[💡 FEATURE REQUEST]: after_init phase command

[rauanmayemir]

Plugin

Server

I have an idea!

It would be nice to have an after_init section similar to server.on_init that would run after the server is initialized and ready to receive requests.

Extra nice option: allowing to have the same section for each plugin (unless it's meaningless because server.after_init would run after all the plugins are done loading). E.g: I can have this command for http or grpc, so that I could start accepting requests without waiting for jobs plugin.

[rustatian]

Done, in testing now, will be in the v2023.2.0 ⚡

[rauanmayemir]

The implementation is a bit quirky as after_init is being called by the server and then called again by each roadrunner plugin.

The initial proposal was supposed to be like this:

1. Allow each plugin to have an after_init phase
2. Add a global after_init that would be called after all the plugins are ready

It is a single-use command that only fires once. Anything else could be done by polling healthz.

I was trying to adopt this feature like this: process manager pre-opens a notification file descriptor and then expects the RR process to write to that FD and then close it. I verified that it works, but only once. I then get two more log lines: signal: broken pipe.

The goal is being able to mark container ready without resorting to polling.

[rustatian]

@rauanmayemir
after_init triggered once per pool initialization (e.g., per-plugin), there are no additional calls. Each plugin does have an after_init configuration to override the server's after_init general command. It can be specified per-pool. This option is not documented, because after_init is a questionable option which has very tiny scope where to use it. It won't be updated until I get additional feedback from the community about use-cases for this feature.

[rauanmayemir]

If I wanted a notification from a plugin, I guess I would have set e.g jobs.after_init and this actually makes sense e.g for dynamically creating and launching job pipelines.

This results in a confusion. You specify server.after_init, but for some reason it's called by all the plugins. So ultimately, it's not server's after_init, it's just a section to specify the default command.

I was the one requesting the feature, but I don't know how I can use this or what for. 😄

[rustatian]

Thus, I would suggest you to describe your request in a new FR.
Describe use cases, proposed configuration options, etc.

[rauanmayemir]

To re-quote the original request:

It would be nice to have an after_init section similar to server.on_init that would run after the server is initialized and ready to receive requests.

It's basically a readiness notification that would allow process manager not to poll /health for readiness, because RR process won't be ready until it sends back the notification. Essentially, something like previously mentioned sdnotify, but for non-systemd use case.

I was confused thinking 'server' encompasses everything, not just a plugin called 'server'.