On Mar 8, 2010, at 4:26 AM, Dirkjan Ochtman wrote: > On Fri, Mar 5, 2010 at 20:27, Adam Kocoloski <[email protected]> wrote: >> If we were to do this, I'd implement it that 2nd way, where the reader only >> hands over contiguous blocks to the writer, and doesn't "get too far ahead >> of itself" > > I think this would be very good to have. If it's really performance > sensitive, maybe making it optional would be useful. > > But in practice, CouchDB already gets this correct some large > percentage of the time (at least it has in continuous replication, in > my application), so I think it would be pretty nice to have. > > Adam, may I file an issue for you? I did try looking at the code > myself, but this code seems rather too involved for someone new at > Erlang.
Sure, go right ahead. > BTW, some semi-related issue: would it be sensible to have an > include_docs option to changes feeds? Right now every line of the > continuous changes feed requires a doc fetch, it seems like a > no-brainer to have an option to just yield the doc along with the seq > update. That makes sense to me, and could be useful in many replication scenarios. For example, if the replicator can deduce that a particular replication is transferring all new updates, it should start skipping the missing_revs check on the target and just throw every update at it. In that case it would be quite nice to have the documents shipped in the changes feed directly, so that only the attachments would need a separate request. Best, Adam
