graceful scale down

[rashtao]

When scaling down the children are killed and not gracefully shutdown.

Eg. In my app code I have:

    process.on('message', function (msg) {
        if (msg == 'shutdown') {
            // log something to file
        }
    });

I launch it with pm2:

$ pm2 start app.js -i 4
[PM2] Process app.js launched
┌──────────┬────┬─────────┬───────┬────────┬─────────┬────────┬─────────────┬──────────┐
│ App name │ id │ mode    │ pid   │ status │ restart │ uptime │ memory      │ watching │
├──────────┼────┼─────────┼───────┼────────┼─────────┼────────┼─────────────┼──────────┤
│ app      │ 93 │ cluster │ 20165 │ online │ 0       │ 110s   │ 25.973 MB   │ disabled │
│ app      │ 94 │ cluster │ 20167 │ online │ 0       │ 110s   │ 25.969 MB   │ disabled │
│ app      │ 95 │ cluster │ 20207 │ online │ 0       │ 105s   │ 25.938 MB   │ disabled │
│ app      │ 96 │ cluster │ 20212 │ online │ 0       │ 105s   │ 25.938 MB   │ disabled │
└──────────┴────┴─────────┴───────┴────────┴─────────┴────────┴─────────────┴──────────┘

and I give the command:

$ pm2 scale app 3

one child is terminated, but the message event is not fired. So I guess that the child is brutally killed.

The message events are correctly fired when I do:

$ pm2 gracefulReload app
[PM2] Reloading process by name app
[PM2] Process app succesfully reloaded
[PM2] Process app succesfully reloaded
[PM2] All processes reloaded

What am I missing?
Is it possible to "gracefullly" scale down?

[Fantus]

i do have the same issue, as i need to close some connection before the worker process gets killed. any suggestions?

[jtokoph]

I know it's been a while since this was closed, but nobody gave any alternative for gracefulStop.

After a bit of digging I found a solution and wanted to post it here, as my Google searches brought me here first.

gracefulReload uses process.on('message') to catch the reload notification. The rest of the stops and restarts send a SIGINT as documented here: http://pm2.keymetrics.io/docs/usage/signals-clean-restart/

Simple solution is to call your same clean shutdown code in a SIGINT handler as documented at the same link. If you want your process to be given more than the default 1.6 seconds to cleanly shutdown, you'll need to set the PM2_KILL_TIMEOUT environment variable to the number of milliseconds you want.

[Unitech]

SIGINT is the way to go:

SIGINT handle now works as expected (it will be called every time the process get exited).

To try it:

$ npm install pm2@next -g
$ pm2 update

[Unitech]

Patch available on pm2@2.1.4 (main):

$ npm install pm2 -g
$ pm2 update