HDDS-2534. scmcli container delete not working

[chimney-lee]

What changes were proposed in this pull request?

fix scmcli container delete not working

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-2534

How was this patch tested?

exec bin/ozone scmcli container delete <containerId>

[xiaoyuyao]

Note this only delete the container from the SCM container DB store. The container report from DN may still report the deleted container replicas. We need to either further send the delete to datanodes upon SCM cli container delete. Or add additional hanlding in ContainerReportHandler#processContainerReplicas to deal with ContainerNotFoundException by sending datanode container delete command via heatbeat.

[chimney-lee]

@xiaoyuyao thanks for comment, should we remove the container in NodeStateMap with NodeStateMap#removeContainer while do SCMClientProtocolServer#deleteContainer. For that, After deal with ContainerNotFoundException in ContainerReportHandler#processContainerReplicas by sending datanode container delete command via heatbeat, then scm handling the first heartbeat from datanode, the NodeStateMap still contains the containerid. this cause the ContainerNotFoundException in ContainerReportHandler#processMissingReplicas, but after the handle of this heartbeat, the NodeStateMap will be updated.

[elek]

What's the plan with this patch? @xiaoyuyao Do you have any suggestion? Do we need to include the logic to delete the container in this patch?

[xiaoyuyao]

Yes, the current patch delete container from SCM DB, container state manager still has the deleted container info in memory. Here are the issues:

1. ContainerReportHandler will not hit ContainerNotFoundException.
2. Container state manager won't get deleted container removed from in-memory map until a restart/reload.
3. In-memory container state inconsistent with the SCM DB.

To solve this, we will need to keep track of the deleting status in SCM until the heartbeat from Dn confirm the deletion cmd execution on three datanodes. This will be similar to what we did for the block deletion log in SCM.

[adoroszlai]

Closing as this is superseded by #389.