[
https://issues.apache.org/jira/browse/HBASE-12684?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14253170#comment-14253170
]
Jurriaan Mous commented on HBASE-12684:
---------------------------------------
{quote}
Why out of interest the below change?
830 return (value != null)? Integer.valueOf(value).intValue(): DEFAULT_TTL;
830 return (value != null) ? Integer.parseInt(value) : DEFAULT_TTL;
{quote}
I was fixing findbugs issues and this performance issue was too easy to ignore.
So I thought why not change it while I was at it. Is it ok or should I revert
it?
Findbug recommendation:
{quote}
Boxing/unboxing to parse a primitive
org.apache.hadoop.hbase.HColumnDescriptor.getMinVersions()
Bug type DM_BOXED_PRIMITIVE_FOR_PARSING (click for details)
In class org.apache.hadoop.hbase.HColumnDescriptor
In method org.apache.hadoop.hbase.HColumnDescriptor.getMinVersions()
Called method Integer.intValue()
Should call Integer.parseInt(String) instead
At HColumnDescriptor.java:[line 846]
{quote}
{quote}
The below is not a repeat because we are replacing old rpc with this new one?
81 private static final byte[] MAGIC = new byte[]{ 'H', 'B', 'a', 's' }
{quote}
That was my original thought. But I am now moving it to RpcClient interface as
a constant since it is so basic to the HBase RPC protocol.
bq. This looks great: 201 f.channel().pipeline().addFirst(saslHandler);
Yes I think that shows the nice pipeline architecture of Netty in 1 line of
code :)
bq. Patch looks great. On the tests that are zombies, they are probably missing
(timeout=XX) annotations after @Test. Feel free to add if it will help you
debug.
Will look into it. Those zombie tests were already indicating issues with the
client. Could add that parameter while I work on the remaining ones.
bq. When you get a chance, make a list of what you will be able to remove after
this is all working. I love removing code.
Will do! It will be mostly RpcClientImpl, Call and anything related with it.
Some methods in HBaseSaslRpcClient can be removed. More cleanup is possible
when AsyncProcess is swapped out for true async calls.
Thanks for the quick review!
> Add new AsyncRpcClient
> ----------------------
>
> Key: HBASE-12684
> URL: https://issues.apache.org/jira/browse/HBASE-12684
> Project: HBase
> Issue Type: Improvement
> Components: Client
> Reporter: Jurriaan Mous
> Assignee: Jurriaan Mous
> Attachments: HBASE-12684-v1.patch, HBASE-12684-v2.patch,
> HBASE-12684-v3.patch, HBASE-12684-v4.patch, HBASE-12684-v5.patch,
> HBASE-12684.patch
>
>
> With the changes in HBASE-12597 it is possible to add new RpcClients. This
> issue is about adding a new Async RpcClient which would enable HBase to do
> non blocking protobuf service communication.
> Besides delivering a new AsyncRpcClient I would also like to ask the question
> what it would take to replace the current RpcClient? This would enable to
> simplify async code in some next issues.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
