Dave,

While migration xapi reserves some memory doing calls to squeezed.
Moreover migration process will fail with Failure("The ballooning
daemon is not running")  if squeezed is stopped. I think problem is
more complex than you've desribed :)

Cheers!

On Tue, Oct 23, 2012 at 5:24 PM, Dave Scott <[email protected]> wrote:
>> > In case it's useful: the most recent versions of xapi (found in
>> XenServer 6.1 and should be in XCP 1.6) can run without squeezed.
>>
>> If squeezed is stopped such functions like migration (and maybe
>> others) can be broken, don't they?
>
> The operations will work as long as there is already enough memory. You're 
> right though, if there isn't enough memory, then stopping squeezed isn't a 
> great idea :-)
>
>>
>> Best wishes,
>> Kakadu
>>
>> On Tue, Oct 23, 2012 at 5:15 PM, Dave Scott <[email protected]>
>> wrote:
>> > Hi,
>> >
>> > Dave asked:
>> >> > When you try to migrate to a host which hasn't got much free
>> memory,
>> >> do you intend to run the existing squeezing logic?
>> >
>> > Kakadu replied:
>> >> Nope, it should do nothing. In future another balooning daemon will
>> be
>> >> running and case there is not enough ememory never gonna happen. But
>> >> now I'm trying to put right variables to vm params (to emulate
>> >> presence of this future daemon).
>> >
>> > In case it's useful: the most recent versions of xapi (found in
>> XenServer 6.1 and should be in XCP 1.6) can run without squeezed. So
>> you can
>> >
>> > service squeezed stop
>> >
>> > and then when you try to start a VM, there won't be any squeezing at
>> all. Your new daemon could do whatever it likes to manage the VM
>> balloon targets independently of xapi.
>> >
>> > Does that help at all?
>> >
>> > Cheers,
>> > Dave
>> >
>> >> Thanks for the link. I'll try to find something useful.
>> >>
>> >> Best wishes,
>> >> Kakadu
>> >>
>> >>
>> >> On Fri, Oct 19, 2012 at 7:27 PM, Dave Scott
>> <[email protected]>
>> >> wrote:
>> >> >
>> >> > Hi,
>> >> >
>> >> > Kakadu wrote:
>> >> >
>> >> >> Hey.
>> >> >>
>> >> >> I have accepted a challenge about disabling some features of
>> >> >> automatical rebalancing. For example, I want squeezed to raise an
>> >> >> exception while migration in case if there is not enough free
>> memory
>> >> >> and some squeezing needed. So I fixed reserve-memory handler in
>> my
>> >> >> squeezed-like daemon [1] and implemented successful result only
>> when
>> >> >> we can get needed memory without squeezing [2].
>> >> >>
>> >> >> My big problem is testing that stuff. I have vm with these
>> >> >> characteristics
>> >> >> $ xe vm-param-list uuid=ec1dec56-ae4f-d854-8065-e106a1139361 |
>> grep
>> >> >> memory
>> >> >>                  memory-actual ( RO): 250609664
>> >> >>                  memory-target ( RO): 250609664
>> >> >>                memory-overhead ( RO): 4194304
>> >> >>              memory-static-max ( RW): 250000000
>> >> >>             memory-dynamic-max ( RW): 250000000
>> >> >>             memory-dynamic-min ( RW): 250000000
>> >> >>              memory-static-min ( RW): 250000000
>> >> >>
>> >> >> and while localhost migration Squeezer says that it can't
>> reserve
>> >> >> memory without squeezing.
>> >> >>
>> >> >> [20121019T14:36:28.830Z|debug|xenhost-1|0|reserve_memory(xapi,
>> >> >> 256428)|xenstore-rpc] using new version of reserve_memory
>> >> >> [20121019T14:36:28.830Z|debug|xenhost-1|0|reserve_memory(xapi,
>> >> >> 256428)|xenops] force_change_host_free_memory required_mem =
>> 265644
>> >> >> KiB
>> >> >> [20121019T14:36:28.831Z|debug|xenhost-1|0|reserve_memory(xapi,
>> >> >> 256428)|xenops] Total non-domain reservations = 811008
>> >> >> [20121019T14:36:28.831Z|debug|xenhost-1|0|reserve_memory(xapi,
>> >> >> 256428)|xenops] domid 3 has been declared inactive
>> >> >> [20121019T14:36:28.831Z|debug|xenhost-1|0|reserve_memory(xapi,
>> >> >> 256428)|xenops] total_range = 0 gamma = 1.000000 gamma' = inf
>> >> >> [20121019T14:36:28.831Z|debug|xenhost-1|0|reserve_memory(xapi,
>> >> >> 256428)|xenops] Total additional memory over dynamic_min =
>> 2675808
>> >> >> KiB; will set gamma = 1.00 (leaving unallocated 0 KiB)
>> >> >> [20121019T14:36:28.831Z|debug|xenhost-1|0|reserve_memory(xapi,
>> >> >> 256428)|xenops] Maximum possible free memory if all active
>> domains
>> >> >> balloon down to dynamic_min = 2941452
>> >> >> [20121019T14:36:28.831Z|debug|xenhost-1|0|reserve_memory(xapi,
>> >> >> 256428)|xenops] current host free mem = 2941452 KiB (aiming for
>> >> 265644
>> >> >> KiB); OK; all domain targets not reached; allocation phase
>> >> >> [20121019T14:36:28.831Z|debug|xenhost-1|0|reserve_memory(xapi,
>> >> >> 256428)|xenops] Can't free 265644 KiB: some domains need
>> squeezing
>> >> >>
>> >> >> I'm a little bit confused why first number is less than second in
>> `
>> >> >> mem = 2941452 KiB (aiming for 265644 KiB)`. Maybe I dont
>> understand
>> >> >> something good enough
>> >> >
>> >> > I think what it's trying to say is:
>> >> >
>> >> > The host has 2.9 GiB free but we'd like to give most of this free
>> >> memory away to domains (by 'balancing'), to leave only 265644 KiB
>> left.
>> >> >
>> >> >> So I have two questions:
>> >> >> 1) How to setup parameters for testing migration with my new
>> daemon?
>> >> >> 2) What else pifalls I will have in my way while disabling
>> >> >> autobalancing?
>> >> >
>> >> > When you try to migrate to a host which hasn't got much free
>> memory,
>> >> do you intend to run the existing squeezing logic?
>> >> >
>> >> > BTW there is some internal squeezed design documentation here,
>> just
>> >> in case its useful:
>> >> >
>> >> > https://github.com/xen-org/xen-api/tree/master/docs/squeezed
>> >> >
>> >> > Cheers,
>> >> > Dave
>> >> >
>> >> >>
>> >> >>
>> >> >> Cheers,
>> >> >> Kakadu
>> >> >>
>> >> >> [1] https://github.com/Kakadu/xcp-
>> >> xapi/blob/master/debian/patches/0101-
>> >> >> serred.patch#L123
>> >> >> [2] https://github.com/Kakadu/xcp-
>> >> xapi/blob/master/debian/patches/0101-
>> >> >> serred.patch#L617
>> >> >>
>> >> >> _______________________________________________
>> >> >> Xen-api mailing list
>> >> >> [email protected]
>> >> >> http://lists.xen.org/cgi-bin/mailman/listinfo/xen-api

_______________________________________________
Xen-api mailing list
[email protected]
http://lists.xen.org/cgi-bin/mailman/listinfo/xen-api

Reply via email to