RFR 8015391: NPG: With -XX:+UseCompressedKlassPointers OOME due to exhausted metadat, a space could occur when metaspace is almost empty

[Jon Masamitsu]

On 6/24/13 11:23 AM, Coleen Phillimore wrote:
>
> On 06/24/2013 01:46 PM, Jon Masamitsu wrote:
>> http://cr.openjdk.java.net/~coleenp/8015391/src/share/vm/memory/metaspace.cpp.frames.html
>>
>>> 2673   size_t specialized_count = 0, small_count = 0, medium_count = 0, large_count = 0;
>>> 2675   size_t cls_specialized_count = 0, cls_small_count = 0, cls_medium_count = 0, cls_large_count = 0;
>>
>> Not introduced by your change but
>>
>> humongous_count instead of large_count
>> and
>> cls_humongous_count instead of cls_large_count
>>
>> would be more consistent.
>
> Yes, it is.  I changed it.
>
>>
>> http://cr.openjdk.java.net/~coleenp/8015391/src/share/vm/classfile/classLoaderData.cpp.frames.html
>>
>> I'd suggest adding Metaspace::AppClassMetaspaceType
>>>   385     } else if (SystemDictionary::is_app_class_loader(class_loader())) {
>>>   386       if (TraceClassLoaderData && Verbose && class_loader() != NULL) {
>>>   387         tty->print_cr("sun/misc/Launcher$AppClassLoader class loader");
>>>   388       }
>>>   389       set_metaspace(new Metaspace(_metaspace_lock, Metaspace::BootMetaspaceType));
>> and adding a flag InitialAppClassLoaderMetaspaceSize.  It's not 
>> obvious to me
>> that we should always be reserving  the same amount of space for the 
>> AppClassLoader
>> as the BootClassLoader.   The same default is OK with me.
>
> Hmm.  Ok.  I don't like adding new flags, so can I add an 
> AppMetaspaceType and give it the same initial size as 
> BootMetaspaceSize until proven that it doesn't need to be the same size.

With the BootClassLoader the space for the initial Metaspace is committed
when the first system class is loaded.  The same would be true for any
Java application (all have some type of AppLoader, right)?   We know we're
going to use the space committed for the BootClassLoader Metaspace. Are
we going to get yelled at for committing 4m and not using it.
>
> I suppose adding an AppMetaspaceInitialSize flag will give me more to 
> talk about at the Permgen Elimination JavaOne talk though (but I'd 
> still rather not).

When you have 100000000000000000 flags, who's going to
notice the 100000000000000001-st :-)

Jon
>
> Coleen
>
>>
>> Rest looks good.
>>
>> Jon
>>
>>
>> On 6/20/13 9:26 AM, Coleen Phillimore wrote:
>>> Summary: Allocate medium chunks for class metaspace when class 
>>> loader has lots of classes
>>>
>>> I originally made class metaspace keep small chunks and not start 
>>> allocating from medium chunks, because I thought with only Klass 
>>> objects, small chunks is enough.  This test has a large set of class 
>>> loaders who have a lot of classes, so allocated thousands of small 
>>> chunks each.  Class loaders with that many classes should start 
>>> allocating from medium chunks, for class metaspace.
>>>
>>> I also increased the ClassMediumChunk size to 4k and measured a lot 
>>> less fragmentation waste than 1K or 2K for this test. Lastly, the 
>>> AppClassLoader instance should have as large of an initial metaspace 
>>> as the bootclass loader.
>>>
>>> Tested with vm.quick.testlist and the failing test.
>>>
>>> open webrev at http://cr.openjdk.java.net/~coleenp/8015391/
>>> bug link at http://bugs.sun.com/view_bug.do?bug_id=8015391
>>> local bug link https://jbs.oracle.com/bugs/browse/JDK-8015391
>>>
>>> Thanks,
>>> Coleen
>>>
>>
>