[OpenAFS-port-darwin] Using AFS home directories under Tiger

[Ian Ward Comfort]

Here at Stanford we've been testing AFS home directories with =20
Kerberized LDAP users under Tiger. Other work prevents me from =20
investigating the code itself but I though others might benefit from =20
our testing experience. So I'm writing to report some issues we've =20
been seeing, in case this helps the active developers or in case =20
anyone has ideas to help us. Feel free to contact me for more detail.

Our setup is as follows. We're using this build (http://=20
www.contrib.andrew.cmu.edu/~cg2v/unreleased/OpenAFS-Tiger.pkg.tgz) of =20=

the AFS client with an aklog binary from here (http://=20
www.contrib.andrew.cmu.edu/~cg2v/unreleased/aklog4-Tiger.bin) (thanks =20=

Chaskiel!). We are on 10.4.2 so we are not munging AuthAuthority for =20
the Kerberos users. AFS root is mounted at startup, and =20
NFSHomeDirectory is mapped for each LDAP user to the appropriate path =20=

in /afs. Aklog is installed in root's loginwindow.plist.

We've been seeing a number of issues, apparently non-deterministic. =20
Some users seem to have more problems than others but I can't tell if =20=

that's chance or not. Even when the machine hangs it does not go =20
down, I can still ssh in with a local admin account and run ps to =20
take a look at the processes.

User A
In heavy use over several days, A observes frequent loginwindow =20
hangs, most common immediately after a reboot or (oddly) a change in =20
the machine's AppleTalk or Bonjour name. A also reports that often =20
the second or third application opened after login (no matter what =20
the app) fails to open any windows and enters uninterruptible wait. =20
Afterwards all other applications accessible by mouse or keyboard, =20
including Finder, also enter uninterruptible wait. The machine must =20
be rebooted to regain the console. Either issue may appear regardless =20=

of whether there is anything in the user's home folder, or whether =20
the user's home is created from the template on the Tiger test =20
machine or from the template on a working Panther machine. Once the =20
machine hung when A attempted to logout.

User B
B was able to log in and out several times and work with several =20
applications for about 15 minutes with no problems, until after one =20
login the third application opened (Dictionary.app) entered =20
uninterruptible wait and took Finder and loginwindow with it.

User C
C worked on the Tiger machine for upwards of 25 minutes with no =20
problems, logging in and out several times, often in quick succession.

User D
D used the machine for 10 or 15 minutes with no problems, browsing =20
the AFS tree and opening several applications. (At one point D mis-=20
entered a key combination in Finder and appeared to delete an AFS =20
directory on which he had insufficient rights. The directory =20
disappeared from Finder though it still could be seen from the =20
Terminal and certainly still existed. D does not know what keys he =20
pressed; this may or may not be related to problems with the AFS =20
client.)

User E
On first login, E's desktop failed to appear, his loginwindow process =20=

hung and the machine had to be rebooted. Afterwards he used the =20
machine for 25 minutes with no further issues.

User F
User F logged in once, tried to open Safari and immediately hung the =20
machine. Safari, Finder, &c. entered uninterruptible wait.

User G
The machine behaved well for G for an hour or more of heavy use =20
(including streaming media and activity in AFS home directory =20
including mounting disk images from there). Then G activated the =20
Dashboard (not for the first time). Two widgets requiring network =20
access failed to complete their display though calculator displayed =20
correctly. All DashboardClient processes entered uninterruptible =20
wait, though several apps below the dashboard layer (including =20
Safari, Finder and iChat) remained alive. The Dashboard could not be =20
dismissed so no console activity was possible, but after a few =20
seconds G saw iChat reporting that he had been unexpectedly logged =20
out of a chat.  Windowserver remained active, and killing Safari and =20
Finder with signals caused their windows to disappear. Loginwindow =20
had entered uninterruptible wait however, and a reboot was necessary =20
to get the console back.

=97IWC