You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@accumulo.apache.org by John McNamee <jp...@terpmail.umd.edu> on 2014/05/22 21:57:28 UTC

Review Request 21826: Accumulo-2632

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/21826/
-----------------------------------------------------------

Review request for accumulo.


Repository: accumulo


Description
-------

The goal was to find a way to increase the write speed in the DfsLogger using compression.
This patch uses the Snappy compression library and was the most optimum way that I could find to increase write performance.
The option to use compression has been made configurable.
Recovery has been modified to uncompress the WAL if needed.


Diffs
-----

  core/src/main/java/org/apache/accumulo/core/conf/Property.java 62b0a33 
  server/tserver/src/main/java/org/apache/accumulo/tserver/log/DecompressingInputStream.java PRE-CREATION 
  server/tserver/src/main/java/org/apache/accumulo/tserver/log/DfsLogger.java bfa51d3 
  server/tserver/src/test/java/org/apache/accumulo/tserver/log/DecompressingInputStreamTest.java PRE-CREATION 

Diff: https://reviews.apache.org/r/21826/diff/


Testing
-------

There is a unit test for the DecryptingInputStream.java, and I ran continuous Ingest/ continuous verify with the agitator on.


Thanks,

John McNamee


Re: Review Request 21826: Accumulo-2632

Posted by John McNamee <jp...@terpmail.umd.edu>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/21826/
-----------------------------------------------------------

(Updated Nov. 24, 2014, 1:26 p.m.)


Review request for accumulo.


Changes
-------

Link to the JIRA issue.


Bugs: ACCUMULO-2632
    https://issues.apache.org/jira/browse/ACCUMULO-2632


Repository: accumulo


Description
-------

The goal was to find a way to increase the write speed in the DfsLogger using compression.
This patch uses the Snappy compression library and was the most optimum way that I could find to increase write performance.
The option to use compression has been made configurable.
Recovery has been modified to uncompress the WAL if needed.


Diffs
-----

  core/src/main/java/org/apache/accumulo/core/conf/Property.java 62b0a33 
  server/tserver/src/main/java/org/apache/accumulo/tserver/log/DecompressingInputStream.java PRE-CREATION 
  server/tserver/src/main/java/org/apache/accumulo/tserver/log/DfsLogger.java bfa51d3 
  server/tserver/src/test/java/org/apache/accumulo/tserver/log/DecompressingInputStreamTest.java PRE-CREATION 

Diff: https://reviews.apache.org/r/21826/diff/


Testing
-------

There is a unit test for the DecryptingInputStream.java, and I ran continuous Ingest/ continuous verify with the agitator on.


Thanks,

John McNamee


Re: Review Request 21826: Accumulo-2632

Posted by Christopher Tubbs <ct...@apache.org>.

> On May 22, 2014, 4:28 p.m., Josh Elser wrote:
> > core/src/main/java/org/apache/accumulo/core/conf/Property.java, line 270
> > <https://reviews.apache.org/r/21826/diff/1/?file=588464#file588464line270>
> >
> >     Should this be tagged with the @Experimental annotation?

I view @Experimental to denote features that are more invasive, optional, and may not work as expected, not simply to denote things that users can experiment with. @Experimental hides these configuration items from a typical user's view, making the target audience developers, rather than end users. While it is currently experimental as a patch, I think the expectation here is that when this feature is ready for merging into a branch, the feature will be targeted to end users and made available for them to tweak, so @Experimental doesn't quite apply.


> On May 22, 2014, 4:28 p.m., Josh Elser wrote:
> > server/tserver/src/main/java/org/apache/accumulo/tserver/log/DfsLogger.java, line 241
> > <https://reviews.apache.org/r/21826/diff/1/?file=588466#file588466line241>
> >
> >     Could we leverage the Compression stuff in Hadoop rather than invoke the snappy lib directly?

I agree, this feature could be improved by allowing more compression types in the configuration, rather than simply enable/disable, and that should probably be done before merging in. However, there are special considerations with other compression types. Snappy allows you to take a byte array and make another byte array. The other types work with streams, and you'd have to ensure the stream is flushed before sync is called. Snappy's byte[]-to-byte[] is a bit easier to experiment with, as a first pass.


- Christopher


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/21826/#review43764
-----------------------------------------------------------


On May 22, 2014, 3:57 p.m., John McNamee wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/21826/
> -----------------------------------------------------------
> 
> (Updated May 22, 2014, 3:57 p.m.)
> 
> 
> Review request for accumulo.
> 
> 
> Repository: accumulo
> 
> 
> Description
> -------
> 
> The goal was to find a way to increase the write speed in the DfsLogger using compression.
> This patch uses the Snappy compression library and was the most optimum way that I could find to increase write performance.
> The option to use compression has been made configurable.
> Recovery has been modified to uncompress the WAL if needed.
> 
> 
> Diffs
> -----
> 
>   core/src/main/java/org/apache/accumulo/core/conf/Property.java 62b0a33 
>   server/tserver/src/main/java/org/apache/accumulo/tserver/log/DecompressingInputStream.java PRE-CREATION 
>   server/tserver/src/main/java/org/apache/accumulo/tserver/log/DfsLogger.java bfa51d3 
>   server/tserver/src/test/java/org/apache/accumulo/tserver/log/DecompressingInputStreamTest.java PRE-CREATION 
> 
> Diff: https://reviews.apache.org/r/21826/diff/
> 
> 
> Testing
> -------
> 
> There is a unit test for the DecryptingInputStream.java, and I ran continuous Ingest/ continuous verify with the agitator on.
> 
> 
> Thanks,
> 
> John McNamee
> 
>


Re: Review Request 21826: Accumulo-2632

Posted by Christopher Tubbs <ct...@apache.org>.

> On May 22, 2014, 4:28 p.m., Josh Elser wrote:
> > server/tserver/src/main/java/org/apache/accumulo/tserver/log/DfsLogger.java, line 572
> > <https://reviews.apache.org/r/21826/diff/1/?file=588466#file588466line572>
> >
> >     It looks like this sync invocation got dropped. Did I just miss it moving somewhere?

I think Keith was working on ACCUMULO-2766. Any feature improvement here should probably be rebased on top of that, to ensure that the concurrency issues are correct.


- Christopher


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/21826/#review43764
-----------------------------------------------------------


On May 22, 2014, 3:57 p.m., John McNamee wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/21826/
> -----------------------------------------------------------
> 
> (Updated May 22, 2014, 3:57 p.m.)
> 
> 
> Review request for accumulo.
> 
> 
> Repository: accumulo
> 
> 
> Description
> -------
> 
> The goal was to find a way to increase the write speed in the DfsLogger using compression.
> This patch uses the Snappy compression library and was the most optimum way that I could find to increase write performance.
> The option to use compression has been made configurable.
> Recovery has been modified to uncompress the WAL if needed.
> 
> 
> Diffs
> -----
> 
>   core/src/main/java/org/apache/accumulo/core/conf/Property.java 62b0a33 
>   server/tserver/src/main/java/org/apache/accumulo/tserver/log/DecompressingInputStream.java PRE-CREATION 
>   server/tserver/src/main/java/org/apache/accumulo/tserver/log/DfsLogger.java bfa51d3 
>   server/tserver/src/test/java/org/apache/accumulo/tserver/log/DecompressingInputStreamTest.java PRE-CREATION 
> 
> Diff: https://reviews.apache.org/r/21826/diff/
> 
> 
> Testing
> -------
> 
> There is a unit test for the DecryptingInputStream.java, and I ran continuous Ingest/ continuous verify with the agitator on.
> 
> 
> Thanks,
> 
> John McNamee
> 
>


Re: Review Request 21826: Accumulo-2632

Posted by Josh Elser <jo...@gmail.com>.

> On May 22, 2014, 8:28 p.m., Josh Elser wrote:
> > Have you done any benchmarks with these changes? I would think that you should see higher CPU usage with less disk I/O, and, overall, quicker ingest. It would be good to see some before and after numbers that show this.

Oops, sorry. I didn't see the attachment you put on the ticket :)


- Josh


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/21826/#review43764
-----------------------------------------------------------


On May 22, 2014, 7:57 p.m., John McNamee wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/21826/
> -----------------------------------------------------------
> 
> (Updated May 22, 2014, 7:57 p.m.)
> 
> 
> Review request for accumulo.
> 
> 
> Repository: accumulo
> 
> 
> Description
> -------
> 
> The goal was to find a way to increase the write speed in the DfsLogger using compression.
> This patch uses the Snappy compression library and was the most optimum way that I could find to increase write performance.
> The option to use compression has been made configurable.
> Recovery has been modified to uncompress the WAL if needed.
> 
> 
> Diffs
> -----
> 
>   core/src/main/java/org/apache/accumulo/core/conf/Property.java 62b0a33 
>   server/tserver/src/main/java/org/apache/accumulo/tserver/log/DecompressingInputStream.java PRE-CREATION 
>   server/tserver/src/main/java/org/apache/accumulo/tserver/log/DfsLogger.java bfa51d3 
>   server/tserver/src/test/java/org/apache/accumulo/tserver/log/DecompressingInputStreamTest.java PRE-CREATION 
> 
> Diff: https://reviews.apache.org/r/21826/diff/
> 
> 
> Testing
> -------
> 
> There is a unit test for the DecryptingInputStream.java, and I ran continuous Ingest/ continuous verify with the agitator on.
> 
> 
> Thanks,
> 
> John McNamee
> 
>


Re: Review Request 21826: Accumulo-2632

Posted by Josh Elser <jo...@gmail.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/21826/#review43764
-----------------------------------------------------------


Have you done any benchmarks with these changes? I would think that you should see higher CPU usage with less disk I/O, and, overall, quicker ingest. It would be good to see some before and after numbers that show this.


core/src/main/java/org/apache/accumulo/core/conf/Property.java
<https://reviews.apache.org/r/21826/#comment78049>

    Should this be tagged with the @Experimental annotation?



server/tserver/src/main/java/org/apache/accumulo/tserver/log/DecompressingInputStream.java
<https://reviews.apache.org/r/21826/#comment78054>

    Same comment as elsewhere, can we use what Hadoop has for compression to abstract out the specifics of Snappy?



server/tserver/src/main/java/org/apache/accumulo/tserver/log/DfsLogger.java
<https://reviews.apache.org/r/21826/#comment78051>

    Could we leverage the Compression stuff in Hadoop rather than invoke the snappy lib directly?



server/tserver/src/main/java/org/apache/accumulo/tserver/log/DfsLogger.java
<https://reviews.apache.org/r/21826/#comment78058>

    It looks like this sync invocation got dropped. Did I just miss it moving somewhere?


- Josh Elser


On May 22, 2014, 7:57 p.m., John McNamee wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/21826/
> -----------------------------------------------------------
> 
> (Updated May 22, 2014, 7:57 p.m.)
> 
> 
> Review request for accumulo.
> 
> 
> Repository: accumulo
> 
> 
> Description
> -------
> 
> The goal was to find a way to increase the write speed in the DfsLogger using compression.
> This patch uses the Snappy compression library and was the most optimum way that I could find to increase write performance.
> The option to use compression has been made configurable.
> Recovery has been modified to uncompress the WAL if needed.
> 
> 
> Diffs
> -----
> 
>   core/src/main/java/org/apache/accumulo/core/conf/Property.java 62b0a33 
>   server/tserver/src/main/java/org/apache/accumulo/tserver/log/DecompressingInputStream.java PRE-CREATION 
>   server/tserver/src/main/java/org/apache/accumulo/tserver/log/DfsLogger.java bfa51d3 
>   server/tserver/src/test/java/org/apache/accumulo/tserver/log/DecompressingInputStreamTest.java PRE-CREATION 
> 
> Diff: https://reviews.apache.org/r/21826/diff/
> 
> 
> Testing
> -------
> 
> There is a unit test for the DecryptingInputStream.java, and I ran continuous Ingest/ continuous verify with the agitator on.
> 
> 
> Thanks,
> 
> John McNamee
> 
>