Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
3.0.0
-
None
-
None
Description
I have use case like : select * from t where ROW = ? (ROW is primary key),so it just a hbase get.
and it turn out a single query will do two RPCs for metadata, they are all calling to
public static ColumnResolver getResolverForQuery(SelectStatement statement, PhoenixConnection connection) throws SQLException { List<TableNode> fromNodes = statement.getFrom(); if (fromNodes.size() == 1) return new SingleTableColumnResolver(connection, (NamedTableNode)fromNodes.get(0), true);
as you can see "SingleTableColumnResolver(connection, (NamedTableNode)fromNodes.get(0), true)", the third parameter is hard coded to true, which means it never use cache. for performance, we let the query use metadata cache, our test show 30% improvement on RT in our use case.
in our patch, we do:
1. add an timer based updater at the global cache (the one in ConnectionQueryServices)
2. the phoenixconnection doesn't have it's own cache, it has a reference to global cache instead
i'm looking forward for your advice, thanks