Agreed.
Unless a test is specifically testing the functionality of
VirtualMachine.list()
it should not be used for locating a pid, which is easily provided
from the ProcessBuilder after the process is started.
Reducing the environmental interference of scanning all running java
processes
is our best bet at this point for these intermittent timeouts.
I do plan to follow up to find the root cause of these hangs,
but at this time we don't have a concrete problem identified.
We just know which swamp to look for the beast.
On 3/24/19, 9:09 PM, David Holmes wrote:
Just an aside, but it seems that other tests that happen to use this
functionality can also run into these transient timeout issues and
likely for the same underlying reasons. See for example:
JDK-8217347: [TESTBUG] runtime/appcds/jvmti/InstrumentationTest.java
timed out
So it would be good to get to the bottom of the underlying problem here.
Cheers,
David
On 23/03/2019 1:00 pm, Chris Plummer wrote:
Ok. No need to write the stress test now, just file the CR.
thanks,
Chris
On 3/22/19 4:07 PM, gary.ad...@oracle.com wrote:
Yes, there will be more follow up after the patch is pushed.
I started a thread on jtreg-dev to discuss the current implementation
and possibly an enhancement to meet the need of the vmTestbase
tests that may have the wrong expectation of the current feature.
e.g. the previous test harness had a nonconcurrent option.
Filing a bug against jtreg will be cleaner after a bit more
discussion there.
You had also suggested creating a stress test to see if we can
capture the jps hang. That may take a bit more time to consider
and is not as urgent as some other work items in progress.
On 3/22/19 3:32 PM, Chris Plummer wrote:
Hi Gary,
The changes look fine, but are you going to file the CRs I
suggested (one for jtreg exclusiveAccess.dirs help text and one to
further look into this issue)?
thanks,
Chris
On 3/22/19 1:38 AM, gary.ad...@oracle.com wrote:
This proposed change will configure the sun/tools/j* and
com/sun/tools/attach tests to be labeled as jtreg
exclusiveAccess.dirs
so the java attaching tests are not run concurrently.
There is only annecdotal observations that multiple
tools are running when these tests are timing out.
1000s of test runs have succeeded with this change in place.
More investigation can be done to find the root cause
of these failures, but for now it's best to make the tests
run more reliably without timing out.
Issues:
https://bugs.openjdk.java.net/browse/JDK-8220242
https://bugs.openjdk.java.net/browse/JDK-8220295
Webrev:
http://cr.openjdk.java.net/~gadams/8220295/webrev/
