[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: >1024 connections in slapd / select->poll



At 05:18 AM 11/14/2004, Volker Lendecke wrote:
>In http://www.openldap.org/faq/index.cgi?file=1126 you mention a workaround for
>the problem that slapd does not allow more than about 1000 concurrent client
>connections. Recompiling glibc is not an option for me,

Note that recompiling glibc and/or kernel is not needed if
you are using modern versions of glibc and kernel.  We've
been running with OPENLDAP_FD_SETSIZE set as high as 8192
without problem on Linux without problem (and without
touching anything other than OpenLDAP Software).

>so I rewrote the main loop in daemon.c to use poll instead of select.
>Thus slapd can ignore the
>FD_SETSIZE #define and rely on sysconf( _SC_OPEN_MAX ); for the maximum number
>of file descriptors.
>
>Is this of any interest for you?

Yes, moving to poll() (and alternatives) is a TODO list item.
We're been moving slow because of a few unknowns, namely
interaction with threads and performance.  Positive operational
experience using poll() here would help move things forward.

>And, is it appropriate to send patches to this mailing list? It's about 11kbytes.

For discussion purposes, yes.  But I note that URLs use is
generally more friendly to the bandwidth challenged.

To actually submit a patch for consideration, as indicated
in the contributing guidelines
<http://www.openldap.org/devel/contributing.html>, it is
necessary to submit it using the Issue Tracking System
<http://www.openldap.org/its/>.  If you have not read
these guidelines, please do.

Thanks, Kurt