Working with the bundle Consul service

Overview

As part of its High Availability stack, GitLab EEP includes a bundled version of Consul that can be managed through /etc/gitlab/gitlab.rb.

A Consul cluster consists of multiple server agents, as well as client agents that run on other nodes which need to talk to the consul cluster.

Operations

Checking cluster membership

To see which nodes are part of the cluster, run the following on any member in the cluster

# /opt/gitlab/embedded/bin/consul members
Node            Address               Status  Type    Build  Protocol  DC
consul-b        XX.XX.X.Y:8301        alive   server  0.9.0  2         gitlab_consul
consul-c        XX.XX.X.Y:8301        alive   server  0.9.0  2         gitlab_consul
consul-c        XX.XX.X.Y:8301        alive   server  0.9.0  2         gitlab_consul
db-a            XX.XX.X.Y:8301        alive   client  0.9.0  2         gitlab_consul
db-b            XX.XX.X.Y:8301        alive   client  0.9.0  2         gitlab_consul

Ideally all nodes will have a Status of alive.

Restarting the server cluster

Note: This section only applies to server agents. It is safe to restart client agents whenever needed.

If it is necessary to restart the server cluster, it is important to do this in a controlled fashion in order to maintain quorum. If quorum is lost, you will need to follow the consul outage recovery process to recover the cluster.

To be safe, we recommend you only restart one server agent at a time to ensure the cluster remains intact.

For larger clusters, it is possible to restart multiple agents at a time. See the Consul consensus document for how many failures it can tolerate. This will be the number of simulateneous restarts it can sustain.

Troubleshooting

Consul server agents unable to communicate

By default, the server agents will attempt to bind to '0.0.0.0', but they will advertise the first private IP address on the node for other agents to communicate with them. If the other nodes cannot communicate with a node on this address, then the cluster will have a failed status.

You will see messages like the following in gitlab-ctl tail consul output if you are running into this issue:

2017-09-25_19:53:39.90821     2017/09/25 19:53:39 [WARN] raft: no known peers, aborting election
2017-09-25_19:53:41.74356     2017/09/25 19:53:41 [ERR] agent: failed to sync remote state: No cluster leader

To fix this:

  1. Pick an address on each node that all of the other nodes can reach this node through.
  2. Update your /etc/gitlab/gitlab.rb
   consul['configuration'] = {
     ...
     bind_addr: 'IP ADDRESS'
   }
  1. Run gitlab-ctl reconfigure

Outage recovery

If you lost enough server agents in the cluster to break quorum, then the cluster is considered failed, and it will not function without manual intervenetion.

Recreate from scratch

By default, GitLab does not store anything in the consul cluster that cannot be recreated. To erase the consul database and reinitialize

# gitlab-ctl stop consul
# rm -rf /var/opt/gitlab/consul/data
# gitlab-ctl start consul

After this, the cluster should start back up, and the server agents rejoin. Shortly after that, the client agents should rejoin as well.

Recover a failed cluster

If you have taken advantage of consul to store other data, and want to restore the failed cluster, please follow the Consul guide to recover a failed cluster.