[1010] in Moira
bug-sipb,linux-dev
daemon@ATHENA.MIT.EDU (Greg Hudson)
Sun Sep 15 20:02:24 1996
Date: Sun, 15 Sep 1996 20:02:16 -0400
From: Greg Hudson <ghudson@MIT.EDU>
To: moiradev@MIT.EDU, moira-admin@MIT.EDU
I sent a request to hesreq about this, but they're probably the wrong
people to deal with it.
Right now, the architecture for attaching Athena lockers depends on
the ordering of DNS records, which is not required by the
specification. Operating systems using newer versions of BIND (all
recent distributions of Linux, as well as the forthcoming NetBSD 1.2
release) often get Hesiod text entries in the wrong order, causing the
wrong locker to be attached. The biggest problem is with the sipb
locker, where the sipb cell version is preferred to the athena cell
version when it is available.
To fix this problem the right way, I've written a patch for "attach"
to interpret preference fields at the end of filsys entries. It's in
the sipb-athena source tree, and I plan to incorporate this patch into
the mainline Athena sources at some point as well. Lower preferences
have priority.
I would like Moira modified so that hesiod entries have preference
fields at the end. For instance, "hesinfo sipb filsys" should return.
AFS /afs/sipb.mit.edu/project/sipb n /mit/sipb 1
AFS /afs/athena.mit.edu/contrib/sipb n /mit/sipb 2
This way, people with the new attach will always get the sipb cell
version if it can be read. Older versions of attach will simply
ignore the preference fields; to test this, all of the SIPB office
machines have been running with the above entries for
sipb.filsys.ns.athena.mit.edu in their named.locals since September 4
with no problems. (We have machines running Solaris, Irix, Ultrix,
AIX, Linux, NetBSD, and BSD 4.3 on both Vaxes and RTs.) (Yes, these
machines are all running the old attach binary, except maybe one of
our Linux machines.)
I've been told by Dan Winship that this change is a simple hack to
hesiod.gen.
