[
https://issues.apache.org/jira/browse/FLINK-7517?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16245330#comment-16245330
]
ASF GitHub Bot commented on FLINK-7517:
---------------------------------------
Github user zhijiangW commented on a diff in the pull request:
https://github.com/apache/flink/pull/4594#discussion_r149886365
--- Diff:
flink-runtime/src/main/java/org/apache/flink/runtime/io/network/netty/NettyBufferPool.java
---
@@ -52,51 +48,61 @@
/** Configured chunk size for the arenas. */
private final int chunkSize;
+ /** We strictly prefer direct buffers and disallow heap allocations. */
+ private static final boolean PREFER_DIRECT = true;
+
+ /**
+ * Arenas allocate chunks of pageSize << maxOrder bytes. With these
defaults, this results in
+ * chunks of 16 MB.
+ *
+ * @see #MAX_ORDER
+ */
+ private static final int PAGE_SIZE = 8192;
+
+ /**
+ * Arenas allocate chunks of pageSize << maxOrder bytes. With these
defaults, this results in
+ * chunks of 16 MB.
+ *
+ * @see #PAGE_SIZE
+ */
+ private static final int MAX_ORDER = 11;
+
/**
* Creates Netty's buffer pool with the specified number of direct
arenas.
*
* @param numberOfArenas Number of arenas (recommended: 2 * number of
task
* slots)
*/
public NettyBufferPool(int numberOfArenas) {
+ super(
+ PREFER_DIRECT,
+ // No heap arenas, please.
+ 0,
+ // Number of direct arenas. Each arena allocates a
chunk of 16 MB, i.e.
+ // we allocate numDirectArenas * 16 MB of direct
memory. This can grow
+ // to multiple chunks per arena during runtime, but
this should only
+ // happen with a large amount of connections per task
manager. We
+ // control the memory allocations with low/high
watermarks when writing
+ // to the TCP channels. Chunks are allocated lazily.
+ numberOfArenas,
+ PAGE_SIZE,
+ MAX_ORDER);
+
checkArgument(numberOfArenas >= 1, "Number of arenas");
--- End diff --
Is it better to checkArgument before call super method?
> let NettyBufferPool extend PooledByteBufAllocator
> -------------------------------------------------
>
> Key: FLINK-7517
> URL: https://issues.apache.org/jira/browse/FLINK-7517
> Project: Flink
> Issue Type: Sub-task
> Components: Network
> Affects Versions: 1.4.0
> Reporter: Nico Kruber
> Assignee: Nico Kruber
>
> {{NettyBufferPool}} wraps {{PooledByteBufAllocator}} but due to this, any
> allocated buffer's {{alloc()}} method is returning the wrapped
> {{PooledByteBufAllocator}} which allowed heap buffers again. By extending the
> {{PooledByteBufAllocator}}, we prevent this loop hole and also fix the
> invariant that a copy of a buffer should have the same allocator.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
