Thanks to (with paraphrased answers)
Simon May
Yep! He sees that too, so have a cronjob to stop/sleep/start it every night
because lpnet has a memory leak.
Luca Pizzinato
Sees the problem when a printqueue on a remote machine but the remote
printer was disconnected.
Jim Harmon
tmpfs has memory fragmentation problems and it serves me right for not
rebooting regularly (the machine had uptime of 120 days).
As you see, nothing definitive. I'll probably do the cron entry. I discount
the remote printer problem, since this machine is the print server, printing
via hpnpf.
I really dislike the idea of tmpfs fragmentation. IMHO a machine should be
able to stay up for 100+ days without problems, and indeed stopping/starting
the lp service returns the "swap -s" output to sane values. A "df" on /tmp
shows the filesystem taking minimal (a few 100K) of memory space, consisting
of "psdata", "upsdata", "jadump", and samba's temporary "lpq" output.
rgds
Stephen