Re: Distributive SQL Joins

2015-08-11 Thread Sergi Vladykin
I've created respective Jira issue https://issues.apache.org/jira/browse/IGNITE-1232 Sergi 2015-08-11 9:31 GMT+03:00 Dmitriy Setrakyan : > On Mon, Aug 10, 2015 at 11:26 PM, Sergi Vladykin > > wrote: > > > I was thinking about protecting users from doing stupid things, > > but ok, we can do a br

Re: Distributive SQL Joins

2015-08-10 Thread Dmitriy Setrakyan
On Mon, Aug 10, 2015 at 11:26 PM, Sergi Vladykin wrote: > I was thinking about protecting users from doing stupid things, > but ok, we can do a broadcast as well. > I think we should print out a warning in this case for sure. Is there a Jira created for this? We should provide a link to this dis

Re: Distributive SQL Joins

2015-08-10 Thread Sergi Vladykin
I was thinking about protecting users from doing stupid things, but ok, we can do a broadcast as well. Sergi 2015-08-08 3:46 GMT+03:00 Dmitriy Setrakyan : > Sergi, > > I personally don't like that for certain types of queries we will be > throwing an exception. > > After analyzing the approaches

Re: Distributive SQL Joins

2015-08-07 Thread Dmitriy Setrakyan
Sergi, I personally don't like that for certain types of queries we will be throwing an exception. After analyzing the approaches you suggested, I can think of cases where A performs better than B, as well as when B performs better than A. However, if you prefer B, I don't mind us taking that ap

Re: Distributive SQL Joins

2015-08-07 Thread Sergi Vladykin
Alexey, 1. Yes, in my plan it should work exactly like that: if both keys in join are affinity keys, then we are fully collocated, if only one then we can run join remotely as described, if none of them we will fail to run the query. 2. I mean we don't have values for these affinity keys in our l

Re: Distributive SQL Joins

2015-08-07 Thread Alexey Kuznetsov
Sergi, Questions about plan "B" :) 1) It is possible to throw exception on query prepare state (fail fast) when we don't know remote affinity key? 2) Could you provide an example when we don't know remote affinity key? I think we always have some default affinity (no?)? -- Alexey Kuznetsov GridG

Distributive SQL Joins

2015-08-06 Thread Sergi Vladykin
Igniters, I'm going to start working on distributed SQL joins soon and want to put my thoughts on this matter here. To provide collocated and non-collocated joins we need to know affinity key for each table. For example we have a table `Organization(id int)` where `id` is affinity key and `Perso