[
https://issues.apache.org/jira/browse/HBASE-5074?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13198472#comment-13198472
]
Phabricator commented on HBASE-5074:
------------------------------------
mbautin has commented on the revision "[jira] [HBASE-5074] Support checksums in
HBase block cache".
@dhruba: thanks for the update! See my replies inline.
INLINE COMMENTS
src/main/java/org/apache/hadoop/hbase/io/hfile/AbstractHFileReader.java:357 I
don't see any overrides of this method in HFileReaderV{1,2} in the patch, and
this particular method looks really confusing, since it takes a parameter,
ignores it, and returns this.hfs instead. Did you mean to override it in a way
that does use the parameter? In that case, could you please add a javadoc here
explaining why the argument is being ignored?
src/main/java/org/apache/hadoop/hbase/io/hfile/HFile.java:402 Agreed. Perhaps
we should avoid replacing all occurrences of FileSystem with HFileSystem. One
class cast is much simpler.
src/main/java/org/apache/hadoop/hbase/io/hfile/HFile.java:417 OK.
src/main/java/org/apache/hadoop/hbase/io/hfile/HFileBlock.java:123 How much
more work is it to make it configurable? Otherwise we would be storing the
bytes-per-checksum field but not actually using it, which would be really
confusing.
src/main/java/org/apache/hadoop/hbase/io/hfile/HFileBlock.java:149 Sounds
good. Could you replace comments with javadocs? That seems to be the convention
in HBase code even for private fields.
src/main/java/org/apache/hadoop/hbase/io/hfile/HFileBlock.java:441-442 OK,
sounds good. I probably just misread the code.
src/main/java/org/apache/hadoop/hbase/io/hfile/HFileBlock.java:788 It is fine
to leave duplicate code between DataInputStream and ByteBuffer implementations
for performance reasons. However, I still think it is better to move these into
a separate utility class, e.g. ByteBufferUtils.
src/main/java/org/apache/hadoop/hbase/io/hfile/HFileBlock.java:866-870 Sounds
good.
src/main/java/org/apache/hadoop/hbase/io/hfile/HFileBlock.java:1525 This is
probably an error, not a warning, as we are about to shut down the regionserver.
src/main/java/org/apache/hadoop/hbase/io/hfile/HFileBlock.java:1520-1527 This
constructor will be called in the same thread that tries to read the block (see
ThreadLocal.get() implementation). I am not sure if throwing a RuntimeException
will shut down the regionserver. But this type of error definitely too serious
to recover from gracefully, so this is probably fine.
Just to make sure: are we planning to swap checksum implementations in
production? In that case, most RPC threads will still keep their associated
PrefetchedHeader instance with the wrong checksum class.
src/main/java/org/apache/hadoop/hbase/io/hfile/HFileBlock.java:1871 Sounds
good. In that case it is probably better to add a method call to an external
utility method here, instead of putting checksum calculation inline.
REVISION DETAIL
https://reviews.facebook.net/D1521
> support checksums in HBase block cache
> --------------------------------------
>
> Key: HBASE-5074
> URL: https://issues.apache.org/jira/browse/HBASE-5074
> Project: HBase
> Issue Type: Improvement
> Components: regionserver
> Reporter: dhruba borthakur
> Assignee: dhruba borthakur
> Attachments: D1521.1.patch, D1521.1.patch
>
>
> The current implementation of HDFS stores the data in one block file and the
> metadata(checksum) in another block file. This means that every read into the
> HBase block cache actually consumes two disk iops, one to the datafile and
> one to the checksum file. This is a major problem for scaling HBase, because
> HBase is usually bottlenecked on the number of random disk iops that the
> storage-hardware offers.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
