Disable Context Propagation and useless IDLE connections Netty detection

[franz1981]

@cescoffier as well

Any other feedback?

[LesnyRumcajs]

@franz1981 Could you please briefly elaborate on those flags? I would like to know if they do not take advantage of the logic simplicity and are generic enough - in other words, would you use the same flags for a regular backend application? If so, is there a reason they are not the default?

[cescoffier]

I would also test with the new gRPC server (using the http server)

[franz1981]

you use the same flags for a regular backend application? If so, is there a reason they are not the default?

3 different sets of changes here:

• disable context propagation ( https://quarkus.io/guides/context-propagation) : this is an advanced feature and right now we don't disable it by default although most users shouldn't require it, because we are not validating it at compile time (and I have no idea if is possible). This is the comment from one of the implementor -> Save setting the same state in the Vertx local data duplicated context quarkusio/quarkus#34132 (comment)
• disable mutiny callback decorators: such are still involved into the context propagation dance. I'm disabling them just in case the former setting isn't enough
• remove the Idle connection handler of Netty: if peer clients/server aren't shield by proxies and TCP FIN work as expected, there's no point for this

So, in short: assuming what the other servers are doing (they do have something to detect half-dead connections/inactive from long time?) i don't see any reason why pay for both features.

Said that, I didn't yet profiled the test nor inspected the load generator itself yet :)

[franz1981]

And I've noticed another thing: https://quarkus.io/guides/grpc-service-implementation#scaling
The default number of quarkus grpc servers seems to be just 1 (vertx verticle) - meaning that what you have observed so far was probably using a single core?

[LesnyRumcajs]

And I've noticed another thing: https://quarkus.io/guides/grpc-service-implementation#scaling The default number of quarkus grpc servers seems to be just 1 (vertx verticle) - meaning that what you have observed so far was probably using a single core?

I do believe it was addressed in #361, right?

[LesnyRumcajs]

Thanks for the write-up; I really appreciate it!

[franz1981]

I do believe it was addressed in #361, right?

Nope. Quarkus engine is using Vertx under the hood (which uses Netty), but th pr you have mentioned is on Vertx alone, not Quarkus, which have its own configuration parameters and slightly different default behaviours.

I have double checked that the runners for quarkus are instead configuring the number of instances hence my previous comment is invalid!

[LesnyRumcajs]

Sounds good, thanks!

[He-Pin]

@LesnyRumcajs Would you rerun the bench on the same box some later time :) ?

[LesnyRumcajs]

@LesnyRumcajs Would you rerun the bench on the same box some later time :) ?

I'll re-run it this or the next weekend.

[franz1981]

any news @LesnyRumcajs ?