[
https://issues.apache.org/jira/browse/SPARK-21517?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16099685#comment-16099685
]
zhoukang commented on SPARK-21517:
----------------------------------
[~kiszk] In our production cluster we use 1.6.1 and 2.1.0 which all reproduce
this case.However we has not upgrade to 2.2.
But i compared with 2.2,the logic here has not been modified.
> Fetch local data via block manager cause oom
> --------------------------------------------
>
> Key: SPARK-21517
> URL: https://issues.apache.org/jira/browse/SPARK-21517
> Project: Spark
> Issue Type: Improvement
> Components: Block Manager, Spark Core
> Affects Versions: 1.6.1, 2.1.0
> Reporter: zhoukang
>
> In our production cluster,oom happens when NettyBlockRpcServer receive
> OpenBlocks message.The reason we observed is below:
> When BlockManagerManagedBuffer call ChunkedByteBuffer#toNetty, it will use
> Unpooled.wrappedBuffer(ByteBuffer... buffers) which use default
> maxNumComponents=16 in low-level CompositeByteBuf.When our component's number
> is bigger than 16, it will execute during buffer copy.
> {code:java}
> private void consolidateIfNeeded() {
> int numComponents = this.components.size();
> if(numComponents > this.maxNumComponents) {
> int capacity =
> ((CompositeByteBuf.Component)this.components.get(numComponents -
> 1)).endOffset;
> ByteBuf consolidated = this.allocBuffer(capacity);
> for(int c = 0; c < numComponents; ++c) {
> CompositeByteBuf.Component c1 =
> (CompositeByteBuf.Component)this.components.get(c);
> ByteBuf b = c1.buf;
> consolidated.writeBytes(b);
> c1.freeIfNecessary();
> }
> CompositeByteBuf.Component var7 = new
> CompositeByteBuf.Component(consolidated);
> var7.endOffset = var7.length;
> this.components.clear();
> this.components.add(var7);
> }
> }
> {code}
> in CompositeByteBuf which will consume some memory during buffer copy.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
