pc_notify_with_timeout inside pc_client_poll will never returns.

[xpol]

The stack trace:

#0  0x00007fff8f016de6 in __psynch_mutexwait ()
#1  0x00000000000cecc2 in _pthread_mutex_lock_wait ()
#2  0x00000000000cc294 in _pthread_mutex_lock_slow ()
#3  0x000000010186867a in pc_mutex_lock [inlined] at /Users/xpol/Workspace/lua-packages/packages/pomelo/deps/libpomelo2/src/pc_mutex.h:67
#4  0x0000000101868671 in tr_uv_tcp_send at /Users/xpol/Workspace/lua-packages/packages/pomelo/deps/libpomelo2/src/tr/uv/tr_uv_tcp_i.c:366
#5  0x0000000101862b3a in pc_notify_with_timeout at /Users/xpol/Workspace/lua-packages/packages/pomelo/deps/libpomelo2/src/pc_pomelo.c:716
#6  0x000000010187c627 in Client_notify(lua_State*) at /Users/xpol/Workspace/lua-packages/packages/pomelo/lua-pomelo.cpp:657
#7  0x00000001017e96b8 in lj_BC_FUNCC ()
#8  0x00000001017f9984 in lua_pcall ()
#9  0x000000010187d34d in lua_request_cb(pc_request_s const*, int, char const*) at /Users/xpol/Workspace/lua-packages/packages/pomelo/lua-pomelo.cpp:570
#10 0x00000001018638ac in pc__trans_resp at /Users/xpol/Workspace/lua-packages/packages/pomelo/deps/libpomelo2/src/pc_trans.c:333
#11 0x0000000101861d45 in pc__handle_event [inlined] at /Users/xpol/Workspace/lua-packages/packages/pomelo/deps/libpomelo2/src/pc_pomelo.c:310
#12 0x0000000101861d22 in pc_client_poll at /Users/xpol/Workspace/lua-packages/packages/pomelo/deps/libpomelo2/src/pc_pomelo.c:374
#13 0x000000010187c229 in pomelo_poll() at /Users/xpol/Workspace/lua-packages/packages/pomelo/lua-pomelo.cpp:126
#14 0x000000010187dbc9 in lib_poll(lua_State*) at /Users/xpol/Workspace/lua-packages/packages/pomelo/lua-pomelo.cpp:134
#15 0x00000001017e96b8 in lj_BC_FUNCC ()
#16 0x00000001017f9984 in lua_pcall ()

just like #63.

The polling mode is completely unstable and untested...

[cynron]

pc_client_cleanup will wait the uv loop to terminate, if you call pc_client_cleanup in callbacks, it will deadlock.

[xpol]

it is pc_notify_with_timeout this time, not pc_client_cleanup.

[cynron]

there are two mutexes here, polling_event_queue_mutex and tcp_send_queue_mutex
there are two threads, polling thread, uv loop thread

polling thread:
acquire event queue mutex -> invoke callback -> call pc_notify_with_timeout -> want send queue mutex

uv loop thread:
accquire send queue mutex -> uv tcp send ok -> add send complete event -> want event queue mutex

then, deadlock!

solution:
uv loop thread: after fetch a send item, release the send queue mutex

[cynron]

solution 2:
when polling the event queue, release the event_mutex before pc__handle_event and re-acquire it after pc__handle_event finished.

[xpol]

here is a pr for solution 2: #67.

[cynron]

ok, goto #67 ,closing here