dccm running out of file descriptors

On Sat, Jan 10, 2004 at 04:22:18PM -0700, Vernon Schryver wrote:
> > From: Gary Mills <mills@cc.UManitoba.CA>
> 
> > This is under Solaris 8.  `dccm' is started with `ulimit -n 4096',
> > and runs with `-j 800'.  I wonder if the context limit calculation
> > in `dccm' needs a bit of updating.
> 
> That is a plausible idea, except that I've looked at the source and
> don't see any files not accounted for.
> 
> Does `lsof` on your system show sockets?  If so, is there anything
> strange what `lsof` says?  With 'j-800' there should be no more
> than the following for each active job:
> 
>     a socket connected to sendmail,
>     socket to talk to the DCC server.
>     3 FDs for log files,
> 
> for a total of 4000 FDs.  There should 2 or 3 dozen other
> miscellaneous FDs including those for shared libraries.

I just took a look.  It's quiet now.  `dccm' was using 109 threads.
Yes, `lsof' shows sockets.  Here are the file descriptors by type:

 783 IPv4
  33 VREG
   4 VCHR
   1 DOOR

The sockets are 730 UDP and 53 TCP.  Of the UDP sockets, 705 look
like this:

  COMMAND  PID   USER   FD   TYPE        DEVICE SIZE/OFF    NODE NAME
  dccm    5100 daemon 1708u  IPv4 0x3000e2575c8      0t0     UDP electra.cc.umanitoba.ca:64103 (Wait_Data_Xfr)

25 look like this:

  dccm    5100 daemon 1929u  IPv4 0x3002f5a3588      0t0     UDP *:39400 (Idle)

> Is there any chance that an overall system limit on open files is being
> reached?  What is the dccm error message?

No, I don't think it's a system limit.  There are zillions of messages
like this:

  Jan  8 00:15:03 electra dccm[20771]: [ID 925838 mail.error] dcc_mkstemp(/usr/local/dcc/whiteclnt): Too many open files

And a few like this:

  Jan  8 00:43:02 electra dccm[20771]: [ID 125918 mail.error] DCC: accept() returned invalid socket (Too many open files), try again


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