dccm failure under load

Gary Mills mills@cc.UManitoba.CA
Sun Jan 5 03:54:40 UTC 2003

On Sat, Jan 04, 2003 at 04:34:31PM -0700, Vernon Schryver wrote:
> Since 1.1.13 that your DCC servers are running, I found and fixed an error
> in the count of the total open files used when per-user log files are
> used.  Installing 1.1.17 or using `dccm -j` to reduce the number of
> simultaneous dccm jobs might stop the "too many open files" problems.

Yes, I knew about that, but since I'm not using per-user log files,
I disregarded it.  `dccm -j' might be useful.  I've just compiled 1.1.17.
I'll upgrade to it when I get a chance.

> The rate of new connections does not matter.   What matters is the
> total number of active connections at any one time.

I notice just now that `dccm' has 153 threads.  It was higher a few
minutes ago, so it must be managing them.  For comparison, our Trend
virus scanner milter has 62 threads.  It too was higher a few minutes
ago, about double.  When I look at `dccm' with `lsof', it has used
598 file descriptors.  Some are TCP connections to sendmail.  Some
are for log files.  However the vast majority are for UDP sockets.
They all look like this:

dccm    11638 daemon  596u  IPv4 0x300029c3328      0t0    UDP electra.cc.umanitoba.ca:50646 (Wait_Data_Xfr)
dccm    11638 daemon  597u  IPv4 0x300057702d8      0t0    UDP electra.cc.umanitoba.ca:50650 (Wait_Data_Xfr)

What are all these used for?  Are they for queries to the DCC server?
Why are there so many of them?  According to `netstat -an', they are
connected to port 6277 on the DCC server host.  When I check there, no
UDP connections are visible, and `dcc' is only using 17 file
descriptors.  I realize that UDP `connections' are not connections in
the same sense as TCP connections.

-Gary Mills-    -Unix Support-    -U of M Academic Computing and Networking-

More information about the DCC mailing list

Contact vjs@rhyolite.com by mail or use the form.