Please correct if I am wrong, but triggering restart node/agent immediately restarts it, even when builds/jobs are still running. Subsequently the queue/jobs counters are partly reset and (redirected) logs are lost.

In analogy to restart Quickbuild waiting for builds, I would in all cases (i.e. by default) that I can think of prefer the following currently manual necessary steps to be automated:
1. set size of queue(s) of corresponding nodes/agents to 0 (remember the previous used size values)
2. wait for all builds/jobs on this nodes/agents to finish
3. restart nodes/agents
4. re-set size of queue(s) to the remembered previous values

Possible challenge: Is there a mapping between nodes/agents and queue?