[
https://issues.apache.org/jira/browse/FLINK-9913?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16580814#comment-16580814
]
ASF GitHub Bot commented on FLINK-9913:
---------------------------------------
zhijiangW commented on a change in pull request #6417: [FLINK-9913][runtime]
Improve output serialization only once in RecordWriter
URL: https://github.com/apache/flink/pull/6417#discussion_r210195493
##########
File path:
flink-runtime/src/main/java/org/apache/flink/runtime/io/network/api/serialization/RecordSerializer.java
##########
@@ -66,29 +66,33 @@ public boolean isFullBuffer() {
}
/**
- * Starts serializing and copying the given record to the target buffer
- * (if available).
+ * Starts serializing the given record to an intermediate data buffer.
*
* @param record the record to serialize
- * @return how much information was written to the target buffer and
- * whether this buffer is full
*/
- SerializationResult addRecord(T record) throws IOException;
+ void serializeRecord(T record) throws IOException;
Review comment:
I agree with not exposing `RecordSerializer`'s internal private fields to
`SerializedRecord` for safe concern.
Regarding with the way of reference of the `RecordSerializer` directly in
`SerializedRecord`. The whole process in `RecordWriter` may be like this:
SerializedRecord serializedRecord =
serializer.serializeRecord(record);
for (int targetChannel = 0; targetChannel < numChannels;
targetChannel++) {
//reset position for multiple copying
serializer.reset();
BufferBuilder bufferBuilder =
getBufferBuilder(targetChannel);
CopyingResult result =
serializedRecord.copyToBufferBuilder(bufferBuilder);
}
serializer.prune();
The `SerializedRecord` only provides the method of `copyToBufferBuilder` and
wraps the internal class `CopyingResult`, but the actual copying operation is
still done in `RecordSerializer`. So the `RecordSerializer` also needs provide
the method `copyToBufferBuilder`. From this point, we actually do not separate
the two steps of `serialization` and `copy` from `RecordSerializer`. So I think
maybe it is not very worth doing that. :)
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
> Improve output serialization only once in RecordWriter
> ------------------------------------------------------
>
> Key: FLINK-9913
> URL: https://issues.apache.org/jira/browse/FLINK-9913
> Project: Flink
> Issue Type: Improvement
> Components: Network
> Affects Versions: 1.6.0
> Reporter: zhijiang
> Assignee: zhijiang
> Priority: Minor
> Labels: pull-request-available
> Fix For: 1.7.0
>
>
> Currently the {{RecordWriter}} emits output into multi channels via
> {{ChannelSelector}} or broadcasts output to all channels directly. Each
> channel has a separate {{RecordSerializer}} for serializing outputs, that
> means the output will be serialized as many times as the number of selected
> channels.
> As we know, data serialization is a high cost operation, so we can get good
> benefits by improving the serialization only once.
> I would suggest the following changes for realizing it.
> # Only one {{RecordSerializer}} is created in {{RecordWriter}} for all the
> channels.
> # The output is serialized into the intermediate data buffer only once for
> different channels.
> # The intermediate serialization results are copied into different
> {{BufferBuilder}}s for different channels.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
