TCP/IP BIND problem

[dschwarz]

Hi,

from time to time, not (yet) reproducable at will, tcpip bind server logs messages like this:

Tue 16 15:19:23 ERROR: zone /IN: transfer: could not set file modification time of '': file not found
. (file exists, that's sure)
.
Tue 16 15:20:22 ERROR: _jnl: open: I/O error
.
.
Tue 16 15:23:38 ERROR: BUILD13$:[TCPIP_V54_BLECO7.SRC.BIND9_SERVER]SOCKET.C;1:1877: unexpected error:
Tue 16 15:23:38 ERROR: internal_accept: accept() failed: too many open files

The last message is repeated five times per second, the bind server does not accept zone information anymore. STOP/START TCPIP$BIND helps until the next time.

Name resolution still works, so it usually takes some time until we see the problem.

We are running a cluster, two DS20, OpenVMS7.3-2, TCPIP 5.4 ECO 7. TCPIP$BIND is running on one of the nodes.

Any idea ?

[MarkOfAus]

Have you tried rndc or ndc (I forget which one) to see if it will give you any more information?

[Steven Schweda]

> [...] too many open files

I know nothing, but might you have a problem
with too few I/O channels (or something like
that)?

mcr sysgen show CHANNELCNT

If you ask it, does AUTOGEN make any
suggestions?

[Volker Halle]

CHANNELCNT does not have the AUTOGEN feedback attribute, but this could well be a 'channel leak'. Check FILLM quota for the BIND server process and also check the no. of channel from time to time:

$ ANAL/SYS
SDA> SET PROC
SDA> READ SYSDEF
SDA> FROMAT JIB ! look for JIB$L_FILLM and JIB$L_FILCNT

SDA> SHOW PROC/CHAN
...

Compare 'Total number of open channels :' against CHANNELCNT.

JIB$L_FILCNT is the 'remaining' File Limit quota.

Volker.

[dschwarz]

Mark,
no, I never tried rndc. I will try it next time we see the problem.

Steven,
channelcnt is 4096, fillm of BIND is 4096, too

Volker,
JIB$L_FILCNT = %X0FFB = 4091
JIB$L_FILLM = %X1000 = 4096

SDA> SHO PROC/CHAN/ID=
says:
Total number of open channels : 411

But BIND is running quietly at the moment.
I'll check that again in times of problems.