[
https://issues.apache.org/jira/browse/HDDS-15052?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tsz-wo Sze updated HDDS-15052:
------------------------------
Description:
HDDS-14040 changed ContainerStateMachine.writeStateMachineData(..) from
throwing an exception to returning a ContainerCommandResponseProto. As a
result, if there is an exception such as disk-out-of-space, the underlying
problem is ignored. Also, failed transactions can be returned success to
clients.
HDDS-12151 is a related JIRA having a similar change, although it may or may
not cause data loss.
The correct fix is to change applyTransaction(..) to detect if
writeStateMachineData(..) has succeeded. If writeStateMachineData(..) has
failed, applyTransaction(..) should
# stop committing it to rocksdb,
# complete the future exceptionally such that the exception will be returned
to client, and
# shut down the server/pipeline if the failure cannot be recovered.
was:
HDDS-14040 changed ContainerStateMachine.writeStateMachineData(..) from
throwing an exception to returning a ContainerCommandResponseProto. As a
result, if there is an exception such as disk-out-of-space, the underlying
problem is ignored. This could cause silent data loss since failed transactions
can be returned success to clients.
HDDS-12151 is a related JIRA having a similar change, although it may or may
not cause data loss.
The correct fix is to change applyTransaction(..) to detect if
writeStateMachineData(..) has succeeded. If writeStateMachineData(..) has
failed, applyTransaction(..) should
# stop committing it to rocksdb,
# complete the future exceptionally such that the exception will be returned
to client, and
# shut down the server/pipeline if the failure cannot be recovered.
Summary: HDDS-14040 may break majority commit contract (was:
HDDS-14040 may cause data loss)
> HDDS-14040 may break majority commit contract
> ---------------------------------------------
>
> Key: HDDS-15052
> URL: https://issues.apache.org/jira/browse/HDDS-15052
> Project: Apache Ozone
> Issue Type: Bug
> Components: Ozone Datanode
> Reporter: Tsz-wo Sze
> Assignee: Sumit Agrawal
> Priority: Major
>
> HDDS-14040 changed ContainerStateMachine.writeStateMachineData(..) from
> throwing an exception to returning a ContainerCommandResponseProto. As a
> result, if there is an exception such as disk-out-of-space, the underlying
> problem is ignored. Also, failed transactions can be returned success to
> clients.
> HDDS-12151 is a related JIRA having a similar change, although it may or may
> not cause data loss.
> The correct fix is to change applyTransaction(..) to detect if
> writeStateMachineData(..) has succeeded. If writeStateMachineData(..) has
> failed, applyTransaction(..) should
> # stop committing it to rocksdb,
> # complete the future exceptionally such that the exception will be returned
> to client, and
> # shut down the server/pipeline if the failure cannot be recovered.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
