Thanks, Damien, that's what I had suspected. Best, Adam
On Jan 26, 2009, at 7:34 PM, Damien Katz wrote:
Everything the replicator does, except for update the replication
records at the end, is idempotent. The updating of documents for
replication must be idempotent as it might be racing the same
document that's replicating in from another source.
-Damien
On Jan 26, 2009, at 7:14 PM, Adam Kocoloski wrote:
Jason and I were working through COUCHDB-197 and discussing whether
to retry failed HTTP requests during replication. GET requests are
surely safe to retry, but after I thought about it for awhile it
seemed like the POSTs that occur during replication are also safe.
The POST to _missing_revs on the target server clearly does not
change the state of the server; I assume it's a POST because the
request has a body. _bulk_docs is not normally an idempotent
operation, but when "new_edits":false is sent along with the docs
it looks like it might be, since no new revisions are generated.
Adam
