Spawned procs should inherit goreman's stdin

[corasaurus-hex]

Processes spawned via foreman inherit stdin from it. You can test this like so:

❯ cat Procfile
web: echo "Hello, $(read user; echo $user)"

❯ foreman start
23:16:12 web.1  | started with pid 91492
cora
23:16:14 web.1  | Hello, cora
23:16:15 web.1  | exited with code 0
23:16:15 system | sending SIGTERM to all processes

Why does this matter? Because some things care if stdin is a TTY or /dev/null (the previous stdin value for all launched procs):

❯ cora-goreman start watch
23:19:17 watch | Starting watch on port 5000
23:19:17 watch | > dev
23:19:17 watch | > npx tailwindcss -i ./app/assets/stylesheets/engine/tailwind-input.css -o ./app/assets/builds/tailwind.css --watch
23:19:18 watch | Rebuilding...
23:19:18 watch | Done in 167ms.
^C23:19:31 watch | Terminating watch

❯ /opt/homebrew/bin/goreman start watch
23:20:35 watch | Starting watch on port 5000
23:20:35 watch | > dev
23:20:35 watch | > npx tailwindcss -i ./app/assets/stylesheets/engine/tailwind-input.css -o ./app/assets/builds/tailwind.css --watch
23:20:35 watch | Terminating watch

You could always redirect /dev/null into your processes if that's what you want but forcing them to inherit your terminal's TTY is impossible if goreman is inserting /dev/null as stdin and so I think this is the better default.

Thoughts?

[mattn]

web1: echo "Hello, $(read user; echo $user)"
web2: echo "Hello, $(read user; echo $user)"
web3: echo "Hello, $(read user; echo $user)"

What happen for this? It is not guaranteed which one will be called first?

[corasaurus-hex]

Yep, there are no guarantees, the point is not to guarantee which tool gets to read from stdin the point is to 1. match how foreman works, 2. provide for tools that expect to be run via a tty.

[corasaurus-hex]

My example was to demonstrate that the processes foreman starts inherit foreman's stdin, not to enable some kind of specific ordering of stdin behavior.

[mattn]

Is it a specification that foreman handles standard input? If so, please show me the specification or the source code that intentionally implements it. I can't implement it without specification.

[corasaurus-hex]

Here's the code: https://github.com/ddollar/foreman/blob/1445adc429d79b5b3dd1d54cf9dcb75d253c924f/lib/foreman/process.rb#L54

Foreman uses Process.spawn and redirects both stdout and stderr to the same destination. If you don't set a value for :in it will inherit the stdin of the current process that is calling Process.spawn.

Since the docs for Process.spawn are so crap and what it's doing is using posix spawn under the hood with file descriptor inheritance here's a demonstration that you can try that proves that not specifying an stdin for a spawned process means the process inherits the spawning process' stdin:

❯ cat test_stdin_inherited.rb
#!/usr/bin/env ruby

require 'time'

current_process_pid = $$

pid = Process.spawn("bash greet_user.sh", :out => $stdout, :err => $stdout)

puts "#{$0} - #{Time.now.iso8601} - #{current_process_pid} - Waiting on process #{pid} to exit!"

Process.waitpid(pid)

puts "#{$0} - #{Time.now.iso8601} - #{current_process_pid} - Process #{pid} exited!"

❯ cat greet_user.sh
#!/bin/sh

pid=$$

read -p "$0 - $(date -Iseconds) - $pid - What is your name? " user

echo "$0 - $(date -Iseconds) - $pid - Hello, $user"

❯ ruby test_stdin_inherited.rb
test_stdin_inherited.rb - 2023-03-30T11:48:05-05:00 - 37677 - Waiting on process 37774 to exit!
greet_user.sh - 2023-03-30T11:48:05-05:00 - 37774 - What is your name? Cora
greet_user.sh - 2023-03-30T11:48:08-05:00 - 37774 - Hello, Cora
test_stdin_inherited.rb - 2023-03-30T11:48:08-05:00 - 37677 - Process 37774 exited!

Tada! Proof that foreman spawns processes such that they all inherit stdin from the main foreman process. I've also demonstrated that this matters to some processes.

[joeblew99]

is this going forward ? stdin is really useful feature. I guess you would all agree.

[mattn]

Thank you, and sorry for the late response. I tested this and the piped case works as advertised:

$ cat Procfile
web: read user && echo "Hello, $user"
$ echo cora | goreman start
10:27:10 web | Hello, cora

However, when goreman runs on an interactive terminal it hangs: subprocesses are placed in their own process group (Setpgid: true), so a proc reading from the inherited TTY stdin receives SIGTTIN and stops, and goreman waits forever. Reproducible with:

$ timeout 5 script -qec "goreman start" /dev/null < /dev/null
# web stops on SIGTTIN, goreman hangs until killed

So this needs some handling for the TTY case before it can be merged — e.g. only inherit stdin when it is not a terminal, or gate it behind a flag.