Hi Stefan, thanks for the response. My responses are in-line.
> Is the parameter -f TLS1 necessary to reproduce the problem?
>
No, same behavior. Long-running children just never free up the memory
as long as I am hitting the SSL port, seemingly regardless if I pass an
algorithm to AB or not. I see radically different (as in this time,
normal) behavior if I hit the non-SSL port.
> Is the URL / of your webserver a php page? If yes, why is the content
> length of the page 0, what does the php script do? If not php, what is
> it? A simple redirect?
>
It is the index.php for the CodeIgniter framework, the redirect is
somewhat contrived, as it's handled by some internal CodeIgniter stuff
that I am not very familiar with. I don't think PHP is involved at all
in the leak, since the non-HTTPS port works fine.
> A workaround may be to add MaxRequestsPerChild 1000 (or an even lower
> value) to your configuration.
>
>
That wouldn't really help since long-running children just hold onto
memory no matter what it seems.
I feel like this issue would be huge and brought up by someone else by
now if it's an actual regression. Does the package information I put in
the beginning look right?
--
Joe McDonagh
Operations Engineer
www.colonfail.com
Hi Stefan, thanks for the response. My responses are in-line.
> Is the parameter -f TLS1 necessary to reproduce the problem?
>
No, same behavior. Long-running children just never free up the memory
as long as I am hitting the SSL port, seemingly regardless if I pass an
algorithm to AB or not. I see radically different (as in this time,
normal) behavior if I hit the non-SSL port.
> Is the URL / of your webserver a php page? If yes, why is the content
> length of the page 0, what does the php script do? If not php, what is
> it? A simple redirect?
>
It is the index.php for the CodeIgniter framework, the redirect is
somewhat contrived, as it's handled by some internal CodeIgniter stuff
that I am not very familiar with. I don't think PHP is involved at all
in the leak, since the non-HTTPS port works fine.
> A workaround may be to add MaxRequestsPerChild 1000 (or an even lower
> value) to your configuration.
>
>
That wouldn't really help since long-running children just hold onto
memory no matter what it seems.
I feel like this issue would be huge and brought up by someone else by
now if it's an actual regression. Does the package information I put in
the beginning look right?
--
Joe McDonagh
Operations Engineer
www.colonfail.com