Hi William, Currently we are not supporting index merge optimizations.
Here is the issue I have raised for the same thing. Would be better to share details there to show the interest. https://issues.apache.org/jira/browse/PHOENIX-1801 Can't you combine the columns in both indexes into a single index to avoid index merge and make the query run faster? Thanks, Rajeshbabu. On Wed, Jul 20, 2016 at 2:33 PM, William <yhxx...@163.com> wrote: > Hi all, > I have a question about global secondary index in Phoenix 4.6. See the > following statements: > > create table yy (pk integer primary key, cf.a integer, cf.b integer); > create index yya on yy(cf.a); > create index yyb on yy(cf.b); > > then upsert some data into table yy; do the following query: > > select pk from yy where cf.a = 1 or cf.b = 2; > select /*+INDEX(yy yya yyb)*/ pk from yy where cf.a = 1 or cf.b = 2; > > I expect that both index tables will be used in this query, and join the > results from both index tables and return. > But unfortunately, no index tables has been used but a full table scan > instead. The statement with a hint have the same behaviour. The explain > plan is : > > FULL SCAN OVER YY > SERVER FILTER BY (CF.A = 1 OR CF.B = 2) > > Another example: > > create index yyi on yy (cf.a) include (cf.b); > select pk from yy where cf.a = 1 or cf.b =2; > > then this query will hit the index table with a filter. The explain plan: > > FULL SCAN OVER YYI > SERVER FILTER BY (TO_INTEGER("A") = 1 OR CF."B" = 2) > > my question is : I only have index yya and yyb, and i want the above > select statement to hit both index tables, do we support this scenario? > if so, how can i use both indexes? > if not, why? It is too hard to implement ? Is there a plan to support it ? > > Thanks > William > > > >
