On Mon, 1 Feb 2021 10:38:30 GMT, Thomas Stuefe <[email protected]> wrote:

> Hi,
> 
> may I please have reviews for this trivial fix.
> 
> We see timeouts with this test on slow large (memory wise) AIX machines, as 
> well as large core files.
> 
> This test is a negative test which tests that the VM corectly recognizes an 
> error condition and aborts. Code goes through jni_FatalError()->os::abort(), 
> writes a core and aborts the process.
> 
> No parameters are given for that VM invocation, so heap size depends on 
> machine size. On my Linux box, the generated core takes about 500m. On AIX we 
> see cores of 16G and more.
> 
> The difference between AIX and other platforms is that AIX does not have the 
> notion of "committing" memory (we have no MAP_NORESERVE flag), so all mmapped 
> memory counts toward the commit charge from the moment it is mapped. That 
> makes core files on AIX annoyingly large. I'd expect a similar behavior on 
> other platforms with overcommit disabled.
> 
> The fix is to run the test without CreateCoredumpOnCrash. For good measure, I 
> also reduced the heap size to 128m.
> 
> Thanks to @ArnoZeller for figuring that one out.
> 
> 
> -----
> Tests: manual jtreg tests, verifying that no cores are written.

Hi Thomas,
The fix looks good. Thank you fixing it.
I've submitted mach5 test job with 40 runs of the 
java/lang/instrument/PremainClass tests.
All are passed while they were failing before.
Also, David is right, the option -XX:-CreateCoreDumpOnCrash can be removed from 
tests:
 test/jdk/java/lang/instrument/PremainClass/NoPremainAgent.java
 test/jdk/java/lang/instrument/PremainClass/ZeroArgPremainAgent.java
 
 But I can file a separate bug on it.
 Also, I've closed bug https://bugs.openjdk.java.net/browse/JDK-8260469 as a 
dup of this one.
  
Thanks,
Serguei

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

Marked as reviewed by sspitsyn (Reviewer).

PR: https://git.openjdk.java.net/jdk/pull/2332

Reply via email to