[Koha-bugs] [Bug 22417] Add a task queue

[bugzilla-daemon at bugs.koha-community.org]

https://bugs.koha-community.org/bugzilla3/show_bug.cgi?id=22417

--- Comment #327 from David Cook <dcook at prosentient.com.au> ---
I just remembered that I have a RabbitMQ on an openSUSE Leap 15.1 server. It's
running RabbitMQ 3.7.14 with Erlang 20.3.8.15 and Erts 9.3.3.6 with 12 CPUs.

I'm not using +sbwt on that system, but my constant CPU usage is 0%.

The 1_scheduler thread is still doing a constant stream of syscalls
(epoll_pwait, clock_gettime, futex), but I don't see it reflected in CPU usage
via top. The "aux" thread is also doing something a little bit but rarely.

Looking back again at Ubuntu 18.04 with RabbitMQ 3.6.10 with Erlang 20.2.2 and
Erts 9.2 with 8 CPUs, I can see that all the scheduler threads are spinning a
bit.

I'm curious if it's due to Erlang/Erts/RabbitMQ differences, or configuration
differences. I'm going to study the configuration options a bit.

--

As for Docker...

I'm running rabbitmq:3.6.10 and CPU usage is typically 0% and occasionally
spikes to 2% on a 2 CPU system. It's not as frequent as the Ubuntu system but
more frequent than the openSUSE system. 

I'd say on openSUSE it stays at 0% for 10 seconds before moving up to maybe
0.333% for .5 seconds. 

On Ubuntu 18.04, it sits at 0% but every 1-2 seconds it'll jump up to 2%. 

On Docker rabbitmq:3.6.10, it looks like it goes up to 2% every 2-10 seconds. 

--

Trying Docker rabbitmq:3.8.9...  it goes to 2% CPU every every 2-5 seconds it
seems. Like openSUSE 3.7.14, it seems to be 1_scheduler thread that is doing
the majority of the spinning. 

When I use strace for it, I do see a huge number of sched_yield syscalls and a
much smaller number of futex syscalls.

I'm going to try using "+sbwt none" on another container and see how it goes.
(I'm also capturing the CLI conf to see if there's anything obvious there.)

-- 
You are receiving this mail because:
You are watching all bug changes.