X-Git-Url: https://osm.etsi.org/gitweb/?a=blobdiff_plain;f=src%2Femuvim%2Fdcemulator%2Fnode.py;h=3f76df9c4779efd954e81603bd91a3468af7369d;hb=d2bbf4505674a8e0ea4f36f9576d53eba2b7af51;hp=336126c72c422a876ddc268b58639778940b355d;hpb=7f89ec08278873d335e15a9cf39034692c666bff;p=osm%2Fvim-emu.git diff --git a/src/emuvim/dcemulator/node.py b/src/emuvim/dcemulator/node.py index 336126c..3f76df9 100755 --- a/src/emuvim/dcemulator/node.py +++ b/src/emuvim/dcemulator/node.py @@ -3,7 +3,13 @@ Distributed Cloud Emulator (dcemulator) (c) 2015 by Manuel Peuster """ from mininet.node import Docker +from mininet.link import Link import logging +import time +import json + +LOG = logging.getLogger("dcemulator") +LOG.setLevel(logging.DEBUG) DCDPID_BASE = 1000 # start of switch dpid's used for data center switches @@ -20,9 +26,9 @@ class EmulatorCompute(Docker): def __init__( self, name, dimage, **kwargs): - logging.debug("Create EmulatorCompute instance: %s" % name) - self.datacenter = None # pointer to current DC - + self.datacenter = kwargs.get("datacenter") # pointer to current DC + self.flavor_name = kwargs.get("flavor_name") + LOG.debug("Starting compute instance %r in data center %r" % (name, str(self.datacenter))) # call original Docker.__init__ Docker.__init__(self, name, dimage, **kwargs) @@ -66,7 +72,7 @@ class Datacenter(object): DC_COUNTER = 1 - def __init__(self, label, metadata={}): + def __init__(self, label, metadata={}, resource_log_path=None): self.net = None # DCNetwork to which we belong # each node (DC) has a short internal name used by Mininet # this is caused by Mininets naming limitations for swtiches etc. @@ -76,8 +82,17 @@ class Datacenter(object): self.label = label # dict to store arbitrary metadata (e.g. latitude and longitude) self.metadata = metadata - self.switch = None # first prototype assumes one "bigswitch" per DC - self.containers = {} # keep track of running containers + # path to which resource information should be logged (e.g. for experiments). None = no logging + self.resource_log_path = resource_log_path + # first prototype assumes one "bigswitch" per DC + self.switch = None + # keep track of running containers + self.containers = {} + # pointer to assigned resource model + self._resource_model = None + + def __repr__(self): + return self.label def _get_next_dc_dpid(self): global DCDPID_BASE @@ -94,12 +109,12 @@ class Datacenter(object): """ self.switch = self.net.addSwitch( "%s.s1" % self.name, dpid=hex(self._get_next_dc_dpid())[2:]) - logging.debug("created data center switch: %s" % str(self.switch)) + LOG.debug("created data center switch: %s" % str(self.switch)) def start(self): pass - def startCompute(self, name, image=None, command=None, network=None): + def startCompute(self, name, image=None, command=None, network=None, flavor_name="tiny"): """ Create a new container as compute resource and connect it to this data center. @@ -107,6 +122,7 @@ class Datacenter(object): :param image: image name (string) :param command: command (string) :param network: networks list({"ip": "10.0.0.254/8"}, {"ip": "11.0.0.254/24"}) + :param flavor_name: name of the flavor for this compute container :return: """ assert name is not None @@ -124,14 +140,69 @@ class Datacenter(object): if len(network) < 1: network.append({}) + # allocate in resource resource model and compute resource limits for new container + cpu_limit = mem_limit = disk_limit = -1 + cpu_period = cpu_quota = None + if self._resource_model is not None: + # call allocate in resource model to calculate resource limit for this container + (cpu_limit, mem_limit, disk_limit) = alloc = self._resource_model.allocate(name, flavor_name) + LOG.debug("Allocation result: %r" % str(alloc)) + # check if we have a cpu_limit given by the used resource model + if cpu_limit > 0: + # calculate cpu period and quota for CFS + # (see: https://www.kernel.org/doc/Documentation/scheduler/sched-bwc.txt) + # TODO consider multi core machines etc! non trivial! + # Attention minimum cpu_quota is 1ms (micro) + cpu_period = 100000 # lets consider a fixed period of 100000 microseconds for now + cpu_quota = cpu_period * cpu_limit # calculate the fraction of cpu time for this container + LOG.debug( + "CPU limit: cpu_quota = cpu_period * cpu_limit = %f * %f = %f" % (cpu_period, cpu_limit, cpu_quota)) + # ATTENTION >= 1000 to avoid a invalid argument system error ... no idea why + if cpu_quota < 1000: + cpu_quota = 1000 + LOG.warning("Increased CPU quota for %r to avoid system error." % name) + # check if we have a mem_limit given by the used resource model + if mem_limit > 0: + LOG.debug( + "MEM limit: mem_limit = %f MB" % mem_limit) + # ATTENTION minimum mem_limit per container is 4MB + if mem_limit < 4: + mem_limit = 4 + LOG.warning("Increased MEM limit for %r because it was less than 4.0 MB." % name) # create the container - d = self.net.addDocker("%s" % (name), dimage=image, dcmd=command) + d = self.net.addDocker( + "%s" % (name), + dimage=image, + dcmd=command, + datacenter=self, + flavor_name=flavor_name, + cpu_period=int(cpu_period) if cpu_limit > 0 else None, # set cpu limits if needed + cpu_quota=int(cpu_quota) if cpu_limit > 0 else None, + mem_limit="%dm" % int(mem_limit) if mem_limit > 0 else None, # set mem limits if needed + memswap_limit="%dm" % int(mem_limit) if mem_limit > 0 else None # lets set swap to mem limit for now + ) # connect all given networks for nw in network: - self.net.addLink(d, self.switch, params1=nw) + # TODO we cannot use TCLink here (see: https://github.com/mpeuster/dockernet/issues/3) + self.net.addLink(d, self.switch, params1=nw, cls=Link) # do bookkeeping self.containers[name] = d - d.datacenter = self + + # write resource log if a path is given + if self.resource_log_path is not None: + l = dict() + l["t"] = time.time() + l["name"] = name + l["compute"] = d.getStatus() + l["flavor_name"] = flavor_name + l["action"] = "allocate" + l["cpu_limit"] = cpu_limit + l["mem_limit"] = mem_limit + l["disk_limit"] = disk_limit + l["rm_state"] = None if self._resource_model is None else self._resource_model.get_state_dict() + # append to logfile + with open(self.resource_log_path, "a") as f: + f.write("%s\n" % json.dumps(l)) return d # we might use UUIDs for naming later on def stopCompute(self, name): @@ -141,10 +212,29 @@ class Datacenter(object): assert name is not None if name not in self.containers: raise Exception("Container with name %s not found." % name) + LOG.debug("Stopping compute instance %r in data center %r" % (name, str(self))) self.net.removeLink( link=None, node1=self.containers[name], node2=self.switch) self.net.removeDocker("%s" % (name)) del self.containers[name] + # call resource model and free resources + if self._resource_model is not None: + self._resource_model.free(name) + + # write resource log if a path is given + if self.resource_log_path is not None: + l = dict() + l["t"] = time.time() + l["name"] = name + l["flavor_name"] = None + l["action"] = "free" + l["cpu_limit"] = -1 + l["mem_limit"] = -1 + l["disk_limit"] = -1 + l["rm_state"] = None if self._resource_model is None else self._resource_model.get_state_dict() + # append to logfile + with open(self.resource_log_path, "a") as f: + f.write("%s\n" % json.dumps(l)) return True def listCompute(self): @@ -165,3 +255,16 @@ class Datacenter(object): "n_running_containers": len(self.containers), "metadata": self.metadata } + + def assignResourceModel(self, rm): + """ + Assign a resource model to this DC. + :param rm: a BaseResourceModel object + :return: + """ + if self._resource_model is not None: + raise Exception("There is already an resource model assigned to this DC.") + self._resource_model = rm + self.net.rm_registrar.register(self, rm) + LOG.info("Assigned RM: %r to DC: %r" % (rm, self)) +