RFR: 8373945: vmTestbase eatMemory/ClassUnloader provoke OOME to force GC and might cause GC in other threads [v5]

[SendaoYan]

On Mon, 22 Dec 2025 19:01:54 GMT, Chris Plummer <cjplummer at openjdk.org> wrote:

>> Another solution is make some sleep after call `unloadClass()`
>> 
>> 
>> diff --git a/test/hotspot/jtreg/vmTestbase/nsk/jvmti/CompiledMethodUnload/compmethunload001.java b/test/hotspot/jtreg/vmTestbase/nsk/jvmti/CompiledMethodUnload/compmethunload001.java
>> index 13058ec7864..15120d3cad6 100644
>> --- a/test/hotspot/jtreg/vmTestbase/nsk/jvmti/CompiledMethodUnload/compmethunload001.java
>> +++ b/test/hotspot/jtreg/vmTestbase/nsk/jvmti/CompiledMethodUnload/compmethunload001.java
>> @@ -47,7 +47,7 @@ public class compmethunload001 {
>>      private final static String CLS_TO_BE_UNLOADED =
>>          "nsk.jvmti.CompiledMethodUnload.compmethunload001u";
>>  
>> -    private final static int MAX_ITERATIONS = 50;
>> +    private final static int MAX_ITERATIONS = 5;
>>  
>>      static {
>>          try {
>> @@ -95,6 +95,7 @@ public static void callHotClass(String location) throws Exception {
>>  
>>          boolean clsUnloaded = clsUnLoader.unloadClass();
>>          clsUnLoader = null;
>> +        Thread.sleep(5000);
>>          System.gc();
>>      }
>
> It seems this is somewhat poorly written test and is also somewhat imprecise and presumpive. I'm guessing the unload is delayed a bit because it happens async, and you can't really guess how long it is going to take for the unload to happen. The loop and the System.gc() calls are just passing time until the unload happens. The System.gc() calls should not be needed since the ClassUnloader.unloadClass() call should have been sufficient. I would suggest removing both System.gc() calls and just have a loop that does short sleeps until the unload happens. Maybe limit the loop to no more than 10 seconds. A more elegant approach would be to wait on a monitor for no more than 10 seconds, and have the native code notify the monitor when the unload happens.

The code "wait on a monitor and notify from JNI" seems too complex for me. So I just limit the loop to no more than 10 seconds.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/28891#discussion_r2641853242