We're still seeing this problem.  I did a quick diff between 2.4.5 and 
2.5c,  and only had a question about the new check on line 134 inside 
get_line function (in the new code):

if(buf[i] == '\0') {
    n_read = i;
    continue;
}

What is it suppose to achieve?  Why put "continue", as this is the final 
statement of the for loop (for(i = n_read; i < bufsize; i++) {)?  Also, 
the value of the n_read is only important if "i<bufsize", which is the 
next if statement right after the loop.  I was thinking this was written 
to handle the NULL character on a line, but nothing really happens, 
except changing the value of n_read.

I can't see how this would cause the problem with pound using up 100% of 
CPU.  I pretty certain one of the threads is stuck in a loop, and thats 
why was looking at this piece specifically.


Dave Steinberg wrote:[...][...][...]

FYI, I commented these lines out in http.c, and haven't had the problem 
for the last 3 days (normally the problem would occur within few hours 
of restarting pound). 

Albert wrote:[...][...]
>>> This is to announce the release of Pound v2.5c. This is an
experimental
>>> version - the third in the 2.5 series (and incidentally also the
first
>>> release candidate for 2.5). Changes since version 2.5b:
>>>
>>> Enhancements:
>>>   - added support for HTTPS backends (with or without a client
>>>     certificate. No back-end checking is done.
>>>
>>> Bug fixes:
>>>   - fixed problem with sub-patterns in session parameters. This
allows
>>>     the usage of parentheses in ID patterns.
>>>
>>> The software is at version 2.5c (beta quality). Further testing
>>> (especially under heavy loads), improvements and suggestions are
>>> welcome.[...][...]