Generally when we thought of this Secondary indexing soln, the index table is generally an inverted table but the definition of the rowkey differs and how many of such inverted table changes. In Huawei's impl it is 1. Next point is, should the CP calls cross the Region server boundaries or it should be with in that RS. Now this is more applicable for the writes. As Jesse rightly points out the index table should be consistent with the main table and making anything asynchronous in this write path would definitely make consistency a big problem and definitely leads to dead locks. The problem of RPC handlers and going across RS is now a tradeoff with colocation. I think we will fall in any one of the category.
Finally, we internally have discussed on scope of having cross region transactions atleast in the case of co-located regions. This is an interesting problem but if we have this we would be able to have this write path(with index) more reliable. Regards Ram On Tue, Jan 21, 2014 at 4:33 AM, Andrew Purtell <apurt...@apache.org> wrote: > On Mon, Jan 20, 2014 at 2:53 PM, Jesse Yates <jesse.k.ya...@gmail.com > >wrote: > > > .In Phoenix, > > we kind of get around the deadlock issue by separating out the write > layer > > for indexes to make it queuable and using our own thread pools to make it > > somewhat async. It just ends up being hard to ensure consistency when you > > don't block on the write being committed :) > > > > Yes and yes. > > > > -- > Best regards, > > - Andy > > Problems worthy of attack prove their worth by hitting back. - Piet Hein > (via Tom White) >
