Chef's RabbitMQ beam.smp process at 100+% CPU on OSX (homebrew) due to excessive queue creation

I'm seeing 100% CPU load from rabbitmq's beam.smp process on OSX the second and all subsequent times I start rabbitmq. RabbitMQ is supporting a local chef server and nothing else. This presented using Chef 10.12.0 and RabbitMQ 2.8.4, but persisted even after a baseline reinstall to chef 0.10.8 and rabbitmq 2.8.2.

td;dr – if beam.smp is going crazy, stuff these lines into your chef config files:

module ::Chef::Expander   ;                    VNODES = 16 ; end
    module ::Chef::IndexQueue ; class AmqpClient ; VNODES = 16 ; end ; end

-------------------------------

Running `rabbitmqctl report` would hang trying to list `Queues on /chef`. Rabbitmq was otherwise responsive to rabbitmqctl commands (in particular, stop and status), and was using ~ 144 MB under no external load; starting and stopping the server would take more than a minute each. Documents continued to be indexed during this time, so the queue was responsive.

• Setup: rabbitmq, couch and erlang installed using homebrew via this chef recipe. erlang version is R15B01 (erts-5.9.1), rabbitMQ is 2.8.4 Nuking it from orbit + re-installing led to the same issue over multiple trials. I also compiled erlang from scratch and reinstalled rabbitmq from source at 2.8.2, no change.
• The rabbitMQ mgmt dashboard reports that I had 3GB high water mark – rabbitMQ was under no real memory pressure. The computer is a mid-2011 Macbook Pro with 8 GB ram.

I then did the following:

• blew away everything – chef, erlang, couch, rabbitmq, and all their installed directories
• installed erlang and rabbitmq, brought up rabbitmq and verified that it was vanilla.
• Make vhost '/chef', user chef, and authed chef and guest on /chef:

rmq_user=chef; rmq_password=testing ; rmq_vhost='/chef'
    rabbitmqctl add_vhost "$rmq_vhost"
    rabbitmqctl add_user "$rmq_user" "$rmq_password"
    rabbitmqctl set_permissions -p "$rmq_vhost" "$rmq_user" ".*" ".*" ".*"
    rabbitmqctl set_permissions -p "$rmq_vhost" "guest"     ".*" ".*" ".*"

• ran chef-expanderctl queue-depth, which then enumerated its 1024 queues – fairly slowly.
• After some seconds, all those queues report as idle, and the rabbitmq process seemed otherwise quiescent.
• Did a rabbitmq stop, which took ~60sec
• Starting rabbitMQ took more than a minute to come fully active, with the CPU at max the whole way. Tens of minutes later, with no activity, the beam.smp process was still at 300% / fan blasting / space heater warp speed.
• no jobs were ever enqueued – only the enumeration of 1024 queues.

The only thing that did help was to turn down the number of vnodes chef server and expander use:

module ::Chef::Expander   ;                    VNODES = 16 ; end
    module ::Chef::IndexQueue ; class AmqpClient ; VNODES = 16 ; end ; end