mike@BRL.MIL (Mike Muuss) (03/06/91)
Our our SGI machines at BRL, we have /etc/syslog.conf set up to use "@host" forwarding to a central collection machine for certain kinds of error messages. Unfortunately, we have discovered that under IRIX 3.3.1, syslogd rarely succeeds in forwarding these messages after the system is rebooted. If syslogd is killed and restarted, then network forwarding of log messages works properly. I believe that this difficulty is a consequence of the order that the various daemons are started from within /etc/rc2.d, namely: S20sysetup: syslogd S30network: routed, portmap, named, inetd. I suspect that syslogd is unable to resolve the host name given, because named is not running yet, so it silently discards the @host forwarding. Phil Dykstra and Bob Reschly report that for BSD UNIX systems, a different ordering is necessary: # NOTE: The ordering of routing, name service, system logging, RPC port # mapping, and finally inetd service is important. All need get off the # ground as soon as possible. System logging needs name service which # is itself dependant on routing. In addition, RPC port mapping must be # running prior to inetd As a consequence of these observations, I am considering that, at a minimum, syslogd startup be delayed until after S30network. Perhaps syslogd could be moved into it's own file, such as S40syslog, or some such. In a few days, I'll probably go ahead and give breaking out a S40syslog a try. Please let me know if there is a better way of dealing with this issue. Thanks, -Mike
srp@babar.mmwb.ucsf.edu (Scott R. Presnell) (03/06/91)
mike@BRL.MIL (Mike Muuss) writes: >Our our SGI machines at BRL, we have /etc/syslog.conf set up to use >"@host" forwarding to a central collection machine for certain kinds of >error messages. Unfortunately, we have discovered that under IRIX >3.3.1, syslogd rarely succeeds in forwarding these messages after the >system is rebooted. If syslogd is killed and restarted, then network >forwarding of log messages works properly. >I believe that this difficulty is a consequence of the order that >the various daemons are started from within /etc/rc2.d, namely: >S20sysetup: syslogd >S30network: routed, portmap, named, inetd. >I suspect that syslogd is unable to resolve the host name given, because >named is not running yet, so it silently discards the @host forwarding. I not positive that this will fix it, but theoretically it should work here if name resolution is the problem. In any case, it's alternative that you might try. (1) We run the the bind name daemon (named) here, and from the *.bak files that are kept for the local zones, I have the machine construct a new /etc/hosts file every night - so this hosts file contains the local domain hosts (plus whatever hard wired stuff you want like "localhost 127.0.0.1"). (2) WRT host reslotion, I have the following set up. === /usr/etc/resolv.conf === domain mmwb.ucsf.EDU hostresorder local bind nameserver 0.0.0.0 === So that the /etc/hosts file will be queried first - and local references should be resolved by that lookup. For us, it helps with two conditions: (1) Mounts and exports at boot time. By the time our machines reach the phase of the boot where mounts and exports are being performed, named hasn't completely loaded. Just makes the boot a little cleaner, and less dependent on DNS. (2) When the named caches have gone stale and/or athoritative host for our zone(s) is unreachable, this allows local resolutions to work. I think it might also solve the above syslogd apparent resolution problem. Hope this helps. - Scott Presnell -- Scott Presnell +1 (415) 476-9890 Pharm. Chem., S-926 Internet: srp@cgl.ucsf.edu University of California UUCP: ...ucbvax!ucsfcgl!srp San Francisco, CA. 94143-0446 Bitnet: srp@ucsfcgl.bitnet