Github user JoshRosen commented on a diff in the pull request:
https://github.com/apache/spark/pull/9241#discussion_r43456699
--- Diff: core/src/main/java/org/apache/spark/memory/TaskMemoryManager.java
---
@@ -101,29 +105,103 @@
private final boolean inHeap;
/**
+ * The size of memory granted to each consumer.
+ */
+ private final HashSet<MemoryConsumer> consumers;
+
+ /**
* Construct a new TaskMemoryManager.
*/
public TaskMemoryManager(MemoryManager memoryManager, long
taskAttemptId) {
this.inHeap = memoryManager.tungstenMemoryIsAllocatedInHeap();
this.memoryManager = memoryManager;
this.taskAttemptId = taskAttemptId;
+ this.consumers = new HashSet<>();
}
/**
- * Acquire N bytes of memory for execution, evicting cached blocks if
necessary.
+ * Acquire N bytes of memory for a consumer. If there is no enough
memory, it will call
+ * spill() of consumers to release more memory.
+ *
* @return number of bytes successfully granted (<= N).
*/
- public long acquireExecutionMemory(long size) {
- return memoryManager.acquireExecutionMemory(size, taskAttemptId);
+ public long acquireExecutionMemory(long required, MemoryConsumer
consumer) {
+ assert(required >= 0);
+ synchronized (this) {
+ long got = memoryManager.acquireExecutionMemory(required,
taskAttemptId);
+
+ // try to release memory from other consumers first, then we can
reduce the frequency of
+ // spilling, avoid to have too many spilled files.
+ if (got < required) {
+ // Call spill() on other consumers to release memory
+ for (MemoryConsumer c: consumers) {
--- End diff --
Does this approach still have the same concern about concurrent
modification of `consumers` while iterating over it?
---
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]
