[2052] in Release_7.7_team
release-team minutes 1/12
daemon@ATHENA.MIT.EDU (Dan Winship)
Wed Jan 12 15:08:58 2000
Date: Wed, 12 Jan 2000 15:08:43 -0500 (EST)
Message-Id: <200001122008.PAA12848@antharia.mit.edu>
From: Dan Winship <danw@MIT.EDU>
To: release-team@mit.edu
Attending: tb, wdc, aurora, ajfox, danw, ghudson | jweiss, rbasch
1. Soft NFS mounts
We don't want "attach" from peoples' .environment to hang forever.
Replace "soft" with "bg,intr"? Attach might not deal with "bg"
right now.
We need to investigate attach's behavior, and possibly add stuff,
but getting rid of "soft" is doable.
2. Linux rc.conf / athena-base
rc.conf doesn't mesh well with the layered model
options
a. keep things as they are
(This will be confusing, because some of the variables won't
work, if you don't have the right packages installed.)
b. split rc.conf into multiple files
(Linux only or everywhere?)
c. one rc.conf, packages add their own variables to it
d. split into multiple files, but on workstations, have syncconf
generate the multiple files from rc.conf
We like "c" best.
What about syncconf? Most of the variables it does are not
athena-package specific. "MAILRELAY" could be dealt with by
sendmail rc script.
Deleting a package will not delete its rc.conf lines. The base
rc.conf should probably mention that fact.
(Unresolved discussion about whether rc.conf would be easier or
harder to deal with if it had lots of comments.)
Now we can get rid of athena-base:
- base rc.conf and mandesc will go into generic build stuff so
they're created by any package that needs them
- /etc/athena/version will go to athena-ws: this will require
some hacking of other packages to make them not necessarily
depend on it. But this is good, because some things that use it
(olc/sendbug) shouldn't be using it in the layered world.
3. Status reports
Solaris install: good (see miki's mail)
Desktop: bad (see forthcoming athena-desktop minutes)
PWS: We've gotten responses. We'll probably meet soon.
Linux install: Where's Karl? (Or at least, where's Karl's work?)
4. Random stuff
wdc: Bob has gotten an SGI bug fixed. We're making progress on
fixing the SGI/GNOME X crash. I'm continuing to bug Sun about the
Access Plus issue. I'm also making progress on getting us into a
hardware beta testing program.
Who will take release-team minutes when Dan leave? Greg says he
will if no one else wants to.
5. Linux update procedure
Linux machines won't have /srvd and /srvd/.rvdinfo to use to know
when to take an update.
They need to check for new RPMs, but where?
We'll use cluster info to point to the RPM location, and the
release will create and maintain a cluster.fallback file for
machines that don't have cluster info. (This means putting out a
full release will requiring putting out a patch release of the
previous version to update cluster.fallback.)
We don't need to have the mechanism in place to do a full update
when we deploy the pilot cluster, as long as we can put it into
place with a patch release. (Or, worst case, Cluster has to
reinstall 10 machines.)
Integrity checking: ideally we want to check files against MD5s in
AFS, but for the pilot we'll probably just check against MD5s on
local disk (and reinstall packages that are wrong).
Will also want to check (on Public workstations) that no additional
packages are installed.