https://bz.apache.org/bugzilla/show_bug.cgi?id=60750

--- Comment #14 from Ty <ty733...@gmail.com> ---
> OOMEs can happen for a number of reasons -- not always complete 
> heap-exhaustion.

Very true, I forgot about the "initialize an object array of
size=Integer.MAX_VALUE" operations that OOME before they even begin.

> I'd be interested to see if JMX reports an increase in the OOME count when 
> these 
> zero-length chunks are dropped from the responses.

Once the heap size was increased, I see a very sharp and short-lived (i.e.
garbage-collectable) increase in OldGen utilization.  From memory, a spike of
about 850MB.  I think that's what you're asking about.

Besides the probably-swallowed OOME, that just leaves the mystery of a response
that (when under memory constraints) is missing the last chunk but otherwise is
complete.  I'd wager a third-party library is to blame -- in our case and in
OP's case, there is a JSON library in the mix-- but what is strange to me is
the response isn't compliant with the RFC spec for Transfer-Encoding: chunked,
at least according to my reading of it:  a response can't be considered
complete without the zero byte chunk, but the server seems to think it is.

-- 
You are receiving this mail because:
You are the assignee for the bug.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org

Reply via email to