RFR: 8202142: jfr/event/io/TestInstrumentation is unstable

Erik Gahlin erik.gahlin at oracle.com
Mon Jan 20 22:34:19 UTC 2020


Hi,

Could I have review of a test fix. The test has been on the problem list 
for a long time, so there has been code rot.

The initial problem was the host string not matching the host field in 
the event. This happens when the tests (invoked by TestInstrumentation) 
are run in a certain order. I don't know why, but it seemed brittle, so 
I fixed this by removing the host validation and only check IP address 
and port.

The reimplementation of the socket API moved around code, so two methods 
can no longer be instrumented. Fix was to remove them from the list of 
methods to instrument.

Events may arrive out of order, possibly more likely due to changes in 
the JVM implementation, so events are now sorted chronologically.

Each test now closes/stops the recording it uses before it return from 
the main method. This reduces the risk that state in one test impacts a 
later.

Bug:
https://bugs.openjdk.java.net/browse/JDK-8202142

Webrev:
http://cr.openjdk.java.net/~egahlin/8202142

Testing:
jdk/jdk/jfr/event/io 30 times

Thanks
Erik



More information about the hotspot-jfr-dev mailing list