Allow for an ADHoc Sending of a signal to a ContainerPilot job. #571
Description
Problem:
Send a HUP, TERM, or KILL signal to a job running in ContainerPilot without having to subscribe to the SIGHUP and SIGUSR2 handlers that CP is running.
I discovered the need for this functionality while dealing with applications that don't support PID Files or a simple way of discovering the pid.
In this particular case ( elastic/logstash#3577 (comment) ) the maintainers refuse to implement the functionality, and have decided this is the job of the init system.
Proposed Solution:
Add "-termjob 'jobname'", "-killjob 'jobname'", and "-reloadjob 'jobname'" to the ContainerPilot control plane.
Behavior:
-termjob, will send the TERM signal to the provided jobname.
-killjob, will send the KILL signal to the provided jobname.
-reloadjob will send the HUP signal to the provided jobname.
Useful example:
In the autopilot pattern we run into situations where we need to change a job's configuration, and then send the appropriate signal to the corresponding jobs.
consul-template \
-template "./logstash.yml.ctmpl:./logstash.yml:containerpilot -reloadjob 'logstash'
or if logstash can't reload our change with a HUP, restart it with a TERM.
consul-template \
-template "./logstash.yml.ctmpl:./logstash.yml:containerpilot -termjob 'logstash'
or if we really really want to be brave, we can KILL it quickly.
consul-template \
-template "./logstash.yml.ctmpl:./logstash.yml:containerpilot -killjob 'logstash'
kill and term should only apply to jobs where restarts are allowed.
Confession:
#513 Can accommodate most of this, but I can envision a circumstance where this pattern might be more convenient and easier for people to use, and reason about. I didn't see 513, until I wrote the first bit of code.
Possible negative impacts:
I can't really find any possible future issues with having this feature in the Control Plane if implemented correctly. The current Signal handler should co-exist without issue.