andygrove commented on code in PR #1561:
URL: https://github.com/apache/datafusion-comet/pull/1561#discussion_r2008018651
##########
spark/src/main/scala/org/apache/comet/CometExecIterator.scala:
##########
@@ -63,9 +64,28 @@ class CometExecIterator(
}.toArray
private val plan = {
val conf = SparkEnv.get.conf
- // Only enable unified memory manager when off-heap mode is enabled.
Otherwise,
- // we'll use the built-in memory pool from DF, and initializes with
`memory_limit`
- // and `memory_fraction` below.
+
+ val offHeapMode = CometSparkSessionExtensions.isOffHeapEnabled(conf)
+ val memoryLimit = if (offHeapMode) {
+ // in unified mode we share off-heap memory with Spark
+ ByteUnit.MiB.toBytes(conf.getSizeAsMb("spark.memory.offHeap.size"))
Review Comment:
That is what I initially thought as well, but there is a new
`CometFairMemoryPool` that does wrap `comet_task_memory_manager` which is the
callback to Spark pool. The `try_grow` method had additional checks before
delegating to Spark though.
```rust
MemoryPoolType::FairUnified => {
// Set Comet fair memory pool for native
let memory_pool =
CometFairMemoryPool::new(comet_task_memory_manager,
memory_pool_config.pool_size);
Arc::new(memory_pool)
}
```
--
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]
