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

RE: [Xen-API] Squeezed + slow domain = broken dynamic memory

  • To: Dave Scott <Dave.Scott@xxxxxxxxxxxxx>
  • From: George Shuklin <george.shuklin@xxxxxxxxx>
  • Date: Fri, 14 Jan 2011 19:23:44 +0300
  • Cc: "xen-api@xxxxxxxxxxxxxxxxxxx" <xen-api@xxxxxxxxxxxxxxxxxxx>
  • Delivery-date: Fri, 14 Jan 2011 08:24:02 -0800
  • Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=subject:from:to:cc:in-reply-to:references:content-type:date :message-id:mime-version:x-mailer:content-transfer-encoding; b=i+/GC+74b3qiNcW7qtJhKoSNKTtSjKnPIF3mE7WsTN3kFplww5uH2bbkx1iBTkbBvr L8DIBUMZp45o3yAUmlVPlMoYg/W/0LiBp+jagDbD6st+RcJ13RVUBtWglS0wvlYuWpa5 xO4fRdzx33usnej36ijj4Pb+bRIH7kklyboCk=
  • List-id: Discussion of API issues surrounding Xen <xen-api.lists.xensource.com>

Ð ÐÑ., 14/01/2011 Ð 16:06 +0000, Dave Scott ÐÐÑÐÑ:
> Hi George,
> > Good day.
> > 
> > I found strange limitation in squeezed code: if domain's balloon does
> > not reply to memory change request it marked as 'uncooperative'.
> > 
> > This cause stops to react to vm-memory-dynamic-set commands and stay
> > until xe toolstak restarted.
> Hm, I thought if you used vm-memory-dynamic-set then the new memory target 
> would still be written in xenstore -- does it remain at the old value?

Yes, I really thought this too. But our memory-on-demand service use
direct interaction with domain's balloon via xenstore and suddenly it
stops to work. Host have huge memory reserves (about 20-30G - more than
sum of all memory-static-max for all VM's), but memory suddenly stops to
adjusting to bigger values. (we can lower, we can come back, but we can
go upper). And xe vm-memory-target fail silently too - it change XenAPI
values for dynamic-min/dynamic-max, but do not change actual TotalMem
value in guest vm (and mem_kb value for xc.domain_getinfo()). After
xe-toolstack-restart problem solves (without guest restart), so it does
not related to 'uncooperative' balloon driver. (other solution is
migrate vm to other host).

This problem is very futile and occur rarely (about one case in 3-5 days
with about 400 VM's on 8 host's pool), but with constant symptoms. 

After squeezed code research I saw error message text and found them in
squeezed.log for 'failing' domains:

20110114T11:14:05.376Z|debug|server|0|reserve_memory_range(xapi, 359424,
359424)|xenops] domid 125 has been declared inactive

Thank you very much for reply.

wBR, George Shuklin

xen-api mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.