Jefffrey commented on code in PR #18540:
URL: https://github.com/apache/datafusion/pull/18540#discussion_r2508467339
##########
datafusion/physical-optimizer/src/coalesce_batches.rs:
##########
@@ -72,11 +72,26 @@ impl PhysicalOptimizerRule for CoalesceBatches {
)
})
.unwrap_or(false);
+
if wrap_in_coalesce {
Ok(Transformed::yes(Arc::new(CoalesceBatchesExec::new(
plan,
target_batch_size,
))))
+ } else if let Some(async_exec) =
plan_any.downcast_ref::<AsyncFuncExec>() {
Review Comment:
I believe the idea is that for async functions, we are specifically
interested in batching together inputs to the function so ideally it is not
called as often (which can be expensive for async function). Whereas for
coalesce batches in general, it looks like it considers the output of a node
and if it is too small it wraps that note in a coalesce.
So for async, we consider input to aync node for coalesce logic; for other
node types we look at their output for coalesce logic
We should copy the comment from `coalesce_async_exec_input.rs` here to not
lose that context, e.g.
> Coalesce inputs to async functions to reduce number of async function
invocations
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
