Reloading after deploy, "high" availability

[aleho]

Lately I've been thinking about a few ways to reload code / clear the OPCache after deploys.

As FrankenPHP and Caddy are tightly coupled, reloading FrankenPHP will drop all active connections to clients.

Looking at Deployer, they used to recommend setting opcache.validate_timestamps = 1, but have recently implemented cachetool to clear the OPCache in FPM through a socket after a deploy. Of course this is not available for FrankenPHP.

What are your ideas to achieve similar "zero-downtime" deploys for FrankenPHP in worker mode?

I could still add a "reload-frankenphp" script to my containers and live with the short (or sometimes longer) downtime, but is there a better way? Maybe related to #556 (reply in thread)?

[AlliBalliBaba]

The [file watcher] (#1013) is being worked on, but it will still take some time to complete.
I think our interests here align a bit since I'm also trying to make reloading workers more graceful and am looking into clearing opcache along the way.

For a deployment it probably would make more sense to expose the same reload behavior some other way (maybe via caddy's admin?)

[aleho]

I actually really don't mind any "manual" work on deploy, I wouldn't even need the file watcher (though it'd be absolutely great if that was a feature of FrankenPHP)!

So, if there's no inotify for whatever reason – I don't think it would have an impact on my projects, but YMMV – I'd be completely fine with calling a FrankenPHP admin endpoint inside the container to "purge" OPCache via opcache_reset().