[PROTON-1310] proton-j reactor transport head never closes

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: proton-0.15.0, proton-0.16.0
• Fix Version/s: proton-j-future
• Component/s: proton-j
• Assignee: Unassigned
• Reporter: Daniel Evans
• Created: 30/Sep/16 17:06
• Updated: 19/Jul/17 21:23

Description

With a server (listener) test class and a client test class, I'm able to disconnect the client such that occasionally on the server the TRANSPORT_ERROR and TRANSPORT_TAIL_CLOSED events fire but TRANSPORT_HEAD_CLOSED and TRANSPORT_CLOSED do not. A timer selectable running to keep the reactor alive then causes repeated TRANSPORT events that all report -1 pending and isClosed == true. This goes on forever.

In the good case, the global event logger sequence is:

TRANSPORT_ERROR
TRANSPORT_TAIL_CLOSED
SELECTABLE_UPDATED
REACTOR_QUIESCED
TRANSPORT_HEAD_CLOSED
TRANSPORT_CLOSED
SELECTABLE_FINAL
CONNECTION_UNBOUND
TRANSPORT

And in the bad case:

TRANSPORT_ERROR
TRANSPORT_TAIL_CLOSED
SELECTABLE_UPDATED
REACTOR_QUIESCED
SELECTABLE_FINAL
REACTOR_QUIESCED
SELECTABLE_UPDATED
TRANSPORT
REACTOR_QUIESCED
SELECTABLE_UPDATED
TRANSPORT
... etc.

Digging around with the debugger I verified that no selectables were writable in the IOHandler and TransportImpl's pop() was never being called when the output pending was -1.