Once more unto the breach, dear friends; once more!

[Dan Scott]

After much discussion with my colleagues in the world of Evergreen
Thursday morning, and after having pared down many of the other
resource drains on our system (for example blocking most bots, and
disabling catalogues and Z39.50 endpoints for long-lost partners that
still attract a surprising amount of traffic) I'm trying a complete
reversal of my previous strategy and instead configuring a much
smaller initial pool of web server processes with smaller increases
once more requests come in. The server will be much more aggressive
about killing off the processes once requests die down and will be
hastier, too, killing a process after it has served only a small
number of requests.

When new processes need to be started up, there will be a little
lag--but this aggressive pruning of web server processes should avoid
the rather horrible result of running out of memory (nothing but
network timeouts), and the lag might be minimal given the overall
capacity of the system.

In initial testing, the results seem promising on both the public
catalogue and staff client sides, but as we all know by now, the real
proof will come tomorrow when everyone is working on the system at the
same time.

The game's afoot!