Access to JVM PID in universal package

[kodemaniak]

I am using the universal packager to build a zip for a backend service. Upon deployment I want to stop the old process before starting the new one. Unfortunately the script does not exec the java process, so the PID I can observe points to the script and does not terminate the JVM. Is there any workaround for that or can the script modified to exec the JVM?

[jsuereth]

Huh, look at #99. Can you explain what we broke with that fix?

[kodemaniak]

Not that I am a guru on this, but AFAIK the exec results in the original process (the script) being replaced with the called process (the JVM), meaning that the PID in $! actually is the PID of the JVM. I write that PID to a file and use it to kill the service later, when I want to start a new version. Currently I get the PID of the start script, and when I kill the script, the JVM continues to run.

Maybe there other ways to achieve the same. It could also be an option to modify the script generation for java_server packages such that start/stop can be provided to the script and the script handles the PID.

[muuki88]

@kardapoltsev proposal looked like this:

${{exec_cmd}} "$java_cmd" 

And in bash script replacements smth like this:

replacements ++= Seq("exec_cmd" -> (if(fork) "exec" else ""))

However we only removed the exec in #99 . I didn't experienced any issues with this yet.

@kodemaniak , can you post a minimal example for this?

[muuki88]

I made some tests with following results.

Execution Command	Startscript	Result
./bin/mukis-test	"$@"	PIDs equal, kill process works
./bin/mukis-test	exec "$@"	PIDs equal, kill process doesn't work
./bin/mukis-test &	"$@"	PIDs differ
./bin/mukis-test &	exec "$@"	PIDs equal, kill process works

• play 2.2.0
• current sbt-native-packager

[kardapoltsev]

I've done the same tests but got different results. And from man page of exec I realize that it should be added back. But I need to test init.d script before.

If exec is specified with command, it shall replace the shell with command without creating a new process.

@muuki88 Did you use pid file created by Play application in your tests?

[kardapoltsev]

I've tested systemV start scripts with exec $java_cmd and now it works fine.

[muuki88]

Yes, play always creates its own PID. So I checked if the PID play outputs and the one thats running are the same. I didnt test it with a bootsystem, just on commandline.

[kodemaniak]

I assume, you don't need an example anymore? I guess what I am doing/experiencing is the same that @kardapoltsev tested.

[muuki88]

I will test it wil Upstart and SystemV again (just to make sure, that a commandline call is different than a call from the bootsystem).

Is an option, like @kardapoltsev proposed in #99, neccessary or just adding it should be fine?

[kodemaniak]

I don't need the option, but if exec is problematic in some scenarios making it configurable is probably the best solution.

[kardapoltsev]

I was totally wrong when asking for removing exec. Now I've tested system V start script, bash script and read man exec. I think we should put in back.

[muuki88]

I think we're all in some kind of learning process here :P
We'll read it and I think we should document this and provide an alternative way (IMHO overriding the start-script-template) to do otherwise.