[
https://issues.apache.org/jira/browse/HADOOP-1134?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12483308
]
Doug Cutting commented on HADOOP-1134:
--------------------------------------
> current proposal slightly relaxes this and expects the clients to trust DFS
> to maintain data
The client is a part of DFS. The primary change here should be how checksums
are stored. Instead of storing complete checksums for the entire file in a
FileSystem-based file, we store checksums per block on the datanodes. But the
checksums should still be computed in the client as data is written, and
validated in the client as it's read. Instead of writing a parallel file, the
client will send to the datanode with each block its checksums, and, when
reading, the client will receive a checksum with each range of data bytes.
Reads should always be aligned with checksum boundaries. Since the
bytes/checksum of a file may differ from the client's configuration, each read
should probably return something like <start, len, checksum, data>, where the
start is the position in the block of the data (at or before the requested
position), length is the length of the data, and the checksum is the checksum
for the data.
So this issue should not alter what's checksummed, when it's checksummed, where
it's checksummed, etc., but only where and how the checksums are stored. Right?
> 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
> Assigned To: Raghu Angadi
>
> 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.
