Drain on restart: finish the job in hand, then restart issue #50
closed: completed andymccall opened this · 0 comments
andymccall
A runner restart kills its job, so runner deploys wait for both runners to be idle, which on a busy fleet is rarely. On SIGTERM (systemctl restart/stop), stop claiming, finish the job in hand (its result reported), then exit; systemd's stop timeout covers the longest job. Deploys then wait only for that runner's own job, and never kill one.
A runner restart kills its job, so runner deploys wait for both runners to be idle, which on a busy fleet is rarely. On SIGTERM (systemctl restart/stop), stop claiming, finish the job in hand (its result reported), then exit; systemd's stop timeout covers the longest job. Deploys then wait only for that runner's own job, and never kill one.