qmemman.py 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350
  1. #!/usr/bin/python2
  2. # -*- coding: utf-8 -*-
  3. #
  4. # The Qubes OS Project, http://www.qubes-os.org
  5. #
  6. # Copyright (C) 2010 Rafal Wojtczuk <rafal@invisiblethingslab.com>
  7. # Copyright (C) 2013 Marek Marczykowski <marmarek@invisiblethingslab.com>
  8. #
  9. # This program is free software; you can redistribute it and/or
  10. # modify it under the terms of the GNU General Public License
  11. # as published by the Free Software Foundation; either version 2
  12. # of the License, or (at your option) any later version.
  13. #
  14. # This program is distributed in the hope that it will be useful,
  15. # but WITHOUT ANY WARRANTY; without even the implied warranty of
  16. # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  17. # GNU General Public License for more details.
  18. #
  19. # You should have received a copy of the GNU General Public License
  20. # along with this program; if not, write to the Free Software
  21. # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
  22. #
  23. #
  24. import xen.lowlevel.xc
  25. import xen.lowlevel.xs
  26. import string
  27. import time
  28. import qmemman_algo
  29. import os
  30. from notify import notify_error_qubes_manager, clear_error_qubes_manager
  31. import logging
  32. no_progress_msg="VM refused to give back requested memory"
  33. slow_memset_react_msg="VM didn't give back all requested memory"
  34. class DomainState:
  35. def __init__(self, id):
  36. self.meminfo = None #dictionary of memory info read from client
  37. self.memory_current = 0 # the current memory size
  38. self.memory_actual = None # the current memory allocation (what VM
  39. # is using or can use at any time)
  40. self.memory_maximum = None #the maximum memory size
  41. self.mem_used = None #used memory, computed based on meminfo
  42. self.id = id #domain id
  43. self.last_target = 0 #the last memset target
  44. self.no_progress = False #no react to memset
  45. self.slow_memset_react = False #slow react to memset (after few tries still above target)
  46. def __repr__(self):
  47. return self.__dict__.__repr__()
  48. class SystemState(object):
  49. def __init__(self):
  50. self.log = logging.getLogger('qmemman.systemstate')
  51. self.log.debug('SystemState()')
  52. self.domdict = {}
  53. self.xc = xen.lowlevel.xc.xc()
  54. self.xs = xen.lowlevel.xs.xs()
  55. self.BALOON_DELAY = 0.1
  56. self.XEN_FREE_MEM_LEFT = 50*1024*1024
  57. self.XEN_FREE_MEM_MIN = 25*1024*1024
  58. # Overhead of per-page Xen structures, taken from OpenStack nova/virt/xenapi/driver.py
  59. # see https://wiki.openstack.org/wiki/XenServer/Overhead
  60. # we divide total and free physical memory by this to get "assignable" memory
  61. self.MEM_OVERHEAD_FACTOR = 1.0 / 1.00781
  62. self.ALL_PHYS_MEM = int(self.xc.physinfo()['total_memory']*1024 * self.MEM_OVERHEAD_FACTOR)
  63. def add_domain(self, id):
  64. self.log.debug('add_domain(id={!r})'.format(id))
  65. self.domdict[id] = DomainState(id)
  66. # TODO: move to DomainState.__init__
  67. target_str = self.xs.read('', '/local/domain/' + id + '/memory/target')
  68. if target_str:
  69. self.domdict[id].last_target = int(target_str) * 1024
  70. def del_domain(self, id):
  71. self.log.debug('del_domain(id={!r})'.format(id))
  72. self.domdict.pop(id)
  73. def get_free_xen_memory(self):
  74. xen_free = int(self.xc.physinfo()['free_memory']*1024 *
  75. self.MEM_OVERHEAD_FACTOR)
  76. # now check for domains which have assigned more memory than really
  77. # used - do not count it as "free", because domain is free to use it
  78. # at any time
  79. # assumption: self.refresh_memactual was called before
  80. # (so domdict[id].memory_actual is up to date)
  81. assigned_but_unused = reduce(
  82. lambda acc, dom: acc + max(0, dom.last_target-dom.memory_current),
  83. self.domdict.values(),
  84. 0
  85. )
  86. # If, at any time, Xen have less memory than XEN_FREE_MEM_MIN,
  87. # it is a failure of qmemman. Collect as much data as possible to
  88. # debug it
  89. if xen_free < self.XEN_FREE_MEM_MIN:
  90. self.log.error("Xen free = {!r} below acceptable value! "
  91. "assigned_but_unused={!r}, domdict={!r}".format(
  92. xen_free, assigned_but_unused, self.domdict))
  93. elif xen_free < assigned_but_unused+self.XEN_FREE_MEM_MIN:
  94. self.log.error("Xen free = {!r} too small for satisfy assignments! "
  95. "assigned_but_unused={!r}, domdict={!r}".format(
  96. xen_free, assigned_but_unused, self.domdict))
  97. return xen_free - assigned_but_unused
  98. #refresh information on memory assigned to all domains
  99. def refresh_memactual(self):
  100. for domain in self.xc.domain_getinfo():
  101. id = str(domain['domid'])
  102. if self.domdict.has_key(id):
  103. # real memory usage
  104. self.domdict[id].memory_current = domain['mem_kb']*1024
  105. # what VM is using or can use
  106. self.domdict[id].memory_actual = max(
  107. self.domdict[id].memory_current,
  108. self.domdict[id].last_target
  109. )
  110. self.domdict[id].memory_maximum = self.xs.read('', '/local/domain/%s/memory/static-max' % str(id))
  111. if self.domdict[id].memory_maximum:
  112. self.domdict[id].memory_maximum = int(self.domdict[id].memory_maximum)*1024
  113. else:
  114. self.domdict[id].memory_maximum = self.ALL_PHYS_MEM
  115. # the previous line used to be
  116. # self.domdict[id].memory_maximum = domain['maxmem_kb']*1024
  117. # but domain['maxmem_kb'] changes in self.mem_set as well, and this results in
  118. # the memory never increasing
  119. # in fact, the only possible case of nonexisting memory/static-max is dom0
  120. # see #307
  121. def clear_outdated_error_markers(self):
  122. # Clear outdated errors
  123. for i in self.domdict.keys():
  124. if self.domdict[i].slow_memset_react and \
  125. self.domdict[i].memory_actual <= self.domdict[i].last_target + self.XEN_FREE_MEM_LEFT/4:
  126. dom_name = self.xs.read('', '/local/domain/%s/name' % str(i))
  127. if dom_name is not None:
  128. clear_error_qubes_manager(dom_name, slow_memset_react_msg)
  129. self.domdict[i].slow_memset_react = False
  130. if self.domdict[i].no_progress and \
  131. self.domdict[i].memory_actual <= self.domdict[i].last_target + self.XEN_FREE_MEM_LEFT/4:
  132. dom_name = self.xs.read('', '/local/domain/%s/name' % str(i))
  133. if dom_name is not None:
  134. clear_error_qubes_manager(dom_name, no_progress_msg)
  135. self.domdict[i].no_progress = False
  136. #the below works (and is fast), but then 'xm list' shows unchanged memory value
  137. def mem_set(self, id, val):
  138. self.log.info('mem-set domain {} to {}'.format(id, val))
  139. self.domdict[id].last_target = val
  140. #can happen in the middle of domain shutdown
  141. #apparently xc.lowlevel throws exceptions too
  142. try:
  143. self.xc.domain_setmaxmem(int(id), int(val/1024) + 1024) # LIBXL_MAXMEM_CONSTANT=1024
  144. self.xc.domain_set_target_mem(int(id), int(val/1024))
  145. except:
  146. pass
  147. self.xs.write('', '/local/domain/' + id + '/memory/target', str(int(val/1024)))
  148. # this is called at the end of ballooning, when we have Xen free mem already
  149. # make sure that past mem_set will not decrease Xen free mem
  150. def inhibit_balloon_up(self):
  151. self.log.debug('inhibit_balloon_up()')
  152. for i in self.domdict.keys():
  153. dom = self.domdict[i]
  154. if dom.memory_actual is not None and dom.memory_actual + 200*1024 < dom.last_target:
  155. self.log.info(
  156. 'Preventing balloon up to {}'.format(dom.last_target))
  157. self.mem_set(i, dom.memory_actual)
  158. #perform memory ballooning, across all domains, to add "memsize" to Xen free memory
  159. def do_balloon(self, memsize):
  160. self.log.info('do_balloon(memsize={!r})'.format(memsize))
  161. CHECK_PERIOD_S = 3
  162. CHECK_MB_S = 100
  163. niter = 0
  164. prev_memory_actual = None
  165. for i in self.domdict.keys():
  166. self.domdict[i].no_progress = False
  167. #: number of loop iterations for CHECK_PERIOD_S seconds
  168. check_period = max(1, int((CHECK_PERIOD_S + 0.0) / self.BALOON_DELAY))
  169. #: number of free memory bytes expected to get during CHECK_PERIOD_S
  170. #: seconds
  171. check_delta = CHECK_PERIOD_S * CHECK_MB_S * 1024 * 1024
  172. #: helper array for holding free memory size, CHECK_PERIOD_S seconds
  173. #: ago, at every loop iteration
  174. xenfree_ring = [0] * check_period
  175. while True:
  176. self.log.debug('niter={:2d}'.format(niter))
  177. self.refresh_memactual()
  178. xenfree = self.get_free_xen_memory()
  179. self.log.info('xenfree={!r}'.format(xenfree))
  180. if xenfree >= memsize + self.XEN_FREE_MEM_MIN:
  181. self.inhibit_balloon_up()
  182. return True
  183. # fail the request if over past CHECK_PERIOD_S seconds,
  184. # we got less than CHECK_MB_S MB/s on average
  185. ring_slot = niter % check_period
  186. if niter >= check_period and xenfree < xenfree_ring[ring_slot] + check_delta:
  187. return False
  188. xenfree_ring[ring_slot] = xenfree
  189. if prev_memory_actual is not None:
  190. for i in prev_memory_actual.keys():
  191. if prev_memory_actual[i] == self.domdict[i].memory_actual:
  192. #domain not responding to memset requests, remove it from donors
  193. self.domdict[i].no_progress = True
  194. self.log.info('domain {} stuck at {}'.format(i, self.domdict[i].memory_actual))
  195. memset_reqs = qmemman_algo.balloon(memsize + self.XEN_FREE_MEM_LEFT - xenfree, self.domdict)
  196. self.log.info('memset_reqs={!r}'.format(memset_reqs))
  197. if len(memset_reqs) == 0:
  198. return False
  199. prev_memory_actual = {}
  200. for i in memset_reqs:
  201. dom, mem = i
  202. self.mem_set(dom, mem)
  203. prev_memory_actual[dom] = self.domdict[dom].memory_actual
  204. self.log.debug('sleeping for {} s'.format(self.BALOON_DELAY))
  205. time.sleep(self.BALOON_DELAY)
  206. niter = niter + 1
  207. def refresh_meminfo(self, domid, untrusted_meminfo_key):
  208. self.log.debug(
  209. 'refresh_meminfo(domid={}, untrusted_meminfo_key={!r})'.format(
  210. domid, untrusted_meminfo_key))
  211. qmemman_algo.refresh_meminfo_for_domain(
  212. self.domdict[domid], untrusted_meminfo_key)
  213. self.do_balance()
  214. #is the computed balance request big enough ?
  215. #so that we do not trash with small adjustments
  216. def is_balance_req_significant(self, memset_reqs, xenfree):
  217. self.log.debug(
  218. 'is_balance_req_significant(memset_reqs={}, xenfree={})'.format(
  219. memset_reqs, xenfree))
  220. total_memory_transfer = 0
  221. MIN_TOTAL_MEMORY_TRANSFER = 150*1024*1024
  222. MIN_MEM_CHANGE_WHEN_UNDER_PREF = 15*1024*1024
  223. # If xenfree to low, return immediately
  224. if self.XEN_FREE_MEM_LEFT - xenfree > MIN_MEM_CHANGE_WHEN_UNDER_PREF:
  225. self.log.debug('xenfree is too low, returning')
  226. return True
  227. for rq in memset_reqs:
  228. dom, mem = rq
  229. last_target = self.domdict[dom].last_target
  230. memory_change = mem - last_target
  231. total_memory_transfer += abs(memory_change)
  232. pref = qmemman_algo.prefmem(self.domdict[dom])
  233. if last_target > 0 and last_target < pref and memory_change > MIN_MEM_CHANGE_WHEN_UNDER_PREF:
  234. self.log.info(
  235. 'dom {} is below pref, allowing balance'.format(dom))
  236. return True
  237. ret = total_memory_transfer + abs(xenfree - self.XEN_FREE_MEM_LEFT) > MIN_TOTAL_MEMORY_TRANSFER
  238. self.log.debug('is_balance_req_significant return {}'.format(ret))
  239. return ret
  240. def print_stats(self, xenfree, memset_reqs):
  241. for i in self.domdict.keys():
  242. if self.domdict[i].meminfo is not None:
  243. self.log.info('stat: dom {!r} act={} pref={}'.format(i,
  244. self.domdict[i].memory_actual,
  245. qmemman_algo.prefmem(self.domdict[i])))
  246. self.log.info('stat: xenfree={} memset_reqs={}'.format(xenfree, memset_reqs))
  247. def do_balance(self):
  248. self.log.debug('do_balance()')
  249. if os.path.isfile('/var/run/qubes/do-not-membalance'):
  250. self.log.debug('do-not-membalance file preset, returning')
  251. return
  252. self.refresh_memactual()
  253. self.clear_outdated_error_markers()
  254. xenfree = self.get_free_xen_memory()
  255. memset_reqs = qmemman_algo.balance(xenfree - self.XEN_FREE_MEM_LEFT, self.domdict)
  256. if not self.is_balance_req_significant(memset_reqs, xenfree):
  257. return
  258. self.print_stats(xenfree, memset_reqs)
  259. prev_memactual = {}
  260. for i in self.domdict.keys():
  261. prev_memactual[i] = self.domdict[i].memory_actual
  262. for rq in memset_reqs:
  263. dom, mem = rq
  264. # Force to always have at least 0.9*self.XEN_FREE_MEM_LEFT (some
  265. # margin for rounding errors). Before giving memory to
  266. # domain, ensure that others have gived it back.
  267. # If not - wait a little.
  268. ntries = 5
  269. while self.get_free_xen_memory() - (mem - self.domdict[dom].memory_actual) < 0.9*self.XEN_FREE_MEM_LEFT:
  270. self.log.debug('do_balance dom={!r} sleeping ntries={}'.format(
  271. dom, ntries))
  272. time.sleep(self.BALOON_DELAY)
  273. self.refresh_memactual()
  274. ntries -= 1
  275. if ntries <= 0:
  276. # Waiting haven't helped; Find which domain get stuck and
  277. # abort balance (after distributing what we have)
  278. for rq2 in memset_reqs:
  279. dom2, mem2 = rq2
  280. if dom2 == dom:
  281. # All donors have been procesed
  282. break
  283. # allow some small margin
  284. if self.domdict[dom2].memory_actual > self.domdict[dom2].last_target + self.XEN_FREE_MEM_LEFT/4:
  285. # VM didn't react to memory request at all, remove from donors
  286. if prev_memactual[dom2] == self.domdict[dom2].memory_actual:
  287. self.log.warning(
  288. 'dom {!r} didnt react to memory request'
  289. ' (holds {}, requested balloon down to {})'
  290. .format(dom2,
  291. self.domdict[dom2].memory_actual,
  292. mem2))
  293. self.domdict[dom2].no_progress = True
  294. dom_name = self.xs.read('', '/local/domain/%s/name' % str(dom2))
  295. if dom_name is not None:
  296. notify_error_qubes_manager(str(dom_name), no_progress_msg)
  297. else:
  298. self.log.warning('dom {!r} still hold more'
  299. ' memory than have assigned ({} > {})'
  300. .format(dom2,
  301. self.domdict[dom2].memory_actual,
  302. mem2))
  303. self.domdict[dom2].slow_memset_react = True
  304. dom_name = self.xs.read('', '/local/domain/%s/name' % str(dom2))
  305. if dom_name is not None:
  306. notify_error_qubes_manager(str(dom_name), slow_memset_react_msg)
  307. self.mem_set(dom, self.get_free_xen_memory() + self.domdict[dom].memory_actual - self.XEN_FREE_MEM_LEFT)
  308. return
  309. self.mem_set(dom, mem)
  310. # for i in self.domdict.keys():
  311. # print 'domain ', i, ' meminfo=', self.domdict[i].meminfo, 'actual mem', self.domdict[i].memory_actual
  312. # print 'domain ', i, 'actual mem', self.domdict[i].memory_actual
  313. # print 'xen free mem', self.get_free_xen_memory()