git.net

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: KVM Max Guests Limit



On 11/9/18 1:33 PM, Andrija Panic wrote:
> Thanks Wido - though I don't seem to be able to find any related setting
> (there is host.overcommit.mem.mb but that is not it - unless you can
> define negative value to it )  ?
> https://github.com/apache/cloudstack/blob/master/agent/conf/agent.properties
> 

host.reserved.mem.mb=32768

That one should be the setting you might want to look at.

In this case 32G is reserved and not available to CS.

Wido

> 
> thx
> 
> On Fri, 9 Nov 2018 at 13:03, Wido den Hollander <wido@xxxxxxxxx
> <mailto:wido@xxxxxxxxx>> wrote:
> 
> 
> 
>     On 11/9/18 12:56 PM, Andrija Panic wrote:
>     > afaik not - but I did run once or twice intom perhaps looselym
>     connected
>     > issue - ACS reports 100% of host RAM (makes sense) asavailable for VM
>     > deployment to ACS - so in 1-2 cases I did run into out of memory
>     killer,
>     > crashing my VMs.
>     >
>     > It would be great to have some amount of "reserve RAM" for host OS
>     - or
>     > simply have PER HOST RAM disableTreshold setting, similar to
>     cluster level
>     > "cluster.memory.allocated.capacity.disablethreshold", just on host
>     level...
>     >
> 
> 
>     You can do that already, in agent.properties you can set reserved
>     memory.
> 
>     But I doubt indeed that we need such a limit in ACS at all, why do we
>     need to limit the amount of Instances on a hypervisor?
> 
>     Or at least set it to a very high number by default.
> 
>     Wido
> 
>     > On Fri, 9 Nov 2018 at 12:03, Rafael Weingärtner
>     <rafaelweingartner@xxxxxxxxx <mailto:rafaelweingartner@xxxxxxxxx>>
>     > wrote:
>     >
>     >> Do we need these logical constraints in ACS at all?
>     >>
>     >> On Fri, Nov 9, 2018 at 6:57 AM Wido den Hollander <wido@xxxxxxxxx
>     <mailto:wido@xxxxxxxxx>> wrote:
>     >>
>     >>>
>     >>>
>     >>> On 11/8/18 11:20 PM, Simon Weller wrote:
>     >>>> I think these is legacy and a guess back in the day. It was 50
>     at one
>     >>> point and it was lifted higher a few releases. ago.
>     >>>>
>     >>>
>     >>> I see. I'm about to do a test with a bunch of 128GB hypervisors and
>     >>> spawning a lot of 128M VMs. Trying to see where the limit might
>     be and
>     >>> also stress the VR a bit by loading a lot of DHCP entries.
>     >>>
>     >>> Wido
>     >>>
>     >>>>
>     >>>>
>     >>>>
>     >>>> ________________________________
>     >>>> From: Ivan Kudryavtsev <kudryavtsev_ia@xxxxxxxxx
>     <mailto:kudryavtsev_ia@xxxxxxxxx>>
>     >>>> Sent: Thursday, November 8, 2018 3:58 PM
>     >>>> To: dev
>     >>>> Subject: Re: KVM Max Guests Limit
>     >>>>
>     >>>> Hi all, +1 for higher numbers.
>     >>>>
>     >>>> чт, 8 нояб. 2018 г. в 16:32, Wido den Hollander <wido@xxxxxxxxx
>     <mailto:wido@xxxxxxxxx>>:
>     >>>>
>     >>>>> Hi,
>     >>>>>
>     >>>>> I see that for KVM we set the limit to 144 guests by default, can
>     >>>>> anybody tell me why we have this limit set to 144?
>     >>>>>
>     >>>>> Searching a bit I found this:
>     >>>>> https://access.redhat.com/articles/rhel-kvm-limits
>     >>>>>
>     >>>>> "This guest limit does not apply to Red Hat Enterprise Linux with
>     >>>>> Unlimited Guests. There is no guest limit for Red Hat Enterprise
>     >>>>> Virtualization"
>     >>>>>
>     >>>>> There is always a limit somewhere, but why do we set it to 144?
>     >>>>>
>     >>>>> I would personally vote for increasing this to 500 or something so
>     >> that
>     >>>>> users don't run into it that easily.
>     >>>>>
>     >>>>> Also, the log line is printed in DEBUG mode only when a host
>     reaches
>     >>>>> this limit, so I created a PR to set this to INFO:
>     >>>>> https://github.com/apache/cloudstack/pull/3013
>     >>>>>
>     >>>>> Any input?
>     >>>>>
>     >>>>> Wido
>     >>>>>
>     >>>>
>     >>>>
>     >>>> --
>     >>>> With best regards, Ivan Kudryavtsev
>     >>>> Bitworks LLC
>     >>>> Cell RU: +7-923-414-1515
>     >>>> Cell USA: +1-201-257-1512
>     >>>> WWW: http://bitworks.software/ <http://bw-sw.com/>
>     >>>>
>     >>>
>     >>
>     >>
>     >> --
>     >> Rafael Weingärtner
>     >>
>     >
>     >
> 
> 
> 
> -- 
> 
> Andrija Panić