Thanks to Jon LaBadie, Gene Heskett, and Jean-Louis Martineau for replies.

Jean-Louis supplied me with a patch which I modified slightly and that I'm 
now using.

It turns out that some code in server-src/driver.c will allow a dump 
to be retried indefinitely, if dumper returns TRYAGAIN and taper returns 
DONE.  See around line 2203 in 2.4.4p3 server-src/driver.c.

I've been using amanda for years with real tapes but I've just started 
using vtapes this past week and this is the first time I've noticed such 
behaviour.  This is a working amanda configuration that I've used to dump 
many hundreds of gigabytes already, both to real tapes in the past and to 
vtapes in recent days.  But consistently with vtapes, if a client is
alive during planning and was added to the schedule, and then this client 
goes offline before the dump starts, this condition will occur whereby 
dumper says TRYAGAIN, taper says DONE, forever until killed (or until 
taper gets an error for some other reason I guess).

--
Todd Pfaff <pfaff AT mcmaster DOT ca>
Research & High-Performance Computing Support
ABB 132, McMaster University, Hamilton, Ontario, Canada, L8S 4M1
phone:  voice: 905-525-9140 ext 22920  fax: 905-528-3773