[Date Prev][Date Next]
Re: syncrepl questions
At 03:31 PM 9/24/2003, Quanah Gibson-Mount wrote:
>syncrepl in persistent mode is working successfully for me now on our development systems, although I am not currently able to test it under the conditions we have in production. I agree about a more sophisticated engine to be in place eventually. Observationally, Netscape kept a store of changes from the time the server was loaded with the DB until it next got reloaded (which would, for us, be when it next corrupted itself). I think that strategy is a bit overkill. Our general strategy has been nightly stop/starts of the master. This allows us to export the DB, and clear up the resources that slapd & slurpd slowly eat over the course of a day. In the bit that Jong was talking about a log store, I was wondering if it'd be useful to have the store record changes between restarts of the server. There is of course some issues with that as well.
Yes, there are lots of issues with histories, especially
I'm thinking that non-durable session-specific histories would
be "good enough" for most uses. These only need only record
which UUIDs left the sync context (you can determined those
changed or entered using current state information). Durability
would be a nice addition but adds significant complexity.
I'm thinking session-independent histories are too expensive
to maintain and use for large scale directories.