RFR [XS] : 8229182: runtime/containers/docker/TestMemoryAwareness.java test fails on SLES12

[Bob Vandette]

Instead of adding this: opts.appendTestJavaOptions = false;


Try adding these lines:

String javaHeapSize = new String(sizeToAllocInMb/2) + “m”;

opts.addJavaOpts("-Xmx" + javaHeapSize);

This will set the max Java Heap to 1/2 of the size that we will attempt to allocate.
This should reliably cause an OOM exception.

Bob.


> On Aug 16, 2019, at 4:14 AM, Baesken, Matthias <matthias.baesken at sap.com> wrote:
> 
>> Can we not set -Xmx explicitly after any test options have been
>> appended so that it takes precedence? 
> 
> Hi Bob,   do you have a good estimate for an Xmx to set ?
> 
>> It seems fragile to disable
>> appinding all test options altogether. Clearly, this test is
>> susceptible to passed -Xmx options.
> 
> We  have here a java - execution  taking place in the docker container, not the test machine itself .
> I am not sure , are the "regular"   test options so good for this special execution ? Some maybe are , some aren’t ...
> 
> E.g.  we ended up  with
> 
> -Xmx768m -Djava.awt.headless=true
> 
>  and more strange settings  in the  "docker run"   call because of this   ...
> 
> Best regards, Matthias
> 
> 
> 
>> Hi Matthias,
>> 
>> On Fri, 2019-08-16 at 06:10 +0000, Baesken, Matthias wrote:
>>>> Unfortunately  we do not see the expected
>> "java.lang.OutOfMemoryError"
>>>> in the nightly test runs .
>>>> 
>>>> When I execute  the  jtreg test locally  on the same (!)  ppc64le  machine ,
>>>> under the same user  but not from  the test framework we use nightly
>>>> the  "java.lang.OutOfMemoryError"      shows up.
>>> 
>>> Hello, after looking a bit more into it,   we observed   that  the  nightly  runs
>> set a higher -Xmx  = 768m  value  that gets inherited  into the
>> DockerRunOptions
>>> And made the  central  test  fail .
>>> So I had to remove this  :
>>> 
>>> 105         // make sure we avoid Xmx settings from the jtreg vmoptions
>>> 106         opts.appendTestJavaOptions = false;
>> 
>> Can we not set -Xmx explicitly after any test options have been
>> appended so that it takes precedence? It seems fragile to disable
>> appinding all test options altogether. Clearly, this test is
>> susceptible to passed -Xmx options.
>> 
>> Failing that, could we just omit the -Xmx option?
>> 
>> Thanks,
>> Severin
>> 
>>> New webrev :
>>> 
>>> http://cr.openjdk.java.net/~mbaesken/webrevs/8229182.2/
>>> 
>>> Best regards, Matthias
>>> 
>>> 
>>>> Hi Christoph,
>>>> 
>>>>> Hi Matthias,
>>>>> 
>>>>> looks good to me in general.
>>>>> 
>>>>> However, I'm not convinced that line 114
>>>>> ".shouldContain("java.lang.OutOfMemoryError");" won't provoke a
>> test
>>>>> error in case the child process was killed by the container, e.g. exited
>> with
>>>>> exit code 137. We should at least run this patch for several days in our
>>>> system
>>>>> to see if it catches all variations of the problem.
>>>>> 
>>>> 
>>>> Unfortunately  we do not see the expected
>> "java.lang.OutOfMemoryError"
>>>> in the nightly test runs .
>>>> 
>>>> When I execute  the  jtreg test locally  on the same (!)  ppc64le  machine ,
>>>> under the same user  but not from  the test framework we use nightly
>>>> the  "java.lang.OutOfMemoryError"      shows up.
>>>> 
>>>> Best regards, Matthias
>>>> 
>>>> 
>