2 # -*- coding: utf-8 -*-
4 from constants import TESTBED_ID
19 from nepi.util import server
20 from nepi.util import parallel
26 class UnresponsiveNodeError(RuntimeError):
29 def _castproperty(typ, propattr):
31 return getattr(self, propattr)
32 def _set(self, value):
33 if value is not None or (isinstance(value, basestring) and not value):
35 return setattr(self, propattr, value)
36 def _del(self, value):
37 return delattr(self, propattr)
38 _get.__name__ = propattr + '_get'
39 _set.__name__ = propattr + '_set'
40 _del.__name__ = propattr + '_del'
41 return property(_get, _set, _del)
45 # Map Node attribute to plcapi filter name
46 'hostname' : 'hostname',
50 # Map Node attribute to (<tag name>, <plcapi filter expression>)
51 # There are replacements that are applied with string formatting,
52 # so '%' has to be escaped as '%%'.
53 'architecture' : ('arch','value'),
54 'operatingSystem' : ('fcdistro','value'),
55 'pl_distro' : ('pldistro','value'),
56 'city' : ('city','value'),
57 'country' : ('country','value'),
58 'region' : ('region','value'),
59 'minReliability' : ('reliability%(timeframe)s', ']value'),
60 'maxReliability' : ('reliability%(timeframe)s', '[value'),
61 'minBandwidth' : ('bw%(timeframe)s', ']value'),
62 'maxBandwidth' : ('bw%(timeframe)s', '[value'),
63 'minLoad' : ('load%(timeframe)s', ']value'),
64 'maxLoad' : ('load%(timeframe)s', '[value'),
65 'minCpu' : ('cpu%(timeframe)s', ']value'),
66 'maxCpu' : ('cpu%(timeframe)s', '[value'),
70 # (<tag name>, <weight>, <default>)
71 ('bw%(timeframe)s', -0.001, 1024.0),
72 ('cpu%(timeframe)s', 0.1, 40.0),
73 ('load%(timeframe)s', -0.2, 3.0),
74 ('reliability%(timeframe)s', 1, 100.0),
77 DEPENDS_PIDFILE = '/tmp/nepi-depends.pid'
78 DEPENDS_LOGFILE = '/tmp/nepi-depends.log'
79 RPM_FUSION_URL = 'http://download1.rpmfusion.org/free/fedora/rpmfusion-free-release-stable.noarch.rpm'
80 RPM_FUSION_URL_F12 = 'http://download1.rpmfusion.org/free/fedora/releases/12/Everything/x86_64/os/rpmfusion-free-release-12-1.noarch.rpm'
82 minReliability = _castproperty(float, '_minReliability')
83 maxReliability = _castproperty(float, '_maxReliability')
84 minBandwidth = _castproperty(float, '_minBandwidth')
85 maxBandwidth = _castproperty(float, '_maxBandwidth')
86 minCpu = _castproperty(float, '_minCpu')
87 maxCpu = _castproperty(float, '_maxCpu')
88 minLoad = _castproperty(float, '_minLoad')
89 maxLoad = _castproperty(float, '_maxLoad')
91 def __init__(self, api=None):
98 self.architecture = None
99 self.operatingSystem = None
100 self.pl_distro = None
105 self.minReliability = None
106 self.maxReliability = None
107 self.minBandwidth = None
108 self.maxBandwidth = None
113 self.min_num_external_ifaces = None
114 self.max_num_external_ifaces = None
117 # Applications and routes add requirements to connected nodes
118 self.required_packages = set()
119 self.required_vsys = set()
121 self.rpmFusion = False
122 self.env = collections.defaultdict(list)
124 # Some special applications - initialized when connected
125 self.multicast_forwarder = None
127 # Testbed-derived attributes
128 self.slicename = None
129 self.ident_path = None
130 self.server_key = None
131 self.home_path = None
132 self.enable_cleanup = False
134 # Those are filled when an actual node is allocated
136 self._yum_dependencies = None
137 self._installed = False
140 self._logger = logging.getLogger('nepi.testbeds.planetlab')
142 def _nepi_testbed_environment_setup_get(self):
143 command = cStringIO.StringIO()
144 command.write('export PYTHONPATH=$PYTHONPATH:%s' % (
145 ':'.join(["${HOME}/"+server.shell_escape(s) for s in self.pythonpath])
147 command.write(' ; export PATH=$PATH:%s' % (
148 ':'.join(["${HOME}/"+server.shell_escape(s) for s in self.pythonpath])
151 for envkey, envvals in self.env.iteritems():
152 for envval in envvals:
153 command.write(' ; export %s=%s' % (envkey, envval))
154 return command.getvalue()
155 def _nepi_testbed_environment_setup_set(self, value):
157 _nepi_testbed_environment_setup = property(
158 _nepi_testbed_environment_setup_get,
159 _nepi_testbed_environment_setup_set)
161 def build_filters(self, target_filters, filter_map):
162 for attr, tag in filter_map.iteritems():
163 value = getattr(self, attr, None)
164 if value is not None:
165 target_filters[tag] = value
166 return target_filters
169 def applicable_filters(self):
170 has = lambda att : getattr(self,att,None) is not None
172 filter(has, self.BASEFILTERS.iterkeys())
173 + filter(has, self.TAGFILTERS.iterkeys())
176 def find_candidates(self, filter_slice_id=None):
177 self._logger.info("Finding candidates for %s", self.make_filter_description())
179 fields = ('node_id',)
180 replacements = {'timeframe':self.timeframe}
182 # get initial candidates (no tag filters)
183 basefilters = self.build_filters({}, self.BASEFILTERS)
184 rootfilters = basefilters.copy()
186 basefilters['|slice_ids'] = (filter_slice_id,)
188 # only pick healthy nodes
189 basefilters['run_level'] = 'boot'
190 basefilters['boot_state'] = 'boot'
191 basefilters['node_type'] = 'regular' # nepi can only handle regular nodes (for now)
192 basefilters['>last_contact'] = int(time.time()) - 5*3600 # allow 5h out of contact, for timezone discrepancies
194 # keyword-only "pseudofilters"
197 extra['peer'] = self.site
199 candidates = set(map(operator.itemgetter('node_id'),
200 self._api.GetNodes(filters=basefilters, fields=fields, **extra)))
202 # filter by tag, one tag at a time
203 applicable = self.applicable_filters
204 for tagfilter in self.TAGFILTERS.iteritems():
205 attr, (tagname, expr) = tagfilter
207 # don't bother if there's no filter defined
208 if attr in applicable:
209 tagfilter = rootfilters.copy()
210 tagfilter['tagname'] = tagname % replacements
211 tagfilter[expr % replacements] = getattr(self,attr)
212 tagfilter['node_id'] = list(candidates)
214 candidates &= set(map(operator.itemgetter('node_id'),
215 self._api.GetNodeTags(filters=tagfilter, fields=fields)))
217 # filter by vsys tags - special case since it doesn't follow
218 # the usual semantics
219 if self.required_vsys:
220 newcandidates = collections.defaultdict(set)
222 vsys_tags = self._api.GetNodeTags(
224 node_id = list(candidates),
225 fields = ['node_id','value'])
228 operator.itemgetter(['node_id','value']),
231 required_vsys = self.required_vsys
232 for node_id, value in vsys_tags:
233 if value in required_vsys:
234 newcandidates[value].add(node_id)
236 # take only those that have all the required vsys tags
237 newcandidates = reduce(
238 lambda accum, new : accum & new,
239 newcandidates.itervalues(),
242 # filter by iface count
243 if self.min_num_external_ifaces is not None or self.max_num_external_ifaces is not None:
244 # fetch interfaces for all, in one go
245 filters = basefilters.copy()
246 filters['node_id'] = list(candidates)
247 ifaces = dict(map(operator.itemgetter('node_id','interface_ids'),
248 self._api.GetNodes(filters=basefilters, fields=('node_id','interface_ids')) ))
250 # filter candidates by interface count
251 if self.min_num_external_ifaces is not None and self.max_num_external_ifaces is not None:
252 predicate = ( lambda node_id :
253 self.min_num_external_ifaces <= len(ifaces.get(node_id,())) <= self.max_num_external_ifaces )
254 elif self.min_num_external_ifaces is not None:
255 predicate = ( lambda node_id :
256 self.min_num_external_ifaces <= len(ifaces.get(node_id,())) )
258 predicate = ( lambda node_id :
259 len(ifaces.get(node_id,())) <= self.max_num_external_ifaces )
261 candidates = set(filter(predicate, candidates))
263 # make sure hostnames are resolvable
265 self._logger.info(" Found %s candidates. Checking for reachability...", len(candidates))
267 hostnames = dict(map(operator.itemgetter('node_id','hostname'),
268 self._api.GetNodes(list(candidates), ['node_id','hostname'])
270 def resolvable(node_id):
272 addr = socket.gethostbyname(hostnames[node_id])
273 return addr is not None
276 candidates = set(parallel.pfilter(resolvable, candidates,
279 self._logger.info(" Found %s reachable candidates.", len(candidates))
283 def make_filter_description(self):
285 Makes a human-readable description of filtering conditions
289 # get initial candidates (no tag filters)
290 filters = self.build_filters({}, self.BASEFILTERS)
292 # keyword-only "pseudofilters"
294 filters['peer'] = self.site
296 # filter by tag, one tag at a time
297 applicable = self.applicable_filters
298 for tagfilter in self.TAGFILTERS.iteritems():
299 attr, (tagname, expr) = tagfilter
301 # don't bother if there's no filter defined
302 if attr in applicable:
303 filters[attr] = getattr(self,attr)
305 # filter by vsys tags - special case since it doesn't follow
306 # the usual semantics
307 if self.required_vsys:
308 filters['vsys'] = ','.join(list(self.required_vsys))
310 # filter by iface count
311 if self.min_num_external_ifaces is not None or self.max_num_external_ifaces is not None:
312 filters['num_ifaces'] = '-'.join([
313 str(self.min_num_external_ifaces or '0'),
314 str(self.max_num_external_ifaces or 'inf')
317 return '; '.join(map('%s: %s'.__mod__,filters.iteritems()))
319 def assign_node_id(self, node_id):
320 self._node_id = node_id
321 self.fetch_node_info()
323 def unassign_node(self):
325 self.__dict__.update(self.__orig_attrs)
327 def rate_nodes(self, nodes):
328 rates = collections.defaultdict(int)
329 tags = collections.defaultdict(dict)
330 replacements = {'timeframe':self.timeframe}
331 tagnames = [ tagname % replacements
332 for tagname, weight, default in self.RATE_FACTORS ]
334 taginfo = self._api.GetNodeTags(
337 fields=('node_id','tagname','value'))
339 unpack = operator.itemgetter('node_id','tagname','value')
340 for value in taginfo:
341 node, tagname, value = unpack(value)
342 if value and value.lower() != 'n/a':
343 tags[tagname][int(node)] = float(value)
345 for tagname, weight, default in self.RATE_FACTORS:
346 taginfo = tags[tagname % replacements].get
348 rates[node] += weight * taginfo(node,default)
350 return map(rates.__getitem__, nodes)
352 def fetch_node_info(self):
355 self._api.StartMulticall()
356 info = self._api.GetNodes(self._node_id)
357 tags = self._api.GetNodeTags(node_id=self._node_id, fields=('tagname','value'))
358 info, tags = self._api.FinishMulticall()
361 tags = dict( (t['tagname'],t['value'])
364 orig_attrs['min_num_external_ifaces'] = self.min_num_external_ifaces
365 orig_attrs['max_num_external_ifaces'] = self.max_num_external_ifaces
366 self.min_num_external_ifaces = None
367 self.max_num_external_ifaces = None
370 replacements = {'timeframe':self.timeframe}
371 for attr, tag in self.BASEFILTERS.iteritems():
374 if hasattr(self, attr):
375 orig_attrs[attr] = getattr(self, attr)
376 setattr(self, attr, value)
377 for attr, (tag,_) in self.TAGFILTERS.iteritems():
378 tag = tag % replacements
381 if hasattr(self, attr):
382 orig_attrs[attr] = getattr(self, attr)
383 if not value or value.lower() == 'n/a':
385 setattr(self, attr, value)
387 if 'peer_id' in info:
388 orig_attrs['site'] = self.site
389 self.site = self._api.peer_map[info['peer_id']]
391 if 'interface_ids' in info:
392 self.min_num_external_ifaces = \
393 self.max_num_external_ifaces = len(info['interface_ids'])
395 if 'ssh_rsa_key' in info:
396 orig_attrs['server_key'] = self.server_key
397 self.server_key = info['ssh_rsa_key']
399 self.__orig_attrs = orig_attrs
402 if self.home_path is None:
403 raise AssertionError, "Misconfigured node: missing home path"
404 if self.ident_path is None or not os.access(self.ident_path, os.R_OK):
405 raise AssertionError, "Misconfigured node: missing slice SSH key"
406 if self.slicename is None:
407 raise AssertionError, "Misconfigured node: unspecified slice"
410 # Mark dependencies installed
411 self._installed = True
413 # Clear load attributes, they impair re-discovery
414 self.minReliability = \
415 self.maxReliability = \
416 self.minBandwidth = \
417 self.maxBandwidth = \
423 def install_dependencies(self):
424 if self.required_packages and not self._installed:
425 # If we need rpmfusion, we must install the repo definition and the gpg keys
427 if self.operatingSystem == 'f12':
428 # Fedora 12 requires a different rpmfusion package
429 RPM_FUSION_URL = self.RPM_FUSION_URL_F12
431 # This one works for f13+
432 RPM_FUSION_URL = self.RPM_FUSION_URL
435 '( rpm -q $(rpm -q -p %(RPM_FUSION_URL)s) || rpm -i %(RPM_FUSION_URL)s ) &&'
437 'RPM_FUSION_URL' : RPM_FUSION_URL
443 (out,err),proc = server.popen_ssh_command(
445 host = self.hostname,
447 user = self.slicename,
449 ident_key = self.ident_path,
450 server_key = self.server_key,
455 raise RuntimeError, "Failed to set up application: %s %s" % (out,err,)
457 # Launch p2p yum dependency installer
458 self._yum_dependencies.async_setup()
460 def wait_provisioning(self, timeout = 20*60):
461 # Wait for the p2p installer
464 while not self.is_alive():
465 time.sleep(sleeptime)
466 totaltime += sleeptime
467 sleeptime = min(30.0, sleeptime*1.5)
469 if totaltime > timeout:
470 # PlanetLab has a 15' delay on configuration propagation
471 # If we're above that delay, the unresponsiveness is not due
473 if not self.is_alive(verbose=True):
474 raise UnresponsiveNodeError, "Unresponsive host %s" % (self.hostname,)
476 # Ensure the node is clean (no apps running that could interfere with operations)
477 if self.enable_cleanup:
480 def wait_dependencies(self, pidprobe=1, probe=0.5, pidmax=10, probemax=10):
481 # Wait for the p2p installer
482 if self._yum_dependencies and not self._installed:
483 self._yum_dependencies.async_setup_wait()
484 self._installed = True
486 def is_alive(self, verbose = False):
487 # Make sure all the paths are created where
488 # they have to be created for deployment
489 (out,err),proc = server.eintr_retry(server.popen_ssh_command)(
491 host = self.hostname,
493 user = self.slicename,
495 ident_key = self.ident_path,
496 server_key = self.server_key,
498 err_on_timeout = False,
504 self._logger.warn("Unresponsive node %s got:\n%s%s", self.hostname, out, err)
506 elif not err and out.strip() == 'ALIVE':
510 self._logger.warn("Unresponsive node %s got:\n%s%s", self.hostname, out, err)
514 if self.enable_cleanup:
517 def do_cleanup(self):
518 if self.testbed().recovering:
522 self._logger.info("Cleaning up %s", self.hostname)
525 "sudo -S killall python tcpdump || /bin/true ; "
526 "sudo -S killall python tcpdump || /bin/true ; "
527 "sudo -S kill $(ps -N -T -o pid --no-heading | grep -v $PPID | sort) || /bin/true ",
528 "sudo -S killall -u %(slicename)s || /bin/true ",
529 "sudo -S killall -u root || /bin/true ",
530 "sudo -S killall -u %(slicename)s || /bin/true ",
531 "sudo -S killall -u root || /bin/true ",
535 (out,err),proc = server.popen_ssh_command(
536 # Some apps need two kills
538 'slicename' : self.slicename ,
540 host = self.hostname,
542 user = self.slicename,
544 ident_key = self.ident_path,
545 server_key = self.server_key,
546 tty = True, # so that ps -N -T works as advertised...
552 def prepare_dependencies(self):
553 # Configure p2p yum dependency installer
554 if self.required_packages and not self._installed:
555 self._yum_dependencies = application.YumDependency(self._api)
556 self._yum_dependencies.node = self
557 self._yum_dependencies.home_path = "nepi-yumdep"
558 self._yum_dependencies.depends = ' '.join(self.required_packages)
560 def routing_method(self, routes, vsys_vnet):
562 There are two methods, vroute and sliceip.
565 Modifies the node's routing table directly, validating that the IP
566 range lies within the network given by the slice's vsys_vnet tag.
567 This method is the most scalable for very small routing tables
568 that need not modify other routes (including the default)
571 Uses policy routing and iptables filters to create per-sliver
572 routing tables. It's the most flexible way, but it doesn't scale
573 as well since only 155 routing tables can be created this way.
575 This method will return the most appropriate routing method, which will
576 prefer vroute for small routing tables.
579 # For now, sliceip results in kernel panics
580 # so we HAVE to use vroute
583 # We should not make the routing table grow too big
584 if len(routes) > MAX_VROUTE_ROUTES:
587 vsys_vnet = ipaddr.IPNetwork(vsys_vnet)
589 dest, prefix, nexthop, metric = route
590 dest = ipaddr.IPNetwork("%s/%d" % (dest,prefix))
591 nexthop = ipaddr.IPAddress(nexthop)
592 if dest not in vsys_vnet or nexthop not in vsys_vnet:
597 def format_route(self, route, dev, method, action):
598 dest, prefix, nexthop, metric = route
599 if method == 'vroute':
601 "%s %s%s gw %s %s" % (
604 (("/%d" % (prefix,)) if prefix and prefix != 32 else ""),
609 elif method == 'sliceip':
611 "route %s to %s%s via %s metric %s dev %s" % (
614 (("/%d" % (prefix,)) if prefix and prefix != 32 else ""),
621 raise AssertionError, "Unknown method"
623 def _annotate_routes_with_devs(self, routes, devs, method):
627 if dev.routes_here(route):
628 dev_routes.append(tuple(route) + (dev.if_name,))
633 if method == 'sliceip':
634 dev_routes.append(tuple(route) + ('eth0',))
636 raise RuntimeError, "Route %s cannot be bound to any virtual interface " \
637 "- PL can only handle rules over virtual interfaces. Candidates are: %s" % (route,devs)
640 def configure_routes(self, routes, devs, vsys_vnet):
642 Add the specified routes to the node's routing table
645 method = self.routing_method(routes, vsys_vnet)
648 # annotate routes with devices
649 dev_routes = self._annotate_routes_with_devs(routes, devs, method)
650 for route in dev_routes:
651 route, dev = route[:-1], route[-1]
655 rules.append(self.format_route(route, dev, method, 'add'))
657 if method == 'sliceip':
658 rules = map('enable '.__add__, tdevs) + rules
660 self._logger.info("Setting up routes for %s using %s", self.hostname, method)
661 self._logger.debug("Routes for %s:\n\t%s", self.hostname, '\n\t'.join(rules))
663 self.apply_route_rules(rules, method)
665 self._configured_routes = set(routes)
666 self._configured_devs = tdevs
667 self._configured_method = method
669 def reconfigure_routes(self, routes, devs, vsys_vnet):
671 Updates the routes in the node's routing table to match
674 method = self._configured_method
676 dev_routes = self._annotate_routes_with_devs(routes, devs, method)
678 current = self._configured_routes
679 current_devs = self._configured_devs
681 new = set(dev_routes)
682 new_devs = set(map(operator.itemgetter(-1), dev_routes))
684 deletions = current - new
685 insertions = new - current
687 dev_deletions = current_devs - new_devs
688 dev_insertions = new_devs - current_devs
693 # Rule deletions first
694 for route in deletions:
695 route, dev = route[:-1], route[-1]
696 rules.append(self.format_route(route, dev, method, 'del'))
698 if method == 'sliceip':
700 rules.extend(map('disable '.__add__, dev_deletions))
703 rules.extend(map('enable '.__add__, dev_insertions))
705 # Rule insertions now
706 for route in insertions:
707 route, dev = route[:-1], dev[-1]
708 rules.append(self.format_route(route, dev, method, 'add'))
711 self.apply_route_rules(rules, method)
713 self._configured_routes = dev_routes
714 self._configured_devs = new_devs
716 def apply_route_rules(self, rules, method):
717 (out,err),proc = server.popen_ssh_command(
718 "( sudo -S bash -c 'cat /vsys/%(method)s.out >&2' & ) ; sudo -S bash -c 'cat > /vsys/%(method)s.in' ; sleep 0.5" % dict(
719 home = server.shell_escape(self.home_path),
721 host = self.hostname,
723 user = self.slicename,
725 ident_key = self.ident_path,
726 server_key = self.server_key,
727 stdin = '\n'.join(rules),
731 if proc.wait() or err:
732 raise RuntimeError, "Could not set routes (%s) errors: %s%s" % (rules,out,err)
734 logger.debug("%s said: %s%s", method, out, err)