hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Konstantin Shvachko (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HDFS-644) Lease recovery, concurrency support.
Date Sat, 26 Sep 2009 01:57:16 GMT

     [ https://issues.apache.org/jira/browse/HDFS-644?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel

Konstantin Shvachko updated HDFS-644:

    Attachment: leaseRecovery.patch

Fixed JavaDoc, and import warnings. 
Also Hairong correctly noted that internalReleaseLease() does not need to throw RecoveryInProgressException
when the hard limit for the lease expires. The exception now is thrown only when the soft
limit expires to let the client know that file can be re-opened as soon as the lease/block
recovery completes.

> Lease recovery, concurrency support.
> ------------------------------------
>                 Key: HDFS-644
>                 URL: https://issues.apache.org/jira/browse/HDFS-644
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>          Components: data-node, hdfs client, name-node
>    Affects Versions: Append Branch
>            Reporter: Konstantin Shvachko
>            Assignee: Konstantin Shvachko
>             Fix For: Append Branch
>         Attachments: leaseRecovery.patch, leaseRecovery.patch
> Lease recovery happens if hdfs client does not renew its leases within an hour. Lease
recovery triggers block recovery if the block has not been completed during normal write or
append operations. Name-node initiates such block recovery by sending BlockRecoveryCommand
to data-nodes (via reply to a heartbeat). There could be a race condition if some data-nodes
still recover the block when the name-node decides to start a new recovery because the previous
one did not complete. This issue should take of the race condition by introducing recovery
ids. It should also implement the new recovery protocol, which is described in the append
design document,

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message