why does rabbitmq eat 25GB of RAM?

asked 2017-04-11 20:31:57 -0500

cfbatncgr gravatar image

Have installed ocata using packstack --allinone. The beam.smp process running as rabbitmq eats up 25GB of RAM on a 1TB RAM system, with nothing configured except for defaults. /bin/nproc reports 128. I have another question concerning the ocata install adding thousands of processes to my otherwise unused system. Maybe these are related?

edit retag flag offensive close merge delete

Comments

I wanted to make a joke about Easter bunnies but let's stay professional. Guess: The beam process might scale automatically with memory size, or many messages are exchanged.

What are those thousands of processes named? On my 5GB packstack, about 200 processes are running.

Bernd Bausch gravatar imageBernd Bausch ( 2017-04-12 18:50:57 -0500 )edit

Thanks. Here's the main culprits, before any attempt to change conf files. There's some name truncation. 654 httpd 259 neutron-server 257 nova-api 129 swift-proxy-ser 129 nova-conductor 129 neutron-metadat 129 glance-registry

cfbatncgr gravatar imagecfbatncgr ( 2017-04-13 18:36:00 -0500 )edit

654 httpd!?!? That's helpful to look at...

cfbatncgr gravatar imagecfbatncgr ( 2017-04-13 18:36:39 -0500 )edit

I've this problem too with Ocata. An all in one installation with a 32Gb Ram server but mysql eat 15Gb and rabbitmq grow up to 15Gb, so the system stuck and we need to reboot to release the server. Do we have some tips to limit rabbitmq ram?

Eric Rakotonirina gravatar imageEric Rakotonirina ( 2017-04-14 00:41:57 -0500 )edit

I have 5GB, 1 CPU and 11 httpd processes. /etc/httpd/conf.d/15-horizon_vhost.conf:

WSGIDaemonProcess apache group=apache processes=3 threads=10 user=apache

Plus a few other WSGI processes for Keystone, Ceilometer etc. On a multi-CPU 1TB machine it's not surprising to see much more than that.

Bernd Bausch gravatar imageBernd Bausch ( 2017-04-14 03:59:24 -0500 )edit