qmemman.py 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235
  1. import xen.lowlevel.xc
  2. import xen.lowlevel.xs
  3. import string
  4. import time
  5. import qmemman_algo
  6. import os
  7. from guihelpers import notify_error_qubes_manager, clear_error_qubes_manager
  8. no_progress_msg="VM refused to give back requested memory"
  9. slow_memset_react_msg="VM didn't give back all requested memory"
  10. class DomainState:
  11. def __init__(self, id):
  12. self.meminfo = None #dictionary of memory info read from client
  13. self.memory_actual = None #the current memory size
  14. self.memory_maximum = None #the maximum memory size
  15. self.mem_used = None #used memory, computed based on meminfo
  16. self.id = id #domain id
  17. self.last_target = 0 #the last memset target
  18. self.no_progress = False #no react to memset
  19. self.slow_memset_react = False #slow react to memset (after few tries still above target)
  20. class SystemState:
  21. def __init__(self):
  22. self.domdict = {}
  23. self.xc = xen.lowlevel.xc.xc()
  24. self.xs = xen.lowlevel.xs.xs()
  25. self.BALOON_DELAY = 0.1
  26. self.XEN_FREE_MEM_LEFT = 50*1024*1024
  27. self.XEN_FREE_MEM_MIN = 25*1024*1024
  28. self.ALL_PHYS_MEM = self.xc.physinfo()['total_memory']*1024
  29. def add_domain(self, id):
  30. self.domdict[id] = DomainState(id)
  31. def del_domain(self, id):
  32. self.domdict.pop(id)
  33. def get_free_xen_memory(self):
  34. return self.xc.physinfo()['free_memory']*1024
  35. # hosts = self.xend_session.session.xenapi.host.get_all()
  36. # host_record = self.xend_session.session.xenapi.host.get_record(hosts[0])
  37. # host_metrics_record = self.xend_session.session.xenapi.host_metrics.get_record(host_record["metrics"])
  38. # ret = host_metrics_record["memory_free"]
  39. # return long(ret)
  40. #refresh information on memory assigned to all domains
  41. def refresh_memactual(self):
  42. for domain in self.xc.domain_getinfo():
  43. id = str(domain['domid'])
  44. if self.domdict.has_key(id):
  45. self.domdict[id].memory_actual = domain['mem_kb']*1024
  46. self.domdict[id].memory_maximum = self.xs.read('', '/local/domain/%s/memory/static-max' % str(id))
  47. if not self.domdict[id].memory_maximum:
  48. self.domdict[id].memory_maximum = self.ALL_PHYS_MEM
  49. # the previous line used to be
  50. # self.domdict[id].memory_maximum = domain['maxmem_kb']*1024
  51. # but domain['maxmem_kb'] changes in self.mem_set as well, and this results in
  52. # the memory never increasing
  53. # in fact, the only possible case of nonexisting memory/static-max is dom0
  54. # see #307
  55. def clear_outdated_error_markers(self):
  56. # Clear outdated errors
  57. for i in self.domdict.keys():
  58. if self.domdict[i].slow_memset_react and \
  59. self.domdict[i].memory_actual <= self.domdict[i].last_target + self.XEN_FREE_MEM_LEFT/4:
  60. dom_name = self.xs.read('', '/local/domain/%s/name' % str(i))
  61. if dom_name is not None:
  62. clear_error_qubes_manager(dom_name, slow_memset_react_msg)
  63. self.domdict[i].slow_memset_react = False
  64. if self.domdict[i].no_progress and \
  65. self.domdict[i].memory_actual <= self.domdict[i].last_target + self.XEN_FREE_MEM_LEFT/4:
  66. dom_name = self.xs.read('', '/local/domain/%s/name' % str(i))
  67. if dom_name is not None:
  68. clear_error_qubes_manager(dom_name, no_progress_msg)
  69. self.domdict[i].no_progress = False
  70. #the below works (and is fast), but then 'xm list' shows unchanged memory value
  71. def mem_set(self, id, val):
  72. print 'mem-set domain', id, 'to', val
  73. self.domdict[id].last_target = val
  74. #can happen in the middle of domain shutdown
  75. #apparently xc.lowlevel throws exceptions too
  76. try:
  77. self.xc.domain_setmaxmem(int(id), int(val/1024) + 1024) # LIBXL_MAXMEM_CONSTANT=1024
  78. self.xc.domain_set_target_mem(int(id), int(val/1024))
  79. except:
  80. pass
  81. self.xs.write('', '/local/domain/' + id + '/memory/target', str(int(val/1024)))
  82. def mem_set_obsolete(self, id, val):
  83. uuid = self.domdict[id].uuid
  84. if val >= 2**31:
  85. print 'limiting memory from ', val, 'to maxint because of xml-rpc lameness'
  86. val = 2**31 - 1
  87. print 'mem-set domain', id, 'to', val
  88. try:
  89. self.xend_session.session.xenapi.VM.set_memory_dynamic_max_live(uuid, val)
  90. self.xend_session.session.xenapi.VM.set_memory_dynamic_min_live(uuid, val)
  91. #can happen in the middle of domain shutdown
  92. except XenAPI.Failure:
  93. pass
  94. # this is called at the end of ballooning, when we have Xen free mem already
  95. # make sure that past mem_set will not decrease Xen free mem
  96. def inhibit_balloon_up(self):
  97. for i in self.domdict.keys():
  98. dom = self.domdict[i]
  99. if dom.memory_actual is not None and dom.memory_actual + 200*1024 < dom.last_target:
  100. print "Preventing balloon up to", dom.last_target
  101. self.mem_set(i, dom.memory_actual)
  102. #perform memory ballooning, across all domains, to add "memsize" to Xen free memory
  103. def do_balloon(self, memsize):
  104. MAX_TRIES = 20
  105. niter = 0
  106. prev_memory_actual = None
  107. for i in self.domdict.keys():
  108. self.domdict[i].no_progress = False
  109. print "do_balloon start"
  110. while True:
  111. self.refresh_memactual()
  112. xenfree = self.get_free_xen_memory()
  113. print 'got xenfree=', xenfree
  114. if xenfree >= memsize + self.XEN_FREE_MEM_MIN:
  115. self.inhibit_balloon_up()
  116. return True
  117. if prev_memory_actual is not None:
  118. for i in prev_memory_actual.keys():
  119. if prev_memory_actual[i] == self.domdict[i].memory_actual:
  120. #domain not responding to memset requests, remove it from donors
  121. self.domdict[i].no_progress = True
  122. print 'domain', i, 'stuck at', self.domdict[i].memory_actual
  123. memset_reqs = qmemman_algo.balloon(memsize + self.XEN_FREE_MEM_LEFT - xenfree, self.domdict)
  124. print 'requests:', memset_reqs
  125. if niter > MAX_TRIES or len(memset_reqs) == 0:
  126. return False
  127. prev_memory_actual = {}
  128. for i in memset_reqs:
  129. dom, mem = i
  130. self.mem_set(dom, mem)
  131. prev_memory_actual[dom] = self.domdict[dom].memory_actual
  132. time.sleep(self.BALOON_DELAY)
  133. niter = niter + 1
  134. def refresh_meminfo(self, domid, untrusted_meminfo_key):
  135. qmemman_algo.refresh_meminfo_for_domain(self.domdict[domid], untrusted_meminfo_key)
  136. self.do_balance()
  137. #is the computed balance request big enough ?
  138. #so that we do not trash with small adjustments
  139. def is_balance_req_significant(self, memset_reqs, xenfree):
  140. total_memory_transfer = 0
  141. MIN_TOTAL_MEMORY_TRANSFER = 150*1024*1024
  142. MIN_MEM_CHANGE_WHEN_UNDER_PREF = 15*1024*1024
  143. # If xenfree to low, return immediately
  144. if self.XEN_FREE_MEM_LEFT - xenfree > MIN_MEM_CHANGE_WHEN_UNDER_PREF:
  145. return True
  146. for rq in memset_reqs:
  147. dom, mem = rq
  148. last_target = self.domdict[dom].last_target
  149. memory_change = mem - last_target
  150. total_memory_transfer += abs(memory_change)
  151. pref = qmemman_algo.prefmem(self.domdict[dom])
  152. if last_target > 0 and last_target < pref and memory_change > MIN_MEM_CHANGE_WHEN_UNDER_PREF:
  153. print 'dom', dom, 'is below pref, allowing balance'
  154. return True
  155. return total_memory_transfer + abs(xenfree - self.XEN_FREE_MEM_LEFT) > MIN_TOTAL_MEMORY_TRANSFER
  156. def print_stats(self, xenfree, memset_reqs):
  157. for i in self.domdict.keys():
  158. if self.domdict[i].meminfo is not None:
  159. print 'dom' , i, 'act/pref', self.domdict[i].memory_actual, qmemman_algo.prefmem(self.domdict[i])
  160. print 'xenfree=', xenfree, 'balance req:', memset_reqs
  161. def do_balance(self):
  162. if os.path.isfile('/var/run/qubes/do-not-membalance'):
  163. return
  164. self.refresh_memactual()
  165. self.clear_outdated_error_markers()
  166. xenfree = self.get_free_xen_memory()
  167. memset_reqs = qmemman_algo.balance(xenfree - self.XEN_FREE_MEM_LEFT, self.domdict)
  168. if not self.is_balance_req_significant(memset_reqs, xenfree):
  169. return
  170. self.print_stats(xenfree, memset_reqs)
  171. prev_memactual = {}
  172. for i in self.domdict.keys():
  173. prev_memactual[i] = self.domdict[i].memory_actual
  174. for rq in memset_reqs:
  175. dom, mem = rq
  176. # Force to always have at least 0.9*self.XEN_FREE_MEM_LEFT (some
  177. # margin for rounding errors). Before giving memory to
  178. # domain, ensure that others have gived it back.
  179. # If not - wait a little.
  180. ntries = 5
  181. while self.get_free_xen_memory() - (mem - self.domdict[dom].memory_actual) < 0.9*self.XEN_FREE_MEM_LEFT:
  182. time.sleep(self.BALOON_DELAY)
  183. ntries -= 1
  184. if ntries <= 0:
  185. # Waiting haven't helped; Find which domain get stuck and
  186. # abort balance (after distributing what we have)
  187. self.refresh_memactual()
  188. for rq2 in memset_reqs:
  189. dom2, mem2 = rq2
  190. if dom2 == dom:
  191. # All donors have been procesed
  192. break
  193. # allow some small margin
  194. if self.domdict[dom2].memory_actual > self.domdict[dom2].last_target + self.XEN_FREE_MEM_LEFT/4:
  195. # VM didn't react to memory request at all, remove from donors
  196. if prev_memactual[dom2] == self.domdict[dom2].memory_actual:
  197. print 'dom %s didnt react to memory request (holds %d, requested balloon down to %d)' % (dom2, self.domdict[dom2].memory_actual, mem2)
  198. self.domdict[dom2].no_progress = True
  199. dom_name = self.xs.read('', '/local/domain/%s/name' % str(dom2))
  200. if dom_name is not None:
  201. notify_error_qubes_manager(str(dom_name), no_progress_msg)
  202. else:
  203. print 'dom %s still hold more memory than have assigned (%d > %d)' % (dom2, self.domdict[dom2].memory_actual, mem2)
  204. self.domdict[dom2].slow_memset_react = True
  205. dom_name = self.xs.read('', '/local/domain/%s/name' % str(dom2))
  206. if dom_name is not None:
  207. notify_error_qubes_manager(str(dom_name), slow_memset_react_msg)
  208. self.mem_set(dom, self.get_free_xen_memory() + self.domdict[dom].memory_actual - self.XEN_FREE_MEM_LEFT)
  209. return
  210. self.mem_set(dom, mem)
  211. # for i in self.domdict.keys():
  212. # print 'domain ', i, ' meminfo=', self.domdict[i].meminfo, 'actual mem', self.domdict[i].memory_actual
  213. # print 'domain ', i, 'actual mem', self.domdict[i].memory_actual
  214. # print 'xen free mem', self.get_free_xen_memory()