JDK-8166236 : Startup regression with Jython
  • Type: Bug
  • Component: performance
  • Affected Version: 9
  • Priority: P3
  • Status: Closed
  • Resolution: Not an Issue
  • Submitted: 2016-09-17
  • Updated: 2017-04-28
  • Resolved: 2016-09-22
Related Reports
Relates :  
JDK-8163880 - Constant pool caching of fields inhibited/delayed unnecessarily
Description
Startup of Jython with JDK 9 (and server VM) appears to have regressed since JDK 8. Consider the following bench.py and execution with JDK 8 and JDK 9 (jdk-9+136)

import java.lang
c = java.lang.Class.forName("java.lang.Boolean")
print c
(exit)

$ time -p java -jar jython-standalone-2.7.0.jar bench.py 
<type 'java.lang.Boolean'>
real 2.26
user 6.38
sys 0.26

$ time -p java --add-exports=java.base/sun.nio.ch=ALL-UNNAMED -jar jython-standalone-2.7.0.jar bench.py 
<type 'java.lang.Boolean'>
real 2.36
user 7.53
sys 0.29

User time on this system has increased. Part of the issue seems to be the switch to G1. Another part may be increased usages of lambdas in the regex code.
Comments
Claes Redestad has tracked down the main issue. It seems that Jython creates a cache on first usage where it stores an index of the packages in each JAR file in the JDK image. It seems to bring that cache by scanning rt.jar. This means startup with JDK 8 is initially slow but is significant faster on subsequent usages. With JDK 9 then rt.jar no longer exists so Jython doesn't have a cache. If Jython is updated to use jrtfs and integrates this with this cache then the startup regression should disappear.
28-04-2017
The problem is in Jython that it is not caching the jimage packages.
22-09-2016