asciidoctor-diagram: Performance regression with JRuby 9
I don’t if we are doing something wrong here but AsciidoctorJ with Asciidoctor Diagram is noticeably slower with JRuby 9:
@SevereOverfl0w made a sample project to reproduce this issue. It’s a Clojure project but it’s relatively easy to run it (and it will generate a flamegraph!): https://github.com/SevereOverfl0w/asciidoctor-diagram-slow-repro
Once Clojure is installed you should be able to run clojure -A:run.
You should see somthing like:
Warming up JIT with 30 calls
Timing 100 calls of .readDocumentHeader
"Elapsed time: 1793.124807 msecs"
Producing flamegraph for 100 runs
Started [cpu] profiling
And the flamegraph will looks weird with spikes on Asciidoctor Diagram.
Now if you edit the deps.edn file and add org.jruby/jruby-complete {:mvn/version "1.7.27"} (just after net.sourceforge.plantuml/plantuml {:mvn/version "1.2018.13"}) then the output will be:
Warming up JIT with 30 calls
Timing 100 calls of .readDocumentHeader
"Elapsed time: 963.936468 msecs"
Producing flamegraph for 100 runs
Started [cpu] profiling
and the spikes on Asciidoctor Diagram are gone…
About this issue
- Original URL
- State: closed
- Created 5 years ago
- Comments: 33 (28 by maintainers)
Commits related to this issue
- #214 Move require statements out of extension registration block — committed to asciidoctor/asciidoctor-diagram by pepijnve 5 years ago
My next big project for JRuby is to fix up the require/load caching logic, which should prevent this from affecting anyone in the future. Glad to see you found a good solution that is not hacky and makes sense anyway!
@pepijnve The code is now faster in JRuby 9:
I’ll take a shot at profiling this with JProfiler later this evening. Perhaps that clarifies the difference in performance between jruby 1.7 and 9. I’ll let you know what I find (if anything).