On Mar 4, 11:06 am, "Ikai Lan (Google)" <ika...@google.com> wrote: > > As far as eventual consistency goes, our users that are on high replication > have not seen any issues. If we weren't always fussing about transparency, > we would probably have been able to not even bring it up, and it's unlikely > anyone would have noticed - we just figured it'd make more sense to > communicate it, though now we are seeing that users are overreacting a bit. > Did you know index creation is also asynchronous? That is, when you save a > property, it returns before indexes have been created? Technically, this > constitutes a consistency issue, but it has no real world impact because it > is instantaneous. We documented eventual consistency issues because they are > an order of magnitude slower than asynchronous index creation, so in some > small cases of "write an entity then do a cross entity group query" the > query results will be off. If you've run any kind of SQL service at scale > that replicates to a read-slave, you should be used to this pattern of > persistence.
Eventual Consistency is really the #1 reason I'm reluctant to make any change to HR. I sell virtual goods, people pay real money for things that exist in the datastore. If they buy something, and it doesn't show up for minute/hours, I get complaints. Your reply seems to indicate that maybe Eventual Consistency isn't quite that inconsistent, but I haven't seen any data to express just how long things are out of sync, so I have to expect the worst. > We plan on setting High Replication as the default for all new applications. > It's not quite removing it, but perhaps we will explore that direction, > though we are current hesitant to do so. > We also plan on having some guest blog posts ... soon. If we had the > manpower, we would work with each and every one of you to migrate over, then > help you benchmark. Unfortunately, we can't. We can, however, try to publish > as much data as we have as it comes in. I think the most useful data here > will be data from third-parties. As far as our tests go, everything seems > alright, but I'm pretty sure you'll all have more confidence in the results > if they're being written by users whose entire businesses depend on the > reliability and performance of App Engine. I think making HR the default is probably a good idea, if that is the preferred path Google wants to take with the service. Then people won't come to expect things that may not exist in the future. As an early adopter, I understand I have to suffer the pain of things going another direction for the sake of improvement, on occasion. > Lastly, I want to apologize if I came off as irate. I'm usually quick to > respond emotionally, and the damn "undo send" button went away too quickly > in my gmail. I don't mean to shut down any kind of joking around, but do > remember that this group is internationally read. Sarcasm doesn't come off > well in email, and it comes off 10x more confusing for non-native English > speakers. If there's any chance any of you guys are coming to Pycon in > Atlanta next week, Wesley and I will be there. Let's smooth things over with > some drinks/food =). It's ok, I tend to do the same. As always I really appreciate all the work everyone on your end has put into GAE and how responsive you guys can be to issues and concerns. It's always good to see from a company. -- You received this message because you are subscribed to the Google Groups "Google App Engine" group. To post to this group, send email to google-appengine@googlegroups.com. To unsubscribe from this group, send email to google-appengine+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/google-appengine?hl=en.
