[
https://issues.apache.org/jira/browse/HADOOP-1134?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12485605
]
Sameer Paranjpye commented on HADOOP-1134:
------------------------------------------
> But actually, now that I think about it, if we're primarily not validating
> checksums against the data, but rather comparing all the checksums for a
> block, then locality may not be
> worthwhile. In that case we'd want a temporary datanode extension that
> permits writing the checksum file for a block. Then the updater map task can
> read through all copies of
> the checksum file, construct the best possible checksum for each block, then
> send these to datanodes. So, in aggregate, 6% of the filesystem would cross
> the wire during the
> upgrade. Could that work?
If the client is going to send CRCs to Datanodes then we can have each split be
a filename or a list of filenames, this will work fine if we don't want to
validate data. If we want validation a map task can be scheduled local to 1
instance of most or all blocks of a file i.e. scheduled on the node where the
file was generated, it can validate local data, fall back to remote data if
local validation fails, then write checksums to all the block instances.
How do we manage blocks that are missing during the upgrade? There are 3 cases
really:
- Blocks that are entirely missing i.e. no instances are available
- Blocks that have some instances missing
- Blocks whose checksums are missing
> 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.
