Single threaded in production

[singpolyma]

We are currently used to using thin via rackup in production for single-threaded but evented (and thus concurrent) request handling with eventmachine.

We are curious about leaving eventmachine for ruby3. The documentation says that using falcon with rackup is not recommended, but is this just because it will be single-threaded? Since our code will all be async, that seems like not such a big deal and probably we can expect performance similar to thin when used in this way?

[ioquatix]

How many processor cores do you have on the instance and do you want to use them all?

[singpolyma]

I'm aware that a single ruby process won't use multiple cores, but our webapps are generally I/O bound not CPU bound so that's never been a concern.

[ioquatix]

The rackup command has a number of limitations.

• Limited control over server multi-process/multi-thread configuration.
• Limited control over protocol selection.
• Limited control over binding address (and frankly pretty strange options for handling it).
• Limited fault tolerate (e.g. app crash / restart process).
• Limited support for rolling restart.

If you've tested your use case and none of these issues bother you, it might be an acceptable solution to use rackup.