The cause of this is not compile.invokedynamics=false. I was not running tomcat correctly. But there is definitely a performance problem when running under tomcat. I am investigating further.

I downloaded the latest version of Wildfly (13.0.0 Final).
I copied my war file to standalone/deployments and started wildfly.

The speed problems are clearer. It appears to take upto 20 seconds to deliver the HTTP response from the server to the browser!

One example of a simple show page:
It took 10ms for 2 SQL Selects, 340ms to generate the views, but then 20 seconds before the page appears on the browser.

I believe this is the same behaviour as in tomcat. Does anyone know why a warbler generated WAR file should display such behaviour? Is puma (the default web server) getting in the way of tomcat or wildfly?

I solved my problem.
IT IS NOT A JRUBY PROBLEM.
I downloaded the latest version of tomcat, 8.5.32, threw my war file into the webapps directory, and started it. No configuration changes!
My application is as quick as when it runs in development.

I still have no idea why both tomcat 7.0.53 and wildfly 13.0.0 Final take so long to deliver the HTML output back to the browser. I should note that generation of the views often took more than a second, which seems rather long for my simple views (just a table with 50 rows and 4 columns).

I'm going to close this ticket now

As I have been able to reproduce this problem with a small sample rails app, I have reopened this issue.

Code and instructions for building the rails app are here: https://github.com/ninkibah/tc2

After starting tomcat If you refresh the page from 5-20 times, you will notice that at least one of the
refresh times is taking a very long time (more than 10 seconds) to come up.
I have noticed that rails itself says it has produced the view (normally in under 500ms).
My guess is that the problem occurs in the jruby/rack code.

Running as a standalone puma app shows no performance problems.

Awesome, thank you for the reproduction. I suspect something about deployment in a container is messing with how we generate invokedynamic calls.

I appreciate how much you and Tom and the rest of the team do, so I feel obliged to make your jobs as easy as possible. However, to make my stripped down app, I failed to create a new rails app. For some reason on deployment, I kept getting errors from bundler not being able to find the rake gem. I suspect, I will have to create a ticket in jruby/warbler, as I completely failed to see any difference between my stripped down app, and a new rails app.

@ninkibah Ok I'm going to re-close this. If you can figure out a good way for us to reproduce it, we can help. It would also help to get your (JVM flag) -XX:+PrintGCDetails output around the time it slows down, and perhaps output from (JRuby flag) -Xjit.logging or (JVM property) jruby.jit.logging=true.