Server closing channel hangs client in recv

[MattiL]

Server closing the channel hangs the client in wait_frame_inner / recv:
http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/2012-April/019361.html
http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/2012-April/019572.html

It seems to me that the client does not recognize the server sending an error message about the publish exchange not existing and therefore can get stuck on a channel that does not get any more messages.

I have encountered a possible fault in librabbitmq (rabbitmq-c-fb6fca832fd2). I am trying to use that latest version customized with Perl Net::RabbitMQ on SLES 11 SP 1.

amqp_simple_wait_frame and recv do not respond to messages being published to an exchange after rabbitmqctl stop_app, reset and start_app. That problem appears also with Net::RabbitMQ versions 0.2.2 and 0.2.0.

I got the following state in gdb with recv:

0x00007f38e70b7935 in recv () from /lib64/libpthread.so.0
(gdb) bt
#0 0x00007f38e70b7935 in recv () from /lib64/libpthread.so.0
#1 0x00007f38e5cfba4d in recv (__flags=,

__n=<value optimized out>, __buf=<value optimized out>, 
__fd=<value optimized out>) at /usr/include/bits/socket2.h:45

#2 wait_frame_inner (__flags=,

__n=<value optimized out>, __buf=<value optimized out>, 
__fd=<value optimized out>) at amqp_socket.c:182

#3 0x00007f38e5ce78b9 in internal_recv (RETVAL=0xbf4f68, conn=0xbe1440,

piggyback=0) at RabbitMQ.xs:77

#4 0x00007f38e5ce8566 in XS_Net__RabbitMQ_recv (

my_perl=<value optimized out>, cv=<value optimized out>) at RabbitMQ.xs:424

#5 0x000000000047e115 in Perl_pp_entersub ()
#6 0x0000000000455ad3 in Perl_runops_debug ()
#7 0x000000000047a1af in perl_run ()
#8 0x000000000042172c in main ()

(gdb) continue
Continuing.
^C
Program received signal SIGINT, Interrupt.
0x00007f38e70b7935 in recv () from /lib64/libpthread.so.0
(gdb) bt
#0 0x00007f38e70b7935 in recv () from /lib64/libpthread.so.0
#1 0x00007f38e5cfba4d in recv (__flags=,

__n=<value optimized out>, __buf=<value optimized out>, 
__fd=<value optimized out>) at /usr/include/bits/socket2.h:45

#2 wait_frame_inner (__flags=,

__n=<value optimized out>, __buf=<value optimized out>, 
__fd=<value optimized out>) at amqp_socket.c:182

#3 0x00007f38e5ce78b9 in internal_recv (RETVAL=0xbf4f68, conn=0xbe1440,

piggyback=0) at RabbitMQ.xs:77

#4 0x00007f38e5ce8566 in XS_Net__RabbitMQ_recv (

my_perl=<value optimized out>, cv=<value optimized out>) at RabbitMQ.xs:424

#5 0x000000000047e115 in Perl_pp_entersub ()
#6 0x0000000000455ad3 in Perl_runops_debug ()
#7 0x000000000047a1af in perl_run ()
#8 0x000000000042172c in main ()

(gdb) up
#1 0x00007f38e5cfba4d in recv (__flags=,

__n=<value optimized out>, __buf=<value optimized out>, 
__fd=<value optimized out>) at /usr/include/bits/socket2.h:45

45 return __recv_alias (__fd, __buf, __n, __flags);
(gdb) up
#2 wait_frame_inner (__flags=,

__n=<value optimized out>, __buf=<value optimized out>, 
__fd=<value optimized out>) at amqp_socket.c:182

182 res = recv(state->sockfd, state->sock_inbound_buffer.bytes,
(gdb) up
#3 0x00007f38e5ce78b9 in internal_recv (RETVAL=0xbf4f68, conn=0xbe1440,

piggyback=0) at RabbitMQ.xs:77

77 result = amqp_simple_wait_frame(conn, &frame);
(gdb) print conn
$1 = (amqp_connection_state_t) 0xbe1440
(gdb) print *conn
$2 = {frame_pool = {pagesize = 131072, pages = {num_blocks = 5,
blocklist = 0xc5d4f0}, large_blocks = {num_blocks = 0, blocklist = 0x0},
next_page = 0, alloc_block = 0x0, alloc_used = 0}, decoding_pool = {
pagesize = 131072, pages = {num_blocks = 1, blocklist = 0xbf22e0},
large_blocks = {num_blocks = 0, blocklist = 0x0}, next_page = 0,
alloc_block = 0x0, alloc_used = 0}, state = CONNECTION_STATE_IDLE,
channel_max = 0, frame_max = 131072, heartbeat = 0, inbound_buffer = {
len = 131072, bytes = 0x0}, inbound_offset = 0, target_size = 7,
outbound_buffer = {len = 131072, bytes = 0xc67f90}, sockfd = 3,
sock_inbound_buffer = {len = 131072, bytes = 0x7f38e7b57010},
sock_inbound_offset = 79, sock_inbound_limit = 79, first_queued_frame = 0x0,
last_queued_frame = 0x0, most_recent_api_result = {
reply_type = AMQP_RESPONSE_NORMAL, reply = {id = 3932181,
decoded = 0x7f38e7af1038}, library_error = 0}}

[MattiL]

I have added a callback to rabbitmq-c for channel close messages in https://github.com/MattiL/Net--RabbitMQ
That way a client can handle channel close messages from the server and not get stuck in recv system call. I would like you to implement a similar feature.

[alanxz]

I've migrated this issue over to alanxz#15. Please continue discussion there.