[rabbitmq-discuss] rabbitmq 2.8.2 beam process consumes 100% of one cpu in tight epoll_wait loop

Thu Jun 14 22:10:10 BST 2012

Hello;

I am running rabbitmq-server 2.8.2 installed from official RPM on a single
CPU RHEL6 64-bit box. Erlang R14B is installed from EPEL 6 repository.
Immediately upon startup, the beam process consumes 100% of the system's 1
CPU and never consumes less.

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+
COMMAND
 8178 rabbitmq  20   0  725m 148m 2944 R 98.5  3.9   4:11.54 beam

Running strace on the process reveals that beam is in a tight loop, with
the following sequence repeated endlessly:

read(5, "00", 32)                       = 2
read(5, 0x7fffc2213e40, 32)             = -1 EAGAIN (Resource temporarily
unavailable)
epoll_wait(0, {}, 256, 0)               = 0
futex(0x7f93c272f8dc, FUTEX_WAKE_OP_PRIVATE, 1, 1, 0x7f93c272f8d8,
{FUTEX_OP_SET, 0, FUTEX_OP_CMP_GT, 1}) = 1
futex(0x7f93c272f8b0, FUTEX_WAKE_PRIVATE, 1) = 1
futex(0x7f93c272f954, FUTEX_WAKE_OP_PRIVATE, 1, 1, 0x7f93c272f950,
{FUTEX_OP_SET, 0, FUTEX_OP_CMP_GT, 1}) = 1
futex(0x7f93c272f928, FUTEX_WAKE_PRIVATE, 1) = 1
epoll_wait(0, {{EPOLLIN, {u32=5, u64=9332386261106688005}}}, 256, 0) = 1

This occurs even when zero clients are connected to rabbitmq. Do you have
any ideas about this? Is this expected behavior?

Con
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20120614/5c0d5b76/attachment.htm>