pidfile handling

[rsluiters]

Sidekiq currently uses two different methods to generate the name for pidfiles: If only one Sidekiq process is running the file is only called sidekiq.pid, if two or more processes are running the pid files are called sidekiq-0.pid, sidekiq-1.pid, ... This leads to problems in deployment, as changing from 1 to 2 processes or back to 1 always leads to not stopped sidekiq workers.

So when e.g. changing the sidekiq_processes from 1 to 2 not the process stored in the sidekiq.pid is stopped but the non-existing processes sidekiq-0/1.pid. So the old process keeps running.

[seuros]

• stop sidekiq
• change settings
• start startkiq.

[rsluiters]

This doesn't really work in continuous deployment (especially over different stages), so you would manually need to stop sidekiq and then deploy. As the number of processes is in the repo(deploy.rb) checking in a changed process count would not adapt sidekiq properly in the automatic deployment.

[seuros]

Ok, send a PR. let have always sidekiq.pid, sidekiq-1.pid

[rsluiters]

Ok, cool, this would solve the problem on scaling up. It just came into my mind that scaling down would likely still create problems, or not? If you have 4 processes and then go back to 3, would it stop all processes or only the 3 because of the sidekiq_processes set to 3 ?

[seuros]

That the responsibility of the deployer, maybe adding note in the Readme will be enough.

[edslocomb]

capistrano-sidekiq is currently overriding the pidfile setting in sidekiq.yml, and that is not acceptable.

Can we please revert this change and send it back to w1mvy to fix it so that it will only behave in this (terrible) way if a config option is set?

[codener]

I opened a new issue on this: #156.