Mark Thomas wrote:
I am trying to bring together all the information I have gleaned on this
so far from the multiple threads to try and find the common factors.
So far I have:
- 7.0.21 is OK
- 7.0.22 has an fd leak
- 7.0.23 has an fd leak and may leak faster than 7.0.22
- occurs with APR/native
- does not occur with BIO
- has been observed in HTTP & HTTPS
- use of Comet does not trigger it
- use of compression does not trigger it
- separate connection and keep-alive timeouts does not trigger it
It may be related to POST processing.
I have tried (and so far failed) to reproduce this. I'll be looking at
POST processing next. In the meantime, here are some further questions
to try and narrow things down:
1. Does the application where this is observed make use of Servlet 3.0
async requests?
2. Does this leak occur when the NIO connector is used?
3. Are there any exceptions in the logs that weren't present in 7.0.21
or earlier?
4. Does the leak occur if sendfile is disabled?
I also have reviewing the 7.0.21 to 7.0.22 changes on my todo list but
there are quite a few as I was refactoring the connectors to reduce code
duplication and ironically, reduce maintenance requirements, at the time.
Suggestion: the "large POST requests" mentioned by a couple of posters suggest file
uploads, which may imply temporary files used to buffer the files being uploaded, no ?
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org
