Op donderdag, 20 januari 2011 17:03 schreef Mark Thomas <ma...@apache.org>:
On 20/01/2011 15:45, Mark Thomas wrote:
> On 20/01/2011 15:29, Mark Thomas wrote:
>> On 20/01/2011 11:50, Ronald Klop wrote:
>>> Hi,
>>>
>>> In this mail thread there is mention of a memory leak which is fixed in
>>> Tomcat 7.0.6.
>>> http://marc.info/?l=tomcat-user&m=129487211630075&w=2
>>>
>>> Is it possible that this leak also exists in 6.0.30 (and not in 6.0.29)?
>>> We are seeing problems since 6.0.30 with the Nio connector. Just a quick
>>> question to rule this option out.
>>
>> Not a chance. I introduced the bug in 7.0.x with some Servlet 3
>> re-factoring. That re-factoring was not back-ported to 6.0.x.
>>
>> It is easy to confirm. With the leak, you would see one
>> Http11NioProcessor object on the heap for every HTTP request made. An
>> OOME soon followed.
>
> Given that there have been two reports of this, I may need to take that
> back. Let me do some tests.
/me swears loudly and eats words re the "Not a chance" comment.
The re-factoring wasn't back-ported but a few lines of code were [1] as
a follow up to a fix in a similar area. Those few lines introduced this bug.
This makes the NIO connector in 6.0.30 pretty much useless.
I'll get the process started to fix this on the dev list.
Mark
[1] http://svn.apache.org/viewvc?view=revision&revision=1033341
Thanks a lot for looking into this! I was just planning to put some time and
effort in making a test case when I saw your last mail. I will visit a Tomcat
conference some day and offer you a beer or two. :-)
Ronald.
