of better Galera support in the official image. eg PR 24.

This image uses entrypoint modifications similar to the ones by Kristian Klausen to provide (better) Galera support for the offcial mariadb:10.1 image.

Also included is Galera Arbitrator (aka garbd) which allows you to maintain quorum with a two node cluster. (Not required when running 3 or more nodes, helps when spanning WAN segments).

• WSREP_NODE_ADDRESS - IP or domain of host interface eg WSREP_NODE_ADDRESS=10.0.0.1
• WSREP_CLUSTER_ADDRESS - List of cluster nodes and ports eg WSREP_CLUSTER_ADDRESS=gcomm://10.0.0.1:4567,10.0.0.2:4567,10.0.0.3:4567
• WSREP_CLUSTER_NAME - Default my_wsrep_cluster
• WSREP_NODE_NAME - Defaults to the container hostname. You'll want to set this to something that doesn't change.

Garbd is available. Just specify garbd as the command.

docker run -d --net host --name galera-garbd \
  -e WSREP_CLUSTER_ADDRESS=$WSREP_CLUSTER_ADDRESS \
  panubo/mariadb-galera \
    garbd

Node 1:

docker run -d --net host --name galera \
  -e WSREP_NODE_ADDRESS=$WSREP_NODE_ADDRESS \
  -e WSREP_CLUSTER_ADDRESS=$WSREP_CLUSTER_ADDRESS \
  -e MYSQL_ROOT_PASSWORD={{ mysql_root_password }} \
  -p 3306:3306 \
  -p 4567:4567/udp \
  -p 4567-4568:4567-4568 \
  -p 4444:4444 \
  -v /mnt/data/galera.service/mysql:/var/lib/mysql:Z \
  panubo/mariadb-galera \
    mysqld \
    --wsrep-new-cluster

Node 2-N:

Create empty mysql dir to skip database initialisation. (Kludge!)

mkdir -p /mnt/data/galera.service/mysql/mysql

Start the container normally (without --wsrep-new-cluster).

docker run -d --net host --name galera \
  -e WSREP_NODE_ADDRESS=$WSREP_NODE_ADDRESS \
  -e WSREP_CLUSTER_ADDRESS=$WSREP_CLUSTER_ADDRESS \
  -p 3306:3306 \
  -p 4567:4567/udp \
  -p 4567-4568:4567-4568 \
  -p 4444:4444 \
  -v /mnt/data/galera.service/mysql:/var/lib/mysql:Z \
  panubo/mariadb-galera \
    mysqld

Recovery when quorum is lost is fairly simple:

First, stop on all nodes. Eg (if using a Systemd unit to run Galera):

systemctl stop galera.service

Secondly, start node with most complete / recent data set with --wsrep-new-cluster argument. EG:

docker run -d --net host --name galera-init \
  -e WSREP_NODE_ADDRESS=$WSREP_NODE_ADDRESS \
  -e WSREP_CLUSTER_ADDRESS=$WSREP_CLUSTER_ADDRESS \
  -e MYSQL_ROOT_PASSWORD=$MYSQL_ROOT_PASSWORD \
  -p 3307:3306 \
  -p 4567:4567/udp \
  -p 4567-4568:4567-4568 \
  -p 4444:4444 \
  -v /mnt/data/galera.service/mysql:/var/lib/mysql:Z \
  panubo/mariadb-galera \
    mysqld \
    --wsrep-new-cluster

Finally, bring up other nodes normally. Eg. (Systemd example)

systemctl start galera.service

1. Whilst it isn't strictly necessary to use the host network (--net host), there seems to be an issue (bug?) whereby Galera gets both the host and the (duplicated) Docker network IP assigned to the node. This causes issues when multiple nodes fail and attempt to rejoin the cluster.

2. Garbd requires an explicit port if it blows up with "Exception in creating receive loop." See issue 312.