on all DHCP events of interest.

Remove jaildog, and use the common watchdog instead.

script both inside and outside the jail.

run with the user group set. Bug reported by Eric.

On linux, do it the old way (talk directly to boss) until I get those
images updated with new stuff.

Add pidfile argument to program-agent.
Add logfile argument to program-agent.
Move uid/gid flip into the program-agent so that it can do a proper
setgroups, instead of the silly way that perl does it.
Add argument so that agent talks to local elvind, instead of boss
elvind (proxy always runs now).

multiplexing of nodes.

inside jails.

routing tables, a jail can be a router quite nicely.

federation, which is not supported in the version we have source to.
Basically, we run an elvind on each node. The proxy on each node
subscribes to all events for that node from the boss elvind, and hands
them to the local elvind, Each client on the node subscribes to the
local elvind, and gets its events via the proxy. This should reduce
the number of connections to boss, and makes it possible to run agents
inside each virtual node without an FD explosion on boss.

Add commented out find -delete command to remove all mysql logs older
than 35 days.  I just run it by hand at the moment, but I'll turn it
on at some point.

update V3 nodes to V4 CDs, and the new image. By inserting an entry
into this table, the node will update its image the next time it
reboots and checks in for instructions. The actual instructions are
hardwired in the web still, but this allows Dave to schedule updates
as needed. Eventually, the instructions will be part of the DB state
as well, but this works.

started before the interval (so, all we see is a swapout).

Note, I'm not dealing with the creator/swapper uid thing yet. TODO.

more useful, by including the reasons and such.

Also add a similar email message when they change the reasons or timeouts.

Epoch listing. Maybe later if we need it (since it reaches back to the
dawn of Emulab, todays experiments are not likely to have a big
influence on the numbers).

Only lookup the Bossnode name once

swap request idle message.

Add -u option to specify the user. Do the uid flip here instead of in
the perl wrapper, but only if root of course. Otherwise runs as the
user that invoked the program-agent.
Add mandatory -e option to speicfy the pid/eid to use in event tuple
instead of the ipaddr, since in jails without their own IP address,
using the ipaddr is broken (all jails see all program events).
Add mandatory -a option to specify a list of object names, so that the
agent will get just the events it should. There is a corresponding new
tmcc command that specifies the list of program objects for the node
(or vnode).

Add -l option to specify the logfile.
Add code to write pidfile, and add -i option to specify pidfile name.
Useful with jails where there is a delay agent per jail, and thus
a logfile and pidfile per jail.