[1504] in Release_7.7_team
Larvnet deployment
daemon@ATHENA.MIT.EDU (Greg Hudson)
Wed Sep 23 14:20:09 1998
Date: Wed, 23 Sep 1998 14:20:03 -0400
From: Greg Hudson <ghudson@MIT.EDU>
To: ops@MIT.EDU
Cc: release-team@MIT.EDU
Okay, here is a concrete plan for Larvnet deployment during the 8.2
release cycle.
0. ops recently established that larvnet will be running on
dill, including during the test deployment phase.
1. Mike is going to make a larvnet.sloc entry pointing to
dill.
2. Some time during the next seven days or so, I expect to put
out 8.2.11 with Larvnet client support (that's liblarv,
busyd, and the xlogin changes) into the dev cell. The
biggest thing we want to know about is whether the liblarv
code breaks xlogin, since that would be especially poor.
So it's important that step 1 happens early so that xlogin
is actually sending off status packets. (It doesn't matter
whether anything is listening to them, since status packets
aren't answered.)
3. Mike will grab a snapshot of the liblarv, libares,
larvnetd, and cviewd code for to run on dill. To get any
interesting information during the testing of 8.2.11, we
should start with a configuration with clusters drawn from
machines which use the dev cell system packs--perhaps
"crash", "alpha", and "beta". I'll draw up the
configuration file if no one else is psyched to.
4. Once steps 2 and 3 are done, we can look at the data on
dill and see if it looks pretty. If it looks pretty and
nothing else has gone wrong, we can put 8.2.11 out to the
Athena cell about two weeks after it goes into the dev
cell.
5. The larvnet configuration on dill should then be pointed at
the real clusters. At this point, we should do testing
with xcluster (using local named records for cview.sloc) to
make sure that it works with larvnetd. (If it doesn't, we
probably just have to fix larvnetd, not the client support,
so there's no great need to do this testing early.)
6. Once everything looks good, we can point cview.sloc at
dill instead of minos.