[jira] [Commented] (HDFS-4452) getAdditionalBlock() can create multiple blocks if the client times out and retries.

["Konstantin Shvachko (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HDFS-4452?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13565738#comment-13565738 ] 

Konstantin Shvachko commented on HDFS-4452:
-------------------------------------------

A bit more details. {{FSNamesystem.getAdditionalBlock()}} consists of two parts both surrounded by {{writeLock}}. First part validates different conditions on the file. The second actually adds a new block. The two parts are separated by {{chooseTarget()}}, which is performed outside the the {{writeLock}}.
In the error scenario the are two threads trying to perform the same operation {{getAdditionalBlock()}} with the same input parameters.
- First thread goes through the first {{writeLock}} section checking the file state and releases control somewhere in {{chooseTarget()}}.
- Then the second thread starts working on the same first {{writeLock}} section and will get the same results as the first thread, because the file state hasn't changed.
- Then both threads execute the second {{writeLock}} section in some order, which doesn't matter, and create two different blocks with different ids and targets.
- The client will receive response only from one of the competing threads and will have no idea about the other block. The client proceeds with sending data to DataNodes.
- When the client tries to create yet another block it will receive {{NotReplicatedYetException}}, because the penultimate block in the file has 0 replicas and will never have more.
                
> getAdditionalBlock() can create multiple blocks if the client times out and retries.
> ------------------------------------------------------------------------------------
>
>                 Key: HDFS-4452
>                 URL: https://issues.apache.org/jira/browse/HDFS-4452
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: namenode
>    Affects Versions: 2.0.2-alpha
>            Reporter: Konstantin Shvachko
>            Assignee: Konstantin Shvachko
>            Priority: Critical
>
> HDFS client tries to addBlock() to a file. If NameNode is busy the client can timeout and will reissue the same request again. The two requests will race with each other in {{FSNamesystem.getAdditionalBlock()}}, which can result in creating two new blocks on the NameNode while the client will know of only one of them. This eventually results in {{NotReplicatedYetException}} because the extra block is never reported by any DataNode, which stalls file creation and puts it in invalid state with an empty block in the middle.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira