Github user mallman commented on a diff in the pull request:
https://github.com/apache/spark/pull/16499#discussion_r101602014
--- Diff: core/src/main/scala/org/apache/spark/storage/BlockManager.scala
---
@@ -813,7 +813,14 @@ private[spark] class BlockManager(
false
}
} else {
- memoryStore.putBytes(blockId, size, level.memoryMode, () =>
bytes)
+ val memoryMode = level.memoryMode
+ memoryStore.putBytes(blockId, size, memoryMode, () => {
+ if (memoryMode == MemoryMode.OFF_HEAP) {
--- End diff --
(Actually, I think we need to check `memoryMode.useOffHeap` here.)
Assume `memoryMode.useOffHeap` is true. We have two cases to consider:
1. `bytes` is on-heap. In this case, we need to copy it into a new direct
buffer, and that's what we're doing here.
2. `bytes` is off-heap. In this case, we assume that the caller upstream is
managing the memory underlying `bytes`, and
`bytes.copy(Platform.allocateDirectBuffer)` becomes a defensive copy. If the
caller is not managing this memory, I would call that a bug in the caller's
behavior.
In either case, I believe we should be calling
`bytes.copy(Platform.allocateDirectBuffer)` when `memoryMode.useOffHeap` is
true.
BTW, in my experience tracing this code in the debugger, `bytes` has always
been an on-heap buffer.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
