Aggregate pushdown v26.5+

Aggregate pushdown controls whether a provider or Kubling evaluates an aggregate operation. It changes where the work is performed, not whether Kubling can execute the SQL query.

Why it matters

With pushdown	Without pushdown
The provider evaluates supported aggregation and returns the reduced result.	The provider returns the rows required to evaluate the aggregation in Kubling.
Fewer rows usually cross the provider boundary.	More rows may cross the provider boundary.
Source-native indexes, partitions or aggregation facilities may be used.	Kubling uses its own memory and execution resources for the operation.
Performance depends mainly on the source-side plan.	Performance depends more heavily on transferred cardinality and engine capacity.

Missing aggregate pushdown does not mean that aggregation is unsupported. Kubling can retain the operation in the federated plan and produce the same logical result.

When Kubling can push down

Kubling delegates an aggregate only when the provider explicitly advertises the complete operation it can preserve. Function support, grouping, DISTINCT and HAVING are independent capabilities; support for one does not imply support for the others.

An aggregate also remains in Kubling when the data source uses its MVCC overlay, because uncommitted or simulated local state must participate in the result.

Capability declaration is a correctness promise, not a preference. Never advertise an aggregate or query shape whose null, type, filter, grouping or overflow semantics differ from Kubling’s contract.

Unsupported or unrepresentable expressions stay in the engine. Kubling does not infer aggregate support from other provider capabilities.

Example

For a provider that advertises grouping, COUNT_STAR and HAVING:


SELECT status, COUNT(*) AS task_count
FROM provider.TASK
GROUP BY status
HAVING COUNT(*) > 1

With pushdown, the provider can return one row per status after applying the group and HAVING condition. Without it, Kubling requests the necessary source rows and performs that work in the federated engine. The result is equivalent, but the amount of transferred data and engine work can be substantially different.

Provider-specific support

The exact function and query-shape matrix belongs to each provider because it can change independently from Kubling. Use the Provider catalog to reach the provider’s documentation and release notes. The catalog’s aggregate-pushdown badge is a discovery aid, not a per-function compatibility matrix.