[jira] Commented: (HADOOP-1134) Block level CRCs in HDFS

["Raghu Angadi (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HADOOP-1134?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12510431 ] 

Raghu Angadi commented on HADOOP-1134:
--------------------------------------


Looks like fixing pread needs more changes than I imagined. Is it ok if I make sure that current patch is not a regression and fix pread properly as a follow up Jira? These are some of the issues I see:

# There is no synchronization around call to chooseDataNode(), updating deadNodes etc.
# Inside chooseDataNode, if not datanode could be found, it does openInfo(). But new updated information is not used in the next iteration of the loop.
# new call to openInfo() might fetch only part of the list of blocks. 
# Ideally the method that invokes chooseDataNode should retry chooseDataNode so thant chooseDataNode could be called with updated block locations.
# making chooseDataNode() synchronized will trigger a findBugs warning since it can sleep for 3 seconds.
# We need to support the case where a file might stay open for a very long time (one hour?) with many simultaneous preads at the same time.

For now, I can make sure that fetchBlockByteRange() (used by pread),  synchronizes around accesses around deadNodes etc. So that it does not mess up the InputStream's state.


> 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, BlockLevelCrc-07032007.patch, BlockLevelCrc-07052007.patch, DfsBlockCrcDesign-05305007.htm, readBuffer.java, readBuffer.java
>
>
> 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.