[
https://issues.apache.org/jira/browse/HDDS-179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16572124#comment-16572124
]
Lokesh Jain commented on HDDS-179:
----------------------------------
[~shashikant] Thanks for working on this! The patch looks good to me. Please
find my comments below.
#
{code:java}
stateMachineMap.computeIfAbsent(requestProto.getContainerID(),
k -> new StateMachineHelper());
{code}
We can do this operation in startTransaction when we are executing the
CreateContainerCommand. This way we do not have to check it in every
writeStateMachineData and applyTransaction operation.
# We are currently not making sure that while CloseContainer is waiting no
further writeChunks are executed. Maybe we need another boolean field in
StateMachineHelper for that purpose.
# I have another proposal for CloseContainer which is similar to what is
currently done for writeChunk and createContainer. We can add a future for
CloseContainer in writeStateMachine data. We can maintain a counter for each
container which is incremented on start of writeChunk op and is reduced when it
completes. On reduction if the counter reaches 0 the close container
stateMachine future can be completed which would lead to execution of
closeContainer op. This way we do not need a per container writeChunk map.
# ContainerStateMachine:351-355 can be moved to handleCloseContainer.
# ContainerStateMachine:415 - We should also check whether the size of
writeChunkFutureMap >0
# Minor comment. Should we rename StateMachineHelper to ContainerStateHelper
or sth like that to give more clarity?
# ContainerStateMachine:210 redundant change.
# Lets move the VisibleForTesting fns to end of the class.
> CloseContainer command should be executed only if all the prior "Write" type
> container requests get executed
> -------------------------------------------------------------------------------------------------------------
>
> Key: HDDS-179
> URL: https://issues.apache.org/jira/browse/HDDS-179
> Project: Hadoop Distributed Data Store
> Issue Type: Bug
> Components: Ozone Client, Ozone Datanode
> Reporter: Shashikant Banerjee
> Assignee: Shashikant Banerjee
> Priority: Major
> Fix For: 0.2.1
>
> Attachments: HDDS-179.01.patch, HDDS-179.02.patch, HDDS-179.03.patch,
> HDDS-179.04.patch, HDDS-179.05.patch, HDDS-179.06.patch, HDDS-179.07.patch
>
>
> When a close Container command request comes to a Datanode (via SCM hearbeat
> response) through the Ratis protocol, all the prior enqueued "Write" type of
> request like WriteChunk etc should be executed first before CloseContainer
> request gets executed. This synchronization needs to be handled in the
> containerStateMachine. This Jira aims to address this.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
