When stem calls controller.close_circuit(circuit_id), circuits don't really close

I'm using master (0.2.8-alpha-dev) and a stem script similar to the example custom_path_selection.py, but that calls controller.close_circuit(circuit_id) after using each circuit.

This may be related to the abort in legacy/trac#18116 (moved).

I get a lot of messages similar to this:

[notice] No circuits are opened. Relaxed timeout for circuit 170239 (a Circuit made by controller 3-hop circuit in state doing handshakes with channel state open) to 60000ms. However, it appears the circuit has timed out anyway. 20 guards are live. [59 similar message(s) suppressed in last 3600 seconds]

It may be that the log message is just wrong, or it may be that stem isn't doing the right thing to close the circuit, or it may be that tor isn't really cleaning up well after controller circuits. The abort in legacy/trac#18116 (moved) suggests the latter.