Hi, Devs,
When I try to develop a new connector, which support aggregate push
down. I found that Flink aggregate pushdown is not user-friendly. The
`AggregateExpression` passed to the connector by
`SupportsAggregatePushDown#applyAggregates` doesn't provide access to
subclasses. This makes it impossible for me to directly determine the type
of agg operator unless I import the planner module, but this is discouraged
and considered a heavyweight action.
Because I cann't access the subclasses of
`AggregateExpression#FunctionDefinition`, like `CountAggFunction` and am
unable to import the planner module, I'm forced to match the agg operators
using hack way that match fully qualified class names like the following
code:
FunctionDefinition functionDefinition =
aggregateExpressions.get(0).getFunctionDefinition();
if (!(functionDefinition
.getClass()
.getCanonicalName()
.equals(
"org.apache.flink.table.planner.functions.aggfunctions.CountAggFunction")
|| functionDefinition
.getClass()
.getCanonicalName()
.equals(
"org.apache.flink.table.planner.functions.aggfunctions.Count1AggFunction")))
{
return false;
}
I think the problem may also exist with other SupportsXxxPushDown. Should
we consider which planner classes can be exposed to developers to
facilitate their use?
Yours,
Swuferhong (Yunhong Zheng).
