[
https://issues.apache.org/jira/browse/KAFKA-527?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14355133#comment-14355133
]
Guozhang Wang commented on KAFKA-527:
-------------------------------------
Hi Yasuhiro,
I thought for compressed writes, the linked list buffers in
BufferingOutputStream still need to be copied to a newly allocated buffer (in
line 54/55 of ByteBufferMessageSet) whereas for MemoryRecord, it append
messages to the compressed stream in-place and no extra copy is required at the
end of the writes, but I may misunderstood Scala's function-parameter syntax
and please let me know if I did.
As for the migration plan, I agree that ByteBufferMessageSet replacement would
not come in the near future, and we can definitely commit the patches now as
compress / de-compress has been a pain for us.
> Compression support does numerous byte copies
> ---------------------------------------------
>
> Key: KAFKA-527
> URL: https://issues.apache.org/jira/browse/KAFKA-527
> Project: Kafka
> Issue Type: Bug
> Components: compression
> Reporter: Jay Kreps
> Assignee: Yasuhiro Matsuda
> Priority: Critical
> Attachments: KAFKA-527.message-copy.history, KAFKA-527.patch,
> java.hprof.no-compression.txt, java.hprof.snappy.text
>
>
> The data path for compressing or decompressing messages is extremely
> inefficient. We do something like 7 (?) complete copies of the data, often
> for simple things like adding a 4 byte size to the front. I am not sure how
> this went by unnoticed.
> This is likely the root cause of the performance issues we saw in doing bulk
> recompression of data in mirror maker.
> The mismatch between the InputStream and OutputStream interfaces and the
> Message/MessageSet interfaces which are based on byte buffers is the cause of
> many of these.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
