Github user JoshRosen commented on a diff in the pull request:
https://github.com/apache/spark/pull/9084#discussion_r41826094
--- Diff:
core/src/main/scala/org/apache/spark/memory/UnifiedMemoryManager.scala ---
@@ -0,0 +1,138 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.memory
+
+import scala.collection.mutable
+
+import org.apache.spark.SparkConf
+import org.apache.spark.storage.{BlockStatus, BlockId}
+
+
+/**
+ * A [[MemoryManager]] that enforces a soft boundary between execution and
storage such that
+ * either side can borrow memory from the other.
+ *
+ * The region shared between execution and storage is a fraction of the
total heap space
+ * configurable through `spark.memory.fraction` (default 0.75). The
position of the boundary
+ * within this space is further determined by
`spark.memory.storageFraction` (default 0.5).
+ * This means the size of the storage region is 0.75 * 0.5 = 0.375 of the
heap space by default.
+ *
+ * Storage can borrow as much execution memory as is free until execution
reclaims its space.
+ * When this happens, cached blocks will be evicted from memory until
sufficient borrowed
+ * memory is released to satisfy the execution memory request.
+ *
+ * Similarly, execution can borrow as much storage memory as is free.
However, execution
+ * memory is *never* evicted by storage due to the complexities involved
in implementing this.
+ * The implication is that attempts to cache blocks may fail if execution
has already eaten
+ * up most of the storage space, in which case the new blocks will be
evicted directly.
+ */
+private[spark] class UnifiedMemoryManager(conf: SparkConf, maxMemory:
Long) extends MemoryManager {
+
+ def this(conf: SparkConf) {
+ this(conf, UnifiedMemoryManager.getMaxMemory(conf))
+ }
+
+ /**
+ * Size of the storage region, in bytes.
+ *
+ * This region is not statically reserved; execution can borrow from it
if necessary.
+ * Cached blocks can be evicted only if actual storage memory usage
exceeds this region.
+ */
+ private val storageRegionSize: Long = {
+ (maxMemory * conf.getDouble("spark.memory.storageFraction",
0.5)).toLong
+ }
+
+ /**
+ * Total amount of memory, in bytes, not currently occupied by either
execution or storage.
+ */
+ private def totalFreeMemory: Long = synchronized {
+ assert(_executionMemoryUsed <= maxMemory)
+ assert(_storageMemoryUsed <= maxMemory)
+ assert(_executionMemoryUsed + _storageMemoryUsed <= maxMemory)
+ maxMemory - _executionMemoryUsed - _storageMemoryUsed
+ }
+
+ /**
+ * Total available memory for execution, in bytes.
+ * In this model, this is equivalent to the amount of memory not
occupied by storage.
+ */
+ override def maxExecutionMemory: Long = synchronized {
+ maxMemory - _storageMemoryUsed
+ }
+
+ /**
+ * Total available memory for storage, in bytes.
+ * In this model, this is equivalent to the amount of memory not
occupied by execution.
+ */
+ override def maxStorageMemory: Long = synchronized {
+ maxMemory - _executionMemoryUsed
+ }
+
+ /**
+ * Acquire N bytes of memory for execution, evicting cached blocks if
necessary.
+ *
+ * This method evicts blocks only up to the amount of memory borrowed by
storage.
+ * Blocks evicted in the process, if any, are added to `evictedBlocks`.
+ * @return number of bytes successfully granted (<= N).
+ */
+ override def acquireExecutionMemory(
+ numBytes: Long,
+ evictedBlocks: mutable.Buffer[(BlockId, BlockStatus)]): Long =
synchronized {
+ assert(numBytes >= 0)
+ val memoryBorrowedByStorage = math.max(0, _storageMemoryUsed -
storageRegionSize)
+ // If there is not enough free memory AND storage has borrowed some
execution memory,
+ // then evict as much memory borrowed by storage as needed to grant
this request
+ val shouldEvictStorage = totalFreeMemory < numBytes &&
memoryBorrowedByStorage > 0
+ if (shouldEvictStorage) {
+ val spaceToEnsure = math.min(numBytes, memoryBorrowedByStorage)
+ memoryStore.ensureFreeSpace(spaceToEnsure, evictedBlocks)
--- End diff --
Is it possible that this call could be expensive? Just wondering whether
it's safe to be performing inside of the `synchronized` block. Note that we can
always optimize the locking later, too.
---
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]
