[ZODB-Dev] Cache Query (why doesn't RAM usage ever drop?)

[Toby Dickenson]

On Friday 25 October 2002 5:44 pm, sean.upton@uniontrib.com wrote:
> This is a good idea, but yet another alternative: if the process that i=
s a
> complex operation needs to be on a server that is seen by the world
> at-large, shouldn't a cluster with 2 ICP-enabled Zope ZEO clients behin=
d
> Squid via ICP+HTTP be able to handle the restart of one of the nodes
> without too much of a bump?

Yes.

With Zope 2.6 you get a small blip. When Zope 2.6 is shut down it severs =
any=20
all sockets instantly. ICP takes care of new requests, but it does cause =
a=20
problem for requests where Zope has half-transmitted the response at the=20
instant of shutdown.

This is resolved in toby-clean-shutdown-branch. This merges cleanly into =
2.6,=20
I am currently using it in production, and I plan to merge it with the tr=
unk=20
for 2.7.