On Thursday, December 4, 2014 1:37:03 PM UTC-8, [email protected] wrote: > > Hi Jens, > > Thank you for your clear reply. I am glad to hear that there is a solid > plan to improve the situation. > > 1. Would you be able to shine more light on the technical details about > how DCP solves the delivery of out of order and delayed updates? > > 2. Am I right that sequences can get permanently lost (e.g. if a sequence > is acquired with an incr call, but then the update fails)? If so, even when > using DCP, how will Sync Gateway be able to differentiate between > permanently lost sequences and delays? > > 3. If we increase the timeout and provision enough hardware resources, but > the changes feed with a gap still gets delivered, the only way one could > get the client that received that feed to request the missed sequence is to > do a sync from the start. > > However, there does not seem to be a way (i.e. there is no API) to trigger > a sync with no "since" parameter in Couchbase Lite. > > One thing we've talked about is the possibility of sending changes that show up on TAP during the out of order waiting interval of a missing sequence, with a lower-than-actual sequence number on the wire (use the last stable sequence number, the one before the missing one.)
So this allows low-latency delivery of newer messages, and then for the sequence number to catch up when any waiting changes come through TAP. The cost is potential revs diff chatter on any sync client that disconnects / reconnects during a window where we are waiting for a sequence to come through TAP. So we can avoid latency issues and instead have the client rollback window remain open at the earlier point, by only ever sending the lower stable sequence number in changes feeds. > As it is very hard to predict the exact load one will see, especially in > AWS where sometimes things like disk and network delays spike, it is > difficult to ensure that not a single update will be missed. In these rare > cases we would love to be able to fix the problem for the customers who > have been hit with it. Without the API to do a sync from scratch, from what > I can see, we have no way to fix the problem, and have nothing to tell our > customers who have hit it. We cannot be in this situation :) We need to be > able to guide a customer to a resolution where they have all their data up > to date. > > 4. Thank you for the suggestion to increase the time out in the Sync > Gateway. > > Would it be possible to do this in your code base? And maybe even accept a > parameter from the command line? > > As you release new versions of Sync Gateway we would love to be able to > deploy them, without having to rebuild our own version each time. > > Thanks a lot for your help, Jens! We look forward to using your solution > and working with you to improve it. > > Cheers, > Anton > > -- You received this message because you are subscribed to the Google Groups "Couchbase Mobile" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/mobile-couchbase/7765a16e-72ea-4d63-97a1-9c7387a8e654%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
