Rational support correctly suspected that we were using the soft option in our automount maps for the mount directories used by our newer views. This could explain failures, since under heavy load, soft mounts do timeout.
Soft mounts were used for two mount directories: /hosts
and /node (and thus /nfs, which was a soft link to /node).
Our vobs all use /nfs (shown with
ct lsvob),
i.e. have been affected as long as this setting has been in use. The
directory used by our views (shown with
ct lsview)
is selected with our ccase.makeview
script, and has changed from /nfs to /hosts (thus this change alone
shouldn't explain anything).
/node was used by the so-called -hosts map (see /etc/auto_master), i.e. implicitly mounts all directories made accessible through the local /etc/exports files.
Hard mounts are written in /etc/fstabs. They may as well be soft (but shouldn't in general).