It has been reported that in certain environments, jtreg when launching a test there's a long delay before running the test code. This appears to be happening for every single test that gets launched in that test run.

Thread dumps have shown the following relevant part:

"DefaultTestRunner:Worker-0:0" #35 [34819] prio=4 os_prio=31 cpu=7.51ms elapsed=4.30s tid=0x000000012e99aa00 nid=34819 runnable [0x0000000171b26000]
   java.lang.Thread.State: RUNNABLE
at java.net.Inet6AddressImpl.getHostByAddr(java.base@23-internal/Native Method)
at java.net.InetAddress$PlatformResolver.lookupByAddress(java.base@23-internal/InetAddress.java:1235)
at java.net.InetAddress.getHostFromNameService(java.base@23-internal/InetAddress.java:850)
at java.net.InetAddress.getCanonicalHostName(java.base@23-internal/InetAddress.java:820)
at com.sun.javatest.regtest.exec.RegressionScript.run(RegressionScript.java:133)
at com.sun.javatest.Script.run(Script.java:431)
at com.sun.javatest.DefaultTestRunner.runTest(DefaultTestRunner.java:183)
at com.sun.javatest.DefaultTestRunner$1.run(DefaultTestRunner.java:76)

It appears that for every test, the jtreg framework code tries to determine the hostname and that can lead to a name resolution which on some setups, like the one where this was causing an issue, can take a long time. The code in question resides at https://github.com/openjdk/jtreg/blob/master/src/share/classes/com/sun/javatest/regtest/exec/RegressionScript.java#L132 where we do:

String hostname;
try {
   hostname = InetAddress.getLocalHost().getCanonicalHostName();
} catch (UnknownHostException e) {
   hostname = "127.0.0.1";
}
testResult.putProperty("hostname", hostname);

I'm not too familiar with the jtreg code, but the only place where I have seen this hostname is in the .jtr file that gets generated for the tests:

#-----testresult-----
description=file\:/jdk/open/test/jdk/java/lang/FooBarTest.java
elapsed=835 0\:00\:00.835
end=Fri Jun 07 20\:31\:30 IST 2024
environment=regtest
execStatus=Passed. Execution successful
harnessLoaderMode=Classpath Loader
harnessVariety=Full Bundle
hostname=foo-bar.example.com
javatestOS=Mac OS X 14.5 (aarch64)

Irrespective of where this gets used, I think one optimization that we can do in jtreg is to compute this hostname once for the entire jtreg launch instead of doing it every test that gets executed as part of that run. That should at least reduce the long delays that accumulate for each test execution.