Parallelize node operations

Who is this for and what problem do they have today?

Today, node operations are queued and executed serially. For example, if a node is decomissioned and another node is added to the cluster, the decommission and drain operation on the first node will complete prior to the second node starting to accept partitions.

For Redpanda users running production clusters, it would be nice to parallelize these operations, as node movement can consume significant time and resources in clusters with large amounts of data throughput and storage.

What are the success criteria?

Node operations can be executed in parallel. From the above example, if a node is decommissioned and a second node is added to the cluster, the decommissioned node will drain partitions to other nodes as usual, but now the new node will start accepting partitions in parallel to the draining operation on the first node rather than waiting for that operation to finish.

Why is solving this problem impactful?

This is impactful for users running large production clusters with significant amounts of stored data (where rebalancing and other node operations can take a significant amount of time and resources).

Additional notes

JIRA Link: CORE-871

No files, tests, or entry points are named. Start by locating the node-operation queue and the decommission, drain, and node-addition paths, then use CORE-871 and the success criteria to define parallel execution. Done means a draining node and a newly added node can move partitions concurrently without waiting for one another.