[ZODB-Dev] ZEO release?

[Toby Dickenson]

On Tue, 08 Jan 2002 13:41:35 -0800, sean.upton@uniontrib.com wrote:

>Jeremy Hylton writes:  >>>>>
>[snip]... The new features are (1)
>automatic failover from one ZSS to another and (2) explicit support
>for read-only clients and servers.

Sounds good.

>Just a curiosity: In regards to #1, I assumed that this was already =
possible
>with existing IP takeover software (i.e. heartbeat) using earlier =
releases
>of ZEO, provided you are willing to wait (perhaps up to 5 minutes) for =
the
>client to reconnect (I'm already doing this with a slightly-modified =
1.0b3).

Why the wait? ClientStorage can be configured to retry reconnections
as fast as you need, and the IP takeover shouldnt take more than a few
tens of seconds?

>Are the improvements just designed to decrease the time the clients are
>disconnected without trying to resetablish the connection?  Or is there
>something more complex (like cache-related issues) that I am currently
>neglecting in my setup that necessitate the mentioned changes to ZEO =
2.0?

Im not sure what ZC are planning for this ZEO enhancement, but
application-level redundancy has many advantages over heartbeat-style
IP takeover. For one, you avoid the problem that the hearteat media is
a single point of failure. Also, the two nodes dont have to be on the
same network.

A good model for application-level redundancy is DNS: Every client
always knows of _more_than_one_ server which can answer every request.
Compared to DNS, heartbeat is an ugly hack.

The disadvantage, of course, is the need for application-specific
support.



Toby Dickenson
tdickenson@geminidataloggers.com