We have seen more frequent issues with new workspaces being created with @number even for agents with no parallelisation. When digging into the issue we find that even when an agent is idle there are several workspaces that are allocated/locked.

A simple script like:
println it.toComputer()?.workspaceList?.inUse

lists several builds that are no longer running, connected to github branches that no longer exists.

With several locked workspaces the disk runs out on our agents and results in failing builds.

I have found no way to get rid of the allocated workspaces. Disconnecting the agent and removing the workspace on agent disk does not help.

I haven't yet tried restarting the Jenkins master, this needs to wait until a planned maintenance window.