Operations grimoire/Restart a Docker engine

This containers historical notes from 2016. This page reflect only partially the current state of our Docker engines. Yet, the broker/Wearg thing is for example still current in 2023.

Restart

A lot of vital components are managed by Docker.

Ideally, we should have redundancy to avoid container lost.

1. Restart Docker engine
2. Restart containers, either manually, either with the docker-containers systemd unit
3. Run production tests

Services needing manual tweaking after container restart

Phabricator instances

If DevCentral complains SSH hosting or daemons isn't available:

1. sv status sshd-hosting (if needed sv start sshd-hosting)
2. sv restart phd

Other instances need only a sv restart phd.

Broker

Wearg needs to be manually reconnected to the broker:

1. .tcl mq disconnect
2. .tcl utimers and kill remaining timers if needed
3. .tcl mq broker::connect
4. .tcl broker::on_tick

That requires a owner access to Wearg (ping Dereckson).

Troubleshooting

When MySQL isn't reachable

If the MySQL container (acquisitariat) IP changed, you need to tweak /etc/hosts in every depending container (Phabricator instances, cachet, pad, login) and ensure there is a <correct IP> mysql line.

What containers need MySQL and symptoms when not reachable?

Container	MySQL priority	Symptom when not running
silly_bardeen	Needed for some CI tasks	Jenkins jobs test-auth-grove-* will fail: PHPUnit test issue: \Tests\Models\UsersTest::testTryGetFromExternalSource PDOException: SQLSTATE[HY000] [2002] php_network_getaddresses: getaddrinfo failed: Name or service not known To test if all works: $ su app $ cd ~/workspace/test-auth-grove-php $ phpunit --no-coverage OK (37 tests, 58 assertions)
cachet	High	App doesn't work
devcentral	High	App doesn't work
etherpad	High	Container doesn't start
wolfphab	High	App doesn't work

When a container doesn't want to restart

Try first to see what happens with docker logs <container name>.

If that doesn't work, go to the services section of the grimoire and reprovision it.

Commit a backup Docker image based on the current content with with docker commit <name> <name>-bak. That will allow investigation.

Error response from daemon: Unknown runtime specified oci

The container doesn't use the current runtime.

It's currently docker-runc but was previously for a short time oci.

You can, change the runtime in the container configuration following instructions from this post.

Do that only for containers you don't want to respin (e.g. devcentral). Others you can safely rm and respin instead.