Query Planner Pools

You can improve the performance of the router's query planner by configuring parallelized query planning.

By default, the query planner plans one operation at a time. It plans one operation to completion before planning the next one. This serial planning can be problematic when an operation takes a long time to plan and consequently blocks the query planner from working on other operations.

Configuring query planner pools

To resolve such blocking scenarios, you can enable parallel query planning. Configure it in router.yaml with supergraph.query_planning.experimental_parallelism:

1supergraph:
2  query_planning:
3    experimental_parallelism: auto # number of available cpus 

The value of experimental_parallelism is the number of query planners in the router's query planner pool. A query planner pool is a preallocated set of query planners from which the router can use to plan operations. The total number of pools is the maximum number of query planners that can run in parallel and therefore the maximum number of operations that can be worked on simultaneously.

Valid values of experimental_parallelism:

• Any integer starting from 1

• The special value auto, which sets the number of query planners equal to the number of available CPUs on the router's host machine

The default value of experimental_parallelism is 1.

In practice, you should tune experimental_parallelism based on metrics and benchmarks gathered from your router.