[2304] in Release_Engineering
Hesiod information for final release
daemon@ATHENA.MIT.EDU (Richard Basch)
Sat May 5 03:42:51 1990
Date: Sat, 5 May 90 03:42:35 -0400
To: dkk@MIT.EDU, release@MIT.EDU
Cc: rel-eng@MIT.EDU, ops@MIT.EDU
From: Richard Basch <probe@MIT.EDU>
Dave, sorry I forgot to get back to you on this...
I have not thought out the following closely, but I think this might be
the cleanest approach without having to make more packs:
Initial update:
- Change Hesiod for the syslib pack, as appropriate
- Swap the srvd's, as appropriate.
Following day, after the updates have occurred:
- Change Hesiod for the usrlib pack to be the AFS link.
From first inspection, this will not cause problems with any machine
types...
Case analysis:
Hesiod information has not propogated (only syslib changed):
6.4 urvd, 7.0 srvd
All links on workstations point through /srvd.
On servers, they point to /urvd/root... this means the servers will
still be running 6.4 binaries if they do the attach. This is the most
likely case to have a few incompatibilities, but none are foreseen. The
update will clear things up. There are no references from /srvd to the
workstation and then directly to the /urvd.
After everything has propogated:
7.0 urvd link, 7.0 srvd
The only risk is that someone has made a local /urvd, but again, /srvd
is the primary packs, and all /usr/* links indirect through /srvd, or
through /site/server/usr, and then they might go to /urvd/root (if the
workstation has not updated, so perhaps the change of the usrlib
information may wish to be delayed a bit, but not too long, as you are
now depending on two RVD's, which isn't bad, as long as you are not
relying on two servers -- twice as many failure points).
These are really the only cases:
7.0 srvd, 6.4 urvd
7.0 srvd, 7.0 urvd symlink
So, as you can see from the last description, staying with usrlib
pointing to 6.4 should not hurt, because there is enough backwards
compatibility. However, if this entails using two RVD servers to
provide pack service, this should only be used as a temporary solution.
Does anyone see any faults?
-Richard
