RFR (S): 8004924: NPG: jmap -heap output should contain ClassMetaspaceSize value

Mikael Gerdin mikael.gerdin at oracle.com
Tue Feb 26 15:19:19 UTC 2013


Erik,

On 2013-02-21 12:41, Erik Helin wrote:
> Hi,
>
> based on internal feedback, I've updated the test. Since the test
> spawned the "jmap -heap" process and connected the jmap process to
> itself, there was a race condition: if the "jmap -heap" process produced
> a lot of output, then the JVM process would not be able to consume this
> output because it has been stopped by jmap. At the same time, jmap would
> wait for the JVM to consume the output from jmap.
>
> The test now pipes the output from "jmap -heap" to a fike and then later
> analyzes the file.

Overall this approach seems much safer than having a java thread copy 
the contents. However in this case you do not drain the stderr from the 
process you created so in some cases, for example if the SA is broken 
then it will throw up a long exception stacktrace on stderr.

Please either use ProcessBuilder.redirectErrorStream to pipe all output 
to one file (and rename it to remove stdout from its name)
or use redirectError and pipe that output to a separate file.

Thanks
/Mikael

>
> Webrev:
> http://cr.openjdk.java.net/~ehelin/8004924/webrev.01/
>
> Thanks,
> Erik
>
> On 02/20/2013 09:50 PM, Erik Helin wrote:
>> Hi all,
>>
>> this change the flag ClassMetapaceSize to the output of jmap -heap. I
>> have also added a test that checks the ouput (the test uses the new
>> process builder framework).
>>
>> Webrev:
>> http://cr.openjdk.java.net/~ehelin/8004924/webrev.00/
>>
>> Bug:
>> http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=8004924
>>
>> Testing:
>> JPRT
>>
>> Thanks,
>> Erik
>



More information about the hotspot-gc-dev mailing list