[Systems] Wiki speed

[Sam P.]

Hi All,

Samuel and I chatted and tested stuff on irc earlier this morning (well,
last night over in you guys part of the world).  Here's where we are at:

*  Docker can actually do cpu limitation in a good way.  We found 2 ways we
can limit the cpu, both of which work pretty well (aka. we tested them)
    - Dedicate docker X cores and use cpu shares.  Effectively saying "if
all of the containers are using the cpu at the same time, you get (my
share)/(total shares) of the cpu pool".  That could work fine, we'd just
need to agree on which cpu cores to use and keep them the same across all
containers
    - We can set a limit to how much (%) of each cpu scheduler period that
the container can use
=> We kinda have found that we can do everything with docker - no hipster
stuff needed

* Docker has good managament/reporting tools.  We deployed (aka. ran the
container) cAdvisor (by Google), which is a nice web interface to the
cpu/ram per container status
    - http://freedom.sugarlabs.org:8080/containers/
    - eg.  Discourse container
http://freedom.sugarlabs.org:8080/containers/docker/205acea16a1bbc6018b3512bf5d771e0906a997945bcac5605b53963e65aee86
    - Will find a solution to why it shows container ids not names on the
list page (because good sysadmins obviously memorise the container id of
every container they launch)

* We now have data to make decisions about what resources to allocate to
the containers and means to limit them to those resources!  Yay!

Thanks,
Sam

On Sat, Jun 27, 2015 at 10:08 PM Sam P. <sam at sam.today> wrote:

> Hi Bernie,
>
> On Sat, Jun 27, 2015 at 3:56 AM Bernie Innocenti <bernie at codewiz.org>
> wrote:
>
>> On 26/06/15 09:14, Samuel Cantero wrote:
>>
> > I like the approach of using systemd for resource management. Besides,
>> > on an operating system which uses systemd as the service manager process
>> > (not only the ones inside of the container) will be placed in a cgroups
>> > tree. BTW, systemd will replace upstart in Ubuntu 15.04.
>>
>> If we need to upgrade to 15.04 for systemd, I'd go for it even if it's
>> not LTS. We have the ability to experiment on freedom while justice
>> keeps serving production traffic.
>>
>
> Yeah, systemd is very useful.  I'm pretty sure it has systemd, but I'll
> spin up a droplet on DO to check it very soon.  From memory, systemd got
> shipped on most of the distros recently.
>
>
>>
>> When (and *if*) we're satisfied with the stability of systemd containers
>> on 15.04, we'll upgrade justice to the same release. And anyway, as soon
>> as the next LTS comes out we'll upgrade both machines to it, in the same
>> staggered fashion which minimizes surprises.
>>
>> However, systemd-nspawn is relatively new and I haven't heard of any
>> large deployment using it. It may very well be that it's not yet ready
>> for production. So don't count too much on it.
>>
>
> Hum, I wasn't thinking of using systemd-nspawn.  It is pretty cool, but it
> might not be as production ready.
>
> I was thinking of using rkt to run containers.  It is backed by coreos so
> it's probably more production ready.  But, since rkt runs in the same UNIX
> style as systemd-nspawn, you can integrate it with systemd unit/.service
> files [1].  This means that you have a simple management system from
> systemd and also a way of allocating resources (CPUQuota and MemoryLimit)
> and monitoring (try "systemctl status display-manager.service").
>
> I was aiming to test that today, but I'm getting there very slowly.  rkt
> runs appc images although you can convert docker images to appc.  I have no
> idea how you build appc images and can't find the link to a registry or
> anything dockery like that.  Running docker images either involves pulling
> them from a docker registry or converting them (like 10minutes of pure disk
> I/O).  Anyway, that's something to research.
>
>
>>
>>
>> > I will look for rtk and runc.io <http://runc.io>.
>>
>> Thanks. Speaking of container management, have you guys tried Kubernetes?
>>
>>    https://github.com/googlecloudplatform/kubernetes
>>
>> Looks a bit too big for our infrastructure, but it also looks pretty
>> well designed and implemented...
>>
>
> Yeah, it's a cool system.  It doesn't seem to solve out resource limits
> issue - it is built on top of docker I think.
>
> What's your use case of it?
>
> Thanks,
> Sam
>
> [1]
> https://github.com/coreos/rkt/blob/master/Documentation/using-rkt-with-systemd.md
>
>
>>
>> --
>>   _ // Bernie Innocenti
>>   \X/  http://codewiz.org
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.sugarlabs.org/private/systems/attachments/20150628/52b2509e/attachment.html>