RFR: 8287135: Calculation of jmm_GetMemoryUsage is wrong

[Thomas Stuefe]

On Fri, 27 May 2022 07:22:11 GMT, Thomas Stuefe <stuefe at openjdk.org> wrote:

>>> I think the right fix is to just convert the MetaspacePool into NonClassMetaspacePool and only report the non-class values.
>> 
>> Yes, it's okay for me. But I have another concern.
>> 
>> The compressed class pool is not directly used by other VM components. However, memory pools are exposed via java management APIs. At JDK level, users could obtain memory pools and choose a specified pool by name. If adapting the first solution, I guess it's somewhat strange for users to understand what is `Non Class Space`:
>> 
>>   private static MemoryPoolMXBean getMemoryPool(String name) {
>>         List<MemoryPoolMXBean> pools = ManagementFactory.getMemoryPoolMXBeans();
>>         for (MemoryPoolMXBean pool : pools) {
>>             if (pool.getName().equals(name)) {
>>                 return pool;
>>             }
>>         }
>> 
>>         throw new RuntimeException("Expected to find a memory pool with name " + name);
>>     }
>>   public static void foo() {
>>      MemoryPoolMXBean a = getMemoryPool("Compressed Class Space");
>>      MemoryPoolMXBean a = getMemoryPool("Non Class Space");
>>   }
>> 
>> If we remove `CompressedClassSpace`, we can only `getMemoryPool("Metaspace")`. Although metaspace is not baked in the specification, IMHO it's easier for developers to understand what is `metaspace` compared to the concepts of `Non Class Space` and `Compressed Class Space`.
>
>> 
>> If we remove `CompressedClassSpace`, we can only `getMemoryPool("Metaspace")`. Although metaspace is not baked in the specification, IMHO it's easier for developers to understand what is `metaspace` compared to the concepts of `Non Class Space` and `Compressed Class Space`.
> 
> I personally think that would be totally fine.

> @tstuefe could you take a look at the test changes. Since we can no longer query the compressed class space individually, I think the tests may become more lenient than before. For example, memoryUsageSmallComp/TestDescription.java uses `MemoryUsageTest::checkForNotGrowing()` which monitors the used bytes in `MetaspaceBaseGC::pool.getUsage().getUsed()`
> 
> * Before this PR, it checks that the usage of the compressed klass space doesn't grow.
> * After this PR, it will allow the compresed klass space to grow, as long as the "other" meta space shrinks by a similar amount.
> 
> Is this OK? Or should we add a new whitebox API to query the compressed vs meta space?

Sorry for the long delay, I was not at work.

I think you may be right, we need a replacement for the old memory bean for these tests. Whitebox seems easiest.

-------------

PR: https://git.openjdk.org/jdk/pull/8831