multi-accept benchmarks fail on Windows

[kellabyte]

I've been debugging why my multi-accept inspired code in my own code base was failing on Windows but working on Linux and OSX. I decided to run the libuv benchmarks to see if the implementations in libuv were working on Windows and I found some issues.

There may be a race condition for when server_handle is initialized between these sequence of calls:

test_tcp() (new thread) -> server_cb() -> get_listen_handle() -> ipc_connect_cb() -> ipc_read_cb() -> server_handle is initialized

test_tcp() (new thread) -> server_cb() ->  uv_listen(server_handle)

https://github.com/joyent/libuv/blob/master/test/benchmark-multi-accept.c#L188
https://github.com/joyent/libuv/blob/master/test/benchmark-multi-accept.c#L269

uv_shutdown_t: 96 bytes
uv_write_t: 144 bytes
uv_connect_t: 96 bytes
uv_udp_send_t: 96 bytes
uv_tcp_t: 256 bytes
uv_pipe_t: 440 bytes
uv_tty_t: 280 bytes
uv_prepare_t: 88 bytes
uv_check_t: 88 bytes
uv_idle_t: 88 bytes
uv_async_t: 160 bytes
uv_timer_t: 128 bytes
uv_fs_poll_t: 72 bytes
uv_fs_event_t: 208 bytes
uv_process_t: 200 bytes
uv_poll_t: 320 bytes
uv_loop_t: 192 bytes
loop_count: 2000000 ticks in 1.26s (1584149/s)
loop_count: 8234660 ticks (1646932 ticks/s)
ping_pongs: 46039 roundtrips/s
1000000 write requests in 5.12s.
tcp_pump100_server: 3.7 gbit/s
tcp_pump100_client: 3.7 gbit/s
tcp_pump1_server: 3.6 gbit/s
tcp_pump1_client: 3.6 gbit/s
tcp-conn-pound-100: 13891 accepts/s (0 failed)
tcp-conn-pound-1000: 12243 accepts/s (0 failed)

`pipe_pump100_client` failed: exit code 3
Output from process `pipe_pump_server`:
Assertion failed in test\benchmark-pump.c on lOutput from process `pipe_pump100_client`:
unknown errorAssertion failed in test\benchmark-pump.c on line 217: status == 0
=============================================================
pipe_pump1_server: 11.1 gbit/s
pipe_pump1_client: 11.1 gbit/s
pipe-conn-pound-100: 99712 accepts/s (481968 failed)
pipe-conn-pound-1000: 101400 accepts/s (518680 failed)

`tcp_multi_accept2` failed: exit code 3
Output from process `tcp_multi_accept2`:
Assertion failed in test\benchmark-multi-accept.c on line 126: 0 == uv_accept(ipc_pipe, (uv_stream_t*) &pc->peer_handle)
=============================================================

`tcp_multi_accept4` failed: exit code 3
Output from process `tcp_multi_accept4`:
Assertion failed in test\benchmark-multi-accept.c on line 126: 0 == uv_accept(ipc_pipe, (uv_stream_t*) &pc->peer_handle)
=============================================================

`tcp_multi_accept8` failed: exit code 3
Output from process `tcp_multi_accept8`:
Assertion failed in test\benchmark-multi-accept.c on line 126: 0 == uv_accept(ipc_pipe, (uv_stream_t*) &pc->peer_handle)
Asserti=============================================================
udp_pummel_1v1: 122100/s received, 122100/s sent. 999997 received, 1000000 sent in 8.2 seconds.
udp_pummel_10v1: 120091/s received, 120091/s sent. 999997 received, 1000000 sent in 8.3 seconds.
udp_pummel_100v1: 119904/s received, 119904/s sent. 999998 received, 1000000 sent in 8.3 seconds.
udp_pummel_1000v1: 118315/s received, 118315/s sent. 999997 received, 1000000 sent in 8.5 seconds.
udp_pummel_10v10: 156764/s received, 156764/s sent. 1000000 received, 1000000 sent in 6.4 seconds.
udp_pummel_100v10: 159719/s received, 159719/s sent. 1000000 received, 1000000 sent in 6.3 seconds.
udp_pummel_1000v10: 158403/s received, 158403/s sent. 1000000 received, 1000000 sent in 6.3 seconds.
udp_pummel_100v100: 154154/s received, 154154/s sent. 1000000 received, 1000000 sent in 6.5 seconds.
udp_pummel_1000v100: 160282/s received, 160282/s sent. 1000000 received, 1000000 sent in 6.2 seconds.
udp_pummel_1000v1000: 128741/s received, 129182/s sent. 996584 received, 1000000 sent in 7.7 seconds.
udp_pummel_1v1: 119722/s received, 119723/s sent. 598130 received, 598135 sent in 5.0 seconds.
udp_pummel_10v1: 121638/s received, 121639/s sent. 608798 received, 608801 sent in 5.0 seconds.
udp_pummel_100v1: 119630/s received, 119631/s sent. 598391 received, 598395 sent in 5.0 seconds.
udp_pummel_1000v1: 119020/s received, 119021/s sent. 592365 received, 592369 sent in 5.0 seconds.
udp_pummel_10v10: 160516/s received, 160518/s sent. 802580 received, 802590 sent in 5.0 seconds.
udp_pummel_100v10: 160724/s received, 160726/s sent. 803780 received, 803790 sent in 5.0 seconds.
udp_pummel_1000v10: 157928/s received, 157930/s sent. 785690 received, 785700 sent in 5.0 seconds.
udp_pummel_100v100: 164163/s received, 164183/s sent. 819500 received, 819600 sent in 5.0 seconds.
udp_pummel_1000v100: 161384/s received, 161404/s sent. 802400 received, 802500 sent in 5.0 seconds.
udp_pummel_1000v1000: 126536/s received, 127431/s sent. 624584 received, 629000 sent in 4.9 seconds.
getaddrinfo: 17794 req/s

`fs_stat` failed: timeout
Output from process `fs_stat`:
1,000,000 stats (sync): 20.13s (49,672/s)
100,000 stats (1 concurrent): 3.04s (32,917/s)
100,000 stats (2 concurrent): 1.71s (58,573/s)
100,000 stats (3 concurrent): 1.32s (75,493/s)
100,000 stats (4 concurrent): 1.26s (79,329/s)
100,000 stats (5 concurrent): 1.29s (77,672/s)
100,000 stats (6 concurrent): 1.27s (79,031/s)
100,000 stats (7 concurrent): 1.27s (79,018/s)
100,000 stats (8 concurrent): 1.27s (78,794/s)
100,000 stats (9 concurrent): 1.27s (78,540/s)
100,000 stats (10 concurrent): 1.26s (79,191/s)
100,000 stats (11 concurrent): 1.26s (79,490/s)
100,000 stats (12 concurrent): 1.25s (79,745/s)
100,000 stats (13 concurrent): 1.25s (79,728/s)
100,000 stats (14 concurrent): 1.25s (79,801/s)
100,000 stats (15 concurrent): 1.26s (79,615/s)
100,000 stats (16 concurrent): 1.25s (79,843/s)
100,000 stats (17 concurrent): 1.25s (79,966/s)
100,000 stats (18 concurrent): 1.25s (80,031/s)
100,000 stats (19 concurrent): 1.26s (79,071/s)
<buffer too small> stats (20 concurrent): 1.24s (80,428/s)
<buffer too small> stats (21 concurrent): 1.24s (<buffer too small>/s)
<buffer too small> stats (22 concurrent): 1.24s (<buffer too small>/s)
<buffer too small> stats (23 concurrent): 1.25s (<buffer too small>/s)
<buffer too small> stats (24 concurrent): 1.25s (<buffer too small>/s)
<buffer too small> stats (25 concurrent): 1.24s (<buffer too small>/s)
<buffer too small> stats (26 concurrent): 1.25s (<buffer too small>/s)
<buffer too small> stats (27 concurrent): 1.25s (<buffer too small>/s)
<buffer too small> stats (28 concurrent): 1.24s (<buffer too small>/s)
<buffer too small> stats (29 concurrent): 1.24s (<buffer too small>/s)
=============================================================

`async1` failed: exit code 3
Output from process `async1`:
A s s e r t i o n   f a i l e d :   ! ( h a n d l e - > f l a g s   &   U V _ _ H A N D L E _ C L O S I N G ) ,   f i l e   s r c
\ w i n \ a s y n c . c ,   l i n e   7 7
 =============================================================

`async2` failed: exit code 3
Output from process `async2`:
A s s e r t i o n   f a i l e d :   ! ( h a n d l e - > f l a g s   &   U V _ _ H A N D L E _ C L O S I N G ) ,   f i l e   s r c
\ w i n \ a s y n c . c ,   l i n e   7 7
 =============================================================

`async4` failed: exit code 3
Output from process `async4`:
A s s e r t i o n   f a i l e d :   ! ( h a n d l e - > f l a g s   &   U V _ _ H A N D L E _ C L O S I N G ) ,   f i l e   s r c
\ w i n \ a s y n c . c ,   l i n e   7 7
 =============================================================

`async8` failed: exit code 3
Output from process `async8`:
A s s e r t i o n   f a i l e d :   ! ( h a n d l e - > f l a g s   &   U V _ _ H A N D L E _ C L O S I N G ) ,   f i l e   s r c
\ w i n \ a s y n c . c ,   l i n e   7 7
 A s s e r t i =============================================================
async_pummel_1: 1,000,000 callbacks in 0.91 seconds (1,104,847/sec)
async_pummel_2: 1,000,000 callbacks in 1.06 seconds (945,629/sec)
async_pummel_4: 1,000,000 callbacks in 1.15 seconds (867,877/sec)

`async_pummel_8` failed: timeout
Output from process `async_pummel_8`: (no output)
=============================================================

`spawn` failed: timeout
Output from process `spawn`: (no output)
=============================================================

`thread_create` failed: timeout
Output from process `thread_create`: (no output)
=============================================================
3,962,526 async events in 5.0 seconds (792,505/s, 1,048,576 unique handles seen)
30.00 seconds total
21.95 seconds init
6.53 seconds dispatch
1.53 seconds cleanup

[kellabyte]

I'm not so sure if the race condition I identified above is actually the problem after some testing today, so don't take that as gospel.