Guide on making a symfony app compatible with frankenphps worker mode

[aszenz]

Just wondering if anyone has gone through this migration for a largish app, and what changes they had to make, how to test and overall guidance.

I know about symfony's reset interface and we are using it to reset stateful services since we use symfony messenger too. So I'm assuming we need to apply that everywhere and rest should work but would like to confirm from experts here first.

Also is there any tooling or monitoring that will help in this migration?

[henderkes]

You don't need to apply it everywhere, only in stateful services, which are a somewhat bad idea in general. The demo application and most other applications should be compatible with worker mode without any changes.

I migrated a fairly large app (~1000 Controllers + Services) and had to only add ResetInterface to a single one that cached data during the request.

[aszenz]

Thing is our app is multi-tenant with each tenant having it's own db + doctrine connection, so we need to ensure doctrine's entity manager (and services that rely on it) are reset too.

While our own services are usually stateless same can't be said about our deps

[aszenz]

Also being compatible is different from running in production for years without any issue, since this stuff is new I wanted to confirm peoples experiences with worker mode.

[henderkes]

Thing is our app is multi-tenant with each tenant having it's own db + doctrine connection, so we need to ensure doctrine's entity manager (and services that rely on it) are reset too.

Same here, although we use different worker threads for different tenants. I'd wager you would notice very quickly if Doctrine kept stateful behaviour, though.

Also being compatible is different from running in production for years without any issue, since this stuff is new I wanted to confirm peoples experiences with worker mode.

Been running in worker mode for about a year with few issues.

[aszenz]

although we use different worker threads for different tenants

Is this for perf reasons (preserving caches + connections) or for better isolation?

[henderkes]

It was different tenants running completely separate systems that later got merged into one VM running from the same code. No particular design decision, just how it happened to come together.