How to fully remove a decommissioned DataNode from ozone admin datanode list?

[Pigueiras]

Hi everyone,

I've successfully decommissioned a DataNode using ozone admin datanode decommission <hostname>. The node's state changes to DECOMMISSIONED, and replication completes as expected. But the node still appears in the output of ozone admin datanode list. Is there a recommended way to fully remove a decommissioned node (as DEAD/DECOMISSIONED)? Or is this expected behavior?

Thanks in advance.

[ivandika3]

Hi @Pigueiras , looking at the implementation, it seems the DEAD and DECOMMISSIONED nodes are stored in SCM indefinitely.

Is there a recommended way to fully remove a decommissioned node (as DEAD/DECOMISSIONED)? Or is this expected behavior?

I think since the number of datanodes should fit well in the SCM memory (even for thousands of datanodes), we currently don't see the urgent need to remove the ones in the terminal states. It is also good to know which nodes are decommissioned or dead to troubleshoot issues (instead of them disappearing).

In the future, we can support a CLI with similar mechanism to Recon's "/datanodes/remove" endpoint to call SCMNodeManager#removeNode when the datanode is DEAD. Feel free to raise a ticket and discuss this further with the community.

[Pigueiras]

Hello, no problem at all, this was mainly for me to understand how it works. If keeping DEAD and DECOMMISSIONED nodes in SCM indefinitely is the expected behavior, that’s totally fine for me. Thanks for the clarification!