Is it possible to restart Unicorn without downtime?

patrickbajao · June 27, 2017, 12:37am

Hi,

I know that we can use sv stop unicorn then sv start unicorn to restart unicorn workers. But that will cause downtime. Killing the master with a USR2 signal is the usual way to do hot restart, as far as I know. Is it still safe to do that for Discourse? Or is there a better way to do this?

I understand that I rarely (or not at all) have to do this. But just want to know in case of emergency.

Thanks!

sam · June 27, 2017, 12:45am

Sort of, we do:

github.com/discourse/docker_manager

lib/docker_manager/upgrader.rb

main


      
          
          percent(80)
          web_server.reload
          reloaded = true
          
          # Flush nginx cache here - this is not critical, and the rake task may not exist yet - ignore failures here.
          percent(85)
          begin
            run("bundle exec rake assets:flush_sw")
          rescue RuntimeError
            log "WARNING: Unable to flush service worker file"
          end
          
          percent(90)
          log("Running post deploy migrations")
          run("bundle exec rake multisite:migrate")
          
          run("bundle exec rake s3:expire_missing_assets") if using_s3_assets
          
          log_version_upgrade
          percent(100)

So you would send the USR2 to the unicorn launcher that will forward the signal for you.

github.com/discourse/docker_manager

lib/docker_manager/upgrader.rb

main


      
          end
          
          def reset!
            @repos.each(&:stop_upgrading)
            clear_logs
            percent(0)
            status(nil)
          end
          
          def upgrade
            return if @repos.any? { |repo| !repo.start_upgrading }
          
            percent(0)
            status("running")
          
            clear_logs
          
            log("********************************************************")
            log("*** Please be patient, next steps might take a while ***")
            log("********************************************************")

patrickbajao · June 27, 2017, 1:07am

Ah. Sweet! That will do it. Thanks @sam!

tgxworld · June 27, 2017, 1:16am

USR2 - reexecute the running binary. A separate QUIT should be sent to the original process once the child is verified to be up and running.

@sam According to the docs, don’t we need to send a QUIT as well?

sam · June 27, 2017, 1:20am

Nope cause of:

github.com/discourse/discourse

config/unicorn_launcher

main


      
            log "exiting"
          }
          
          function on_reload_unicorn() {
            log "Reloading unicorn ($UNICORN_PID)"
            kill -s USR2 $UNICORN_PID
            unset NEW_UNICORN_PID
          
            count=0
            while [ "$count" -lt 180 -a -z "$NEW_UNICORN_PID" ]; do
              NEW_UNICORN_PID=$(pgrep -f -P $UNICORN_PID "unicorn master")

Topic		Replies	Views
Sv hup unicorn dropping traffic Product Support	2	149	September 1, 2026
Too many unicorn workers - failed to stop for upgrade? Self-hosting	4	968	November 3, 2020
Browser upgrade fails when discourse is configured to use one unicorn worker Bug	5	543	October 18, 2023
Upgrade abort - not enough Unicorn workers running Self-hosting	4	207	January 8, 2026
"Waiting for unicorn to reload", then update stops Product Support	30	5138	June 20, 2021

Is it possible to restart Unicorn without downtime?

AI-generated summary

Related topics