[
https://issues.apache.org/jira/browse/HADOOP-1134?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12486429
]
Raghu Angadi commented on HADOOP-1134:
--------------------------------------
> I think we should handle this the same way we'll handle things if/when a CRC
> file is missing after the upgrade.
After the upgrade, I think it is cleaner and simpler to treat this as hard
error on the block. i.e, block will be considered badly corrupt handled
accordingly.
> That shouldn't happen, but it might, and we need to think about what we
> should do in that case. My guess is that we should
> return a null checksum with the data when it is read, and let the client
> decide whether to accept or reject the unchecksummed data
How do we handle transfer from datanode to another.
I understand it will be better to be flexible, but one way or the we have to
deal with real hard errors (mostly because of hardware errors). If our software
is so buggy that we need to expect CRC file not to exists and handle it as an
'expected condition', I think it would be better to spend more time fixing
those bugs. I vote against treating this as soft error.
Regd option to serve possibly corrupt data, I was thinking of making client to
explicitly ask datanode to ignore checksum errors at the beginning of reading
data from the datanode (possibly based on client config). Since CRC is served
inline on the connection, we should have some conventions like 'checksum of
0000 followed by some magic 8 bytes means checksum is incorrect' or some such
thing.
> 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.
