[ZODB-Dev] Lag in ZEO updating across clients?

[Dieter Maurer]

Dario Lopez-Kästen wrote at 2004-4-29 10:23 +0200:
> ...
>However I am seeing a lag of up to a minute or so before the ZEO clients 
>actually see the changes. The files we are moving range in size from a 
>few kilobytes to tens of megabytes.
>
>Is this normal behaviour?

A lag is normal (and cannot be avoided).

However, 1 min is quite large.
How long does it take to write your largest files to disk?

ZEO is essentially single threaded (not quite true: communication
with the clients is handled by a medusa thread, packing is done
by separate threads but the main storage server is in a single
thread). Therefore, while it is performing a large transaction
commit, it is doing nothing else. After the commit, it sends
invalidation messages to the clients. They observe the invalidation
after the invalidation arrived and after the current transaction
is completed (which may again be quite long, if it handles a huge file).

>Can I tune this?

I doubt it...

>what factors affect the 
>propagation of changes accross the clients?

The duration of a transaction, the number of invalidated objects
and the number of clients.

-- 
Dieter