[BUG] ENV['APP_ENV'] isn't passed along to the server

[sdubinsky]

Whatever I set ENV['APP_ENV'] to, when I run it with rackup directly, the server always starts in development mode. I would expect it to respect the env var, as it does when running it directly. This is on Ubuntu with ruby 3.0.1 and

    rack (3.0.0.beta1)
    rackup (0.2.2)
      rack (>= 3.0.0.beta1)
      webrick
    webrick (1.7.0)

My config.ru is

run do |env|
  [200, {}, ["Hello World"]]
end

What I mean is, when I run APP_ENV=production rackup config.ru I get:

Rack::Handler is deprecated and replaced by Rackup::Handler
Puma starting in single mode...
* Puma version: 5.6.5 (ruby 3.0.1-p64) ("Birdie's Version")
*  Min threads: 0
*  Max threads: 5
*  Environment: development
*          PID: 561633
* Listening on http://127.0.0.1:9292
* Listening on http://[::1]:9292
Use Ctrl-C to stop

but when I run APP_ENV=production puma I get:

Puma starting in single mode...
* Puma version: 5.6.5 (ruby 3.0.1-p64) ("Birdie's Version")
*  Min threads: 0
*  Max threads: 5
*  Environment: production
*          PID: 561750
* Listening on http://0.0.0.0:9292
Use Ctrl-C to stop

with the environment set to production as expected.

Honestly, I'm not sure whether the bug is in rack, rackup, or puma. If there's a better location for this, let me know. Thanks!

[ioquatix]

Sorry, I thought this issue was created on the rack project, not rackup. This is the correct place to discuss this.

Uh, so I don't have a strong opinion about it, but I've never liked RACK_ENV and felt it was very confusing for users (it only supports development or deployment and doesn't really understand production).

Personally, I'd like to see us remove the handling of environments from rackup - I suspect what's happening here is that rackup is defaulting to RACK_ENV | development and forcing puma to consume that.

I think for Rack 3.0 release, we should make a 1.0 release of this which replicates the existing Rackup behaviour, but then I'd be totally fine with us cutting loose on a Rackup 2.0 release with a bunch of improvements and a rethink of what we want to provide (option handling, etc).

[sdubinsky]

I suspect what's happening here is that rackup is defaulting to RACK_ENV | development and forcing puma to consume that.

This does seem to be what's happening:

deus-ex:~/code/puma_mwe$ RACK_ENV=production bundle exec rackup config.ru
Rack::Handler is deprecated and replaced by Rackup::Handler
Puma starting in single mode...
* Puma version: 5.6.5 (ruby 3.0.1-p64) ("Birdie's Version")
*  Min threads: 0
*  Max threads: 5
*  Environment: production
*          PID: 619341
* Listening on http://0.0.0.0:9292
Use Ctrl-C to stop

I think the current behavior is a bug, and I don't think you should preserve it in a 1.0 release. Puma expects the env to be something like production or development, not deployment.

[ioquatix]

That seems reasonable to me, do you want to make a PR?

[sdubinsky]

Sure.

Looking at the code, there doesn't seem to be one obvious place to do it. As far as I can tell, rackup is just passing all of its options to the handler, and counting on the handler to ignore whatever it doesn't need. I can see two possibilities:

1. Rename environment to rack_environment in the rackup options. This has the advantage of being clearer with intent.
2. add environment to the list of ignored options. This has the advantage of being a much smaller code change.

What do you think is best?

[ioquatix]

I think we should aim to remove RACK_ENV completely, so I think option (2) makes more sense.

[sdubinsky]

I took a reasonable stab at it, and it's turning into a bit of a game of whack-a-mole trying to sort through everything. I have a workaround for my problem by running puma instead of rackup, and I think when you get around to removing RACK_ENV that will fix this anyway, so I'm OK with closing this.