RFR 8030628: MemoryMXBean/CollectionUsageThreshold.java is not applicable for VM SQE testing

[Mandy Chung]

On 3/12/2014 2:19 AM, Mattias Tobiasson wrote:
> Hi Mandy,
> I have talked to Leonid Mesnik, and he says there are plans for a jtreg flag that would solve this problem in a general way (maybe @requires). I do not know the time plan for that jtreg feature. The test currently fails in nightlies, so we need a temporary fix.

There are other tests (other than java.lang.management tests) that also 
hardcodes with -XX:+UseSerialGC and other flags.  They should fail in 
nightlies.

> I don't know when/if all different GC configurations are used in nightly tests. But for these tests that depend on the GC, I think it's good to always run with different GC configurations. The test is fast so all 4 runs take about a second.

I'll defer this to Staffan and the serviceability team.

Your approach is okay but add some unnecessary complexity in my opinion. 
   If Leniod is going to resolve this with the jtreg support soon, it 
might be simpler to just have the test to run with the GC configuration 
the test run specifies as stop-gap.

thanks
Mandy

> I did not notice that there are more tests in the same dir. I will change them too.
>
> Mattias
>
> ----- Original Message -----
> From: mandy.chung at oracle.com
> To: mattias.tobiasson at oracle.com, serviceability-dev at openjdk.java.net, jonathan.gibbons at oracle.com
> Sent: Tuesday, March 11, 2014 8:48:04 PM GMT +01:00 Amsterdam / Berlin / Bern / Rome / Stockholm / Vienna
> Subject: Re: RFR 8030628: MemoryMXBean/CollectionUsageThreshold.java is not applicable for VM SQE testing
>
> Hi Mattias,
>
> Have you discussed with Jon Gibbons about the jtreg @requires support?
> I thought it was partly motivated by the requirement to specify a test
> to run in different collector.
>
> The reason why these regression tests explicitly specifies different GC
> flags was to increase the coverage and ensure to catch regression early
> during development cycle.  At that time, the VM nightly testing rotates
> the test execution with different GC configuration that delayed to catch
> a regression that occurs in one collector while the nightly testing is
> testing another collector.  For integration, I don't recall if different
> collectors are tested rather than default.  It may be time to revisit
> the test execution with different collectors.  If the verification of
> different collectors is well covered in nightly, perhaps these tests no
> longer need one @run per each collector.
>
> There are other regression tests that specifies the -XX:+UseXXGC flag in
> the @run tag.  It makes sense to modify them in the same patch (perhaps
> at least tests under test/java/lang/management).
>
> Mandy
>
> On 3/11/14 6:15 AM, Mattias Tobiasson wrote:
>> Hi,
>> Could you please review this test fix.
>>
>> The test fails because it specifies its own GC command line options, for example:
>> @run main/othervm/timeout=300 -XX:+PrintGCDetails -XX:+UseSerialGC CollectionUsageThreshold
>>
>> When the framework also specifies GC version, then the test will fail with this log:
>> "Conflicting collector combinations in option list; Error: Could not create the Java Virtual Machine."
>>
>> The solution is to run the test in a separate JVM with controlled GC options.
>> The test will be run multiple times.
>> First with the command line specified from the framework, without test specific GC options.
>> Then once for each GC version specified in the test. When the test specifies the GC, any GC options from the framework are removed.
>>
>> webrev: http://cr.openjdk.java.net/~ykantser/8030628/webrev.01
>> bug: https://bugs.openjdk.java.net/browse/JDK-8030628
>>
>> Mattias
>>