CodingCat commented on code in PR #53190:
URL: https://github.com/apache/spark/pull/53190#discussion_r2596980404
##########
core/src/main/scala/org/apache/spark/internal/config/package.scala:
##########
@@ -445,6 +445,34 @@ package object config {
"Ensure that memory overhead is a double greater than 0")
.createWithDefault(0.1)
+ private[spark] val EXECUTOR_BURSTY_MEMORY_OVERHEAD_ENABLED =
+ ConfigBuilder("spark.executor.memoryOverheadBursty.enabled")
+ .doc("Whether to enable memory overhead bursty")
+ .version("4.2.0")
+ .booleanConf
+ .createWithDefault(false)
+
+ private[spark] val EXECUTOR_BURSTY_MEMORY_OVERHEAD_FACTOR =
+ ConfigBuilder("spark.executor.memoryOverheadBurstyFactor")
+ .doc("the bursty control factor controlling the size of memory overhead
space shared with" +
+ s" other processes, newMemoryOverhead=oldMemoryOverhead-MIN((onheap +
memoryoverhead) *" +
+ s" (this value - 1), oldMemoryOverhead)")
+ .version("4.2.0")
+ .doubleConf
+ .checkValue((v: Double) => v >= 1.0,
+ "the value of bursty control factor has to be no less than 1")
+ .createWithDefault(1.2)
+
+ private[spark] val EXECUTOR_BURSTY_MEMORY_OVERHEAD = ConfigBuilder(
+ "spark.executor.burstyMemoryOverhead")
+ .doc(s"The adjusted amount of memoryOverhead to be allocated per executor"
+
+ s" (the adjustment happens if
${EXECUTOR_BURSTY_MEMORY_OVERHEAD_ENABLED.key} is enabled," +
+ " in MiB unless otherwise specified. This parameter is here only for UI
demonstration," +
+ " there is not effect when user sets it directly")
+ .version("4.2.0")
+ .bytesConf(ByteUnit.MiB)
+ .createOptional
+
Review Comment:
in theory, any cluster manager supporting request/limit-like facilities can
use this feature, we just add limit in resource profile and then consume in K8S
module
in practice, seems only k8s fits the role here until so far ...there is some
possibility to make the whole code locate within k8s module, but that will
significantly change the method signatures across classes.... so we just didn't
go too far along that path
--
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]
