Hardcoded default port numbers prevent concurrent execution - consider testing if port is in use

[garikAsplund]

I had a couple projects open at the same time and noticed that the default ports are hardcoded and thus collide.

I'm using spin 2.4.2.

I'd be willing to give this a shot tomorrow. The issue is in crates/trigger-http/src/lib.rs.

I propose amending the serve and serve_tls functions to go through ~10 ports if the one listening is in use. This would not just work for default_value but also if the "--listen ADDRESS" is also in use.

[lann]

I don't think we should do this by default; if an app starts successfully but the port is hosting a different app than expected you could end up with frustration at best and a security issue at worst.

From personal experience I also just think this would be very surprising behavior; do you have any examples of other tools that work this way?

[garikAsplund]

No worries. I'm used to webdev and Vite has this feature.

It might not be that common of an issue and it's an easy enough quickfix by adding the --listen flag.

[lann]

Vite has this feature.

I've used Vite and never noticed this! 😅

As pointed out out-of-band, another option that sort-of works today is spin up --listen localhost:0, which picks a random unused port. I say "sort-of" because it doesn't currently log the actual selected port, it just says http://127.0.0.1:0 which isn't very helpful. That is definitely something we should try to fix.

[lann]

#2471

[garikAsplund]

Thanks! I'll take a look.

[itowlson]

Given #2471 and #2473, is it okay to close this as addressed via the port 0 workaround?

[garikAsplund]

I believe they're separate issues--it's a matter of preference for changing the logic to increment if a port, likely 3000, is in use. @lann @itowlson I'll leave that to you to decide.

[lann]

I'm somewhat opposed to this behavior being default - spin up is used in production in some places (for better or worse...) - but I wouldn't be opposed to an opt-in feature (flag, env var and/or runtime config option).

[itowlson]

Fixed via #3182