On Thu, Sep 02, 2004 at 09:37:24AM -0700, Kevin Baker wrote: > This could never happen. > > The message id is based on the time that it comes in. > > If one server goes down, and messages come in while it is > down... when it comes back up the messages are all in the > timeslot while it was down (ids based on this timeslot) > > Now any new messages on the server that just came back up > are also based on the time... meaning now, so they will go > after the messages it is syncing ...
I was thinking about the case where a server doesn't go down, but is disconnected from the rest of the servers. Any latency in synchronizing the databases illustrates the problem, though. Example: Say there two smtp servers each injecting messages into separate database servers which are setup in a multiple master scenario. One server inserts message 2004090202104527001 and the other inserts message 2004090202104527002. Before the databases have synchronized, an IMAP server connects to the second database server and asks for a list of messages. It gets 2004090202104527002. Later, after the databases have synchronized, it again asks for the messages and gets 2004090202104527001 and 2004090202104527002. >From the client's perspective, the unique id's have now decreased. > A prev message in thread discussed this solution in > project. Apparently it works great. Are you referring to Simon Cocking's message? I don't think he was referring to IMAP. xn
