document service restart/hup behavior for config and version changes
When omnibus is upgraded and a reconfigure is issued some tasks may happen for the user automatically:
- gitaly restarted if the version changes
- unicorn restarted (given a hup)
- gitlab-workhorse is restarted when the version is updated
Other things don't happen automatically, like a pages restart if the gitlab-pages version changes as well as other components I assume such as workhorse, nginx, etc.
While I don't think we should change the existing behavior for preserving backwards compatability with community deployment scripts, it would be very helpful for GitLab.com deployments to have consistent behavior across all services.
This issue is to propose that we have a ENV var set that would skip service restarts for all services, on reconfigure.
Service summary
service | restart for config change | restart for version change | restart by deployer on version change |
---|---|---|---|
alertmanager |
|
TODO | N/A |
gitaly |
|
|
|
gitlab-monitor |
|
TODO | N/A |
gitlab-workhorse |
|
|
|
nginx |
|
|
|
postgres-exporter |
|
TODO | N/A |
postgresql | ⓘ (Gets a reload) | ⓘ (Shows a warning to users to restart) | N/A |
prometheus | ⓘ (Gets a reload) | TODO | N/A |
redis |
|
TODO | N/A |
redis-exporter | TODO | N/A | |
registry |
|
|
|
sidekiq / sidekiq-cluster | TODO | TODO | ⓘ (both unicorn and sidekiq-cluster are issued hups on upgrade) |
unicorn | ⓘ not working due to issue | ⓘ not working due to issue | ⓘ (HUP on upgrade) |
gitlab-pages |
|
ⓘ not working due to issue |
|
- N/A: means the service is not part of normal gitlab.com deployments
cc @twk3
Edited by John Jarvis