hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Shashikant Banerjee (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDDS-935) Avoid creating an already created container on a datanode in case of disk removal followed by datanode restart
Date Tue, 26 Feb 2019 13:46:00 GMT

    [ https://issues.apache.org/jira/browse/HDDS-935?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16777923#comment-16777923
] 

Shashikant Banerjee commented on HDDS-935:
------------------------------------------

Thanks [~arpitagarwal] for the review. Patch v6 addresses your review comments.
{code:java}
Why do we add the container set to DispatcherContext? We just need to update this set once
the container is successfully created right?
{code}
The Dispatcher gets passed two set of 2 infos : ContainercommandRequestProto which is client
visible plus DispatcherContext which is ratis StateMachine specific info to execute the ContainercommandRequest
properly. The createContainerSet is maintained per ContainerStateMachine (per pipeline) into
the snapshot file and since its specific to StatMachine, we add the containerSet to DispatcherContext
while executing createContainer on the Dispatcher.

> Avoid creating an already created container on a datanode in case of disk removal followed
by datanode restart
> --------------------------------------------------------------------------------------------------------------
>
>                 Key: HDDS-935
>                 URL: https://issues.apache.org/jira/browse/HDDS-935
>             Project: Hadoop Distributed Data Store
>          Issue Type: Improvement
>          Components: Ozone Datanode
>    Affects Versions: 0.4.0
>            Reporter: Rakesh R
>            Assignee: Shashikant Banerjee
>            Priority: Major
>         Attachments: HDDS-935.000.patch, HDDS-935.001.patch, HDDS-935.002.patch, HDDS-935.003.patch,
HDDS-935.004.patch, HDDS-935.005.patch, HDDS-935.006.patch
>
>
> Currently, a container gets created when a writeChunk request comes to HddsDispatcher
and if the container does not exist already. In case a disk on which a container exists gets
removed and datanode restarts and now, if a writeChunkRequest comes , it might end up creating
the same container again with an updated BCSID as it won't detect the disk is removed. This
won't be detected by SCM as well as it will have the latest BCSID. This Jira aims to address
this issue.
> The proposed fix would be to persist the all the containerIds existing in the containerSet
when a ratis snapshot is taken in the snapshot file. If the disk is removed and dn gets restarted,
the container set will be rebuild after scanning all the available disks and the the container
list stored in the snapshot file will give all the containers created in the datanode. The
diff between these two will give the exact list of containers which were created but were
not detected after the restart. Any writeChunk request now should validate the container Id
from the list of missing containers. Also, we need to ensure container creation does not happen
as part of applyTransaction of writeChunk request in Ratis.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

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


Mime
View raw message