[
https://issues.apache.org/jira/browse/HADOOP-1134?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12500476
]
Doug Cutting commented on HADOOP-1134:
--------------------------------------
Re the patch: As I've mentioned before, I'd prefer it if the CRC code could be
shared with CheckSumFileSystem. In particular, it seems to me that
FSInputChecker and FSOutputSummer could be extended to support pluggable
sources and sinks for checksums, respectively, and DFSDataInputStream and
DFSDataOutputStream could use these. Advantages of this are: (a) single
implementation of checksum logic to debug and maintain; (b) keeps checksumming
as close to possible to data generation and use. This patch computes checksums
after data has been buffered, and validates them before it is buffered. We
sometimes use large buffers and would like to guard against in-memory errors.
The current checksum code catches a lot of such errors. So we should compute
checksums after minimal buffering (just bytesPerChecksum, ideally) and validate
them at the last possible moment (e.g., through the use of a small final buffer
with a larger buffer behind it). I do not think this will significantly affect
performance, and data integrity is a high priority.
> Block level CRCs in HDFS
> ------------------------
>
> Key: HADOOP-1134
> URL: https://issues.apache.org/jira/browse/HADOOP-1134
> Project: Hadoop
> Issue Type: New Feature
> Components: dfs
> Reporter: Raghu Angadi
> Assignee: Raghu Angadi
> Attachments: bc-no-upgrade-05302007.patch,
> DfsBlockCrcDesign-05305007.htm
>
>
> Currently CRCs are handled at FileSystem level and are transparent to core
> HDFS. See recent improvement HADOOP-928 ( that can add checksums to a given
> filesystem ) regd more about it. Though this served us well there a few
> disadvantages :
> 1) This doubles namespace in HDFS ( or other filesystem implementations ). In
> many cases, it nearly doubles the number of blocks. Taking namenode out of
> CRCs would nearly double namespace performance both in terms of CPU and
> memory.
> 2) Since CRCs are transparent to HDFS, it can not actively detect corrupted
> blocks. With block level CRCs, Datanode can periodically verify the checksums
> and report corruptions to namnode such that name replicas can be created.
> We propose to have CRCs maintained for all HDFS data in much the same way as
> in GFS. I will update the jira with detailed requirements and design. This
> will include same guarantees provided by current implementation and will
> include a upgrade of current data.
>
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
