Shashikant Banerjee commented on HDDS-226:

Thanks [~anu], for the review comments.

Another orthogonal Question: Sorry for these random comments. I see we have 
BlockID class then we create a new class called OmBlockInfo and add one more 
field, blockLength. Why is this not added as part of BlockID
As a part of BlockCommitProtocol, we need to update each Block with the actual 
block length written as well as BlockCommitSequenceId.
BlockId class as i suppose is used for identifying a unique block across the 
system. So, I thought it would be less confusing to maintain
per Block associated Info for OzoneManager in a separate class. 
BlockCommitSequenceId will be added to the OzoneBlockInfo class in subsequent 
patches(Mentioned in a TODO item)

I am very confused with this code. Can you please check? Why are checking 
keyArgs, did you intend to check blockInfoList?

Yes, it was a mistake. Addressed in the latest patch.

OmKeysArgs.java: More of a question, when would this sum be not equal to the 
datasize ?

In the earlier patch, when we create a key we set Keylength and once we commit 
the key to OM, it was just updating each of the blockLengths , not exactly 
updating total size. Hence, once the key is committed, the actual key size will 
be determined by the sum total of all block lengths.
Patch v2 now, updates each of the block lengths as well as the total size of 
the Key, so the related code which you pointed out is removed.

Rest all the review comments are addressed.

> Client should update block length in OM while committing the key
> ----------------------------------------------------------------
>                 Key: HDDS-226
>                 URL: https://issues.apache.org/jira/browse/HDDS-226
>             Project: Hadoop Distributed Data Store
>          Issue Type: Bug
>          Components: Ozone Manager
>            Reporter: Mukul Kumar Singh
>            Assignee: Shashikant Banerjee
>            Priority: Major
>             Fix For: 0.2.1
>         Attachments: HDDS-226.00.patch, HDDS-226.01.patch
> Currently the client allocate a key of size with SCM block size, however a 
> client can always write smaller amount of data and close the key. The block 
> length in this case should be updated on OM.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org

Reply via email to