[
https://issues.apache.org/jira/browse/HBASE-2941?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12909574#action_12909574
]
HBase Review Board commented on HBASE-2941:
-------------------------------------------
Message from: "Ryan Rawson" <[email protected]>
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://review.cloudera.org/r/846/#review1217
-----------------------------------------------------------
src/main/java/org/apache/hadoop/hbase/ipc/HBaseServer.java
<http://review.cloudera.org/r/846/#comment4131>
this seems to be some kind of coordinate hand off so that the reader gets
notified immediately of new things to read.
this is mostly a straightforward port of the hadoop bug so i cant answer to
the suitability of the design really... but it does work... reliably.
src/main/java/org/apache/hadoop/hbase/ipc/HBaseServer.java
<http://review.cloudera.org/r/846/#comment4129>
Not really sure, i dont think it should be possible, since this thread will
only have 'accept' bits set on the sockets... but this is just a
straightforward port of the HADOOP bug.
src/main/java/org/apache/hadoop/hbase/ipc/HBaseServer.java
<http://review.cloudera.org/r/846/#comment4130>
they used to accept up to 10 connections, then read from 1, then go back
into the select loop.
now the listener thread only accepts 1, then passes that to another thread
to do the reads, so it doesnt have to attempt to keep socket accept() delays
down by accepting batches at a time.
- Ryan
> port HADOOP-6713 - threading scalability for RPC reads - to HBase
> -----------------------------------------------------------------
>
> Key: HBASE-2941
> URL: https://issues.apache.org/jira/browse/HBASE-2941
> Project: HBase
> Issue Type: Bug
> Affects Versions: 0.20.6, 0.89.20100621
> Reporter: ryan rawson
> Assignee: ryan rawson
> Fix For: 0.90.0
>
> Attachments: HBASE-2941.patch, HBASE-2941.txt, HBASE-2941.xlsx,
> HBaseServer.java, Rplot001.png
>
>
> HADOOP-6713 has patch to fix the read scalability of hadoop rpc. Right now
> a single thread accepts() then receives the RPC payload for every single RPC
> in hbase. Including object creation, writable deserialization, etc.
> Apply the patch from that issue to our own forked HBaseRPC code.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
