xintongsong commented on a change in pull request #16771:
URL: https://github.com/apache/flink/pull/16771#discussion_r686600519
##########
File path: flink-core/src/main/java/org/apache/flink/api/dag/Transformation.java
##########
@@ -273,7 +273,8 @@ public ResourceSpec getPreferredResources() {
* @param managedMemoryUseCase The use case that this transformation
declares needing managed
* memory for.
* @param weight Use-case-specific weights for this transformation. Used
for sharing managed
- * memory across transformations for OPERATOR scope use cases.
+ * memory across transformations for OPERATOR scope use cases. For
consistency, the APIs
+ * declare their weights as a kibibyte value.
Review comment:
I'm ok with either KB or MB.
- For KB, it needs 2^31 total weights to get an overflow. That's roughly
2PB, per slot, which is not likely the case.
- For MB, I also don't see much chances that an operator needs to tune its
memory as fine-grained as less than 1MB.
--
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]
