[
https://issues.apache.org/jira/browse/HADOOP-1134?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12486164
]
Raghu Angadi commented on HADOOP-1134:
--------------------------------------
Sameer talked to me about the following approach for upgrade. The following is
based on the discussion :
1) Namenode consideres upgrade complete only when 100% of blocks have at least
one replica upgraded (as described below). We will still have a manual
override call to mark upgrade complete (i.e. finalizeUpgrade() call that was
mentioned earlier).
2) During upgrade, datanodes fetch all the replicas of CRC data (C1, C2, C3.. )
for a given block B. If the CRC replicas match, it stores generates new CRC
file with the data. There is no verification with the block data in this case.
If not all CRCs match but majority of them match, then majority CRC will be
used. If all are different then, we just pick one (with least block id?).
3) If there is no existing CRC data available for some reason, we generate from
existing block data. A warning will be printed.
4) If a datanode comes into cluster with old data directories and notices that
namenode has completed upgrade, it upgrades each of the blocks like this:
4.a) It gets the datanode that has an 'upgraded replica' for a block and
contacts that datanode for the CRC data.
4.b) If can not find an upgraded replica, it generates local CRC.. Or
should it delete it?
4.c) It waits for some time (half an hour) before deciding if an upgraded
replica is 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.
