task-spooler (tsp) shows an empty queue when the queue was created with a custom TS_SOCKET
Symptom: on a worker box, tsp or tsp -l printed only the header line, even though jobs were clearly running there.
Cause: the job wrapper enqueued with TS_SOCKET=/data/tsp/socket tsp -L <label> .... task-spooler keeps one queue per socket. Without that variable it uses the default per-user socket under /tmp, so a plain tsp -l starts a fresh, empty server instead of showing the real queue.
Fix: pass the same socket to every tsp call, e.g. TS_SOCKET=/data/tsp/socket tsp -l, and the same for -r <id> (remove a queued job) and -k <id>. Find the path by grepping the submit script for TS_SOCKET.
Also: if the job itself runs inside systemd-run --scope (for a memory cap), tsp -k only signals the wrapper shell. The scope can keep running, so stop the scope unit itself.
Confirmed again: an in-job 'no other running jobs' check read the default socket, so it reported 0 even though it couldn't see anything, its own job included. A cheap guard: require the listing to show the job's own label as running before trusting a count of 0; if it doesn't, the check proved nothing, so refuse.