mridulm commented on code in PR #38064:
URL: https://github.com/apache/spark/pull/38064#discussion_r1005275497
##########
core/src/main/scala/org/apache/spark/util/io/ChunkedByteBuffer.scala:
##########
@@ -84,6 +91,74 @@ private[spark] class ChunkedByteBuffer(var chunks:
Array[ByteBuffer]) {
}
}
+ /**
+ * write to ObjectOutput with zero copy if possible
+ */
+ override def writeExternal(out: ObjectOutput): Unit = {
+ // we want to keep the chunks layout
+ out.writeInt(chunks.length)
+ chunks.foreach(buffer => out.writeInt(buffer.limit()))
+ chunks.foreach(buffer => out.writeBoolean(buffer.isDirect))
+ var buffer: Array[Byte] = null
+ val bufferLen = ChunkedByteBuffer.COPY_BUFFER_LEN
+
+ getChunks().foreach { chunk => {
+ if (chunk.hasArray) {
+ // zero copy if the bytebuffer is backed by bytes array
+ out.write(chunk.array(), chunk.arrayOffset(), chunk.limit())
+ } else {
+ // fallback to copy approach
+ if (buffer == null) {
+ buffer = new Array[Byte](bufferLen)
+ }
+ var bytesToRead = Math.min(chunk.remaining(), bufferLen)
+ while (bytesToRead > 0) {
+ chunk.get(buffer, 0, bytesToRead)
+ out.write(buffer, 0, bytesToRead)
+ bytesToRead = Math.min(chunk.remaining(), bufferLen)
+ }
Review Comment:
There is `Utils.writeByteBuffer` which writes a buffer to an `OutputStream`.
It does not do a very good job, since it allocates a buffer the same size as
`remaining` - instead, we should enhance it to do what this method is doing.
Additionally, we can use a `ThreadLocal[Array[Byte]]` in `Utils` for use
with this copy (the `buffer` here)
--
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]
