[
https://issues.apache.org/jira/browse/HBASE-17278?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15744192#comment-15744192
]
Enis Soztutar commented on HBASE-17278:
---
- Can you remove the DLOG statements (or maybe move them to a debug method):
{code}
+#if 0
{code}
- Let's keep Cell class to be only the data structure, and move parsing /
encoding logic to KeyValueCodec class. We can use the same Codec structure here
as well with the Encoder and Decoder classes.
- Normally, CellScanner is an interface in Java with different
implementations. {{Result}} also implements the CellScanner so, it maybe useful
to have this to be an abstract base class, and the KeyValueDecoder to be an
implementation.
- Let's use ntohs and friends if possible for dealing with endianness.
(https://linux.die.net/man/3/ntohs) rather than SwapByteOrder methods.
- Can we do this with zero-copy by overtaking the buffer:
{code}
+void CellScanner::SetData(char *cell_block_data, int data_length) {
+ cell_block_data_.reset(new char[data_length]);
+ std::memcpy(cell_block_data_.get(), cell_block_data, data_length);
+ data_length_ = data_length;
+ cur_pos_ = 0;
+}
{code}
The only CellScanner implementation for now will be the scanner that will work
on top of the IPC buffers, so we do not need to memcpy the bytes. Better yet,
we should use the
https://github.com/facebook/folly/blob/master/folly/io/IOBuf.h for the
CellScanner since that is what will be coming from the network using wangle.
- Do we need the second Cell wrapper here?
{code}
new Cell(*Cell::ParseCellData(*current_cell_data))
{code}
> Cell Scanner Implementation to be used by ResultScanner
> ---
>
> Key: HBASE-17278
> URL: https://issues.apache.org/jira/browse/HBASE-17278
> Project: HBase
> Issue Type: Sub-task
>Reporter: Sudeep Sunthankar
>Assignee: Sudeep Sunthankar
> Attachments: HBASE-17278.HBASE-14850.v1.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
