> On Feb 11, 2008, at 4:19 PM, Christian Theune wrote:
>> Context could be given as a reference date that is opaque to the
>> client
>> and can differ from storage to storage, a file pointer could serve
>> this
>> purpose. The API still might include the `length` of the data returned
>> to minimize round-trips.
> Actually, a file position doesn't work either, because the storage
> server can't know that much about what it is serving.

I would have thought that this wouldn't matter if we declare the context
`opaque`. As long as the storage can hand out a token that represents
context and that can be transferred through zrpc ...

> I suspect the only sane approach is to make the storage server
> maintain the iterator returned by the underlying storage on behalf of
> the client.  This will cause resources to be allocated on the server
> that might only be freed when the client disconnects.

Right, now this makes me feel uneasy. :)


For more information about ZODB, see the ZODB Wiki:

ZODB-Dev mailing list  -  ZODB-Dev@zope.org

Reply via email to