src/emuvim/dcemulator/node.py

   1 """
   2 Distributed Cloud Emulator (dcemulator)
   3 (c) 2015 by Manuel Peuster <manuel.peuster@upb.de>
   4 """
   5 from mininet.node import Docker
   6 from mininet.link import Link
   7 import logging
   8 import time
   9 import json
  10
  11 LOG = logging.getLogger("dcemulator")
  12 LOG.setLevel(logging.DEBUG)
  13
  14
  15 DCDPID_BASE = 1000  # start of switch dpid's used for data center switches
  16
  17
  18 class EmulatorCompute(Docker):
  19     """
  20     Emulator specific compute node class.
  21     Inherits from Dockernet's Docker host class.
  22     Represents a single container connected to a (logical)
  23     data center.
  24     We can add emulator specific helper functions to it.
  25     """
  26
  27     def __init__(
  28             self, name, dimage, **kwargs):
  29         self.datacenter = kwargs.get("datacenter")  # pointer to current DC
  30         self.flavor_name = kwargs.get("flavor_name")
  31         LOG.debug("Starting compute instance %r in data center %r" % (name, str(self.datacenter)))
  32         # call original Docker.__init__
  33         Docker.__init__(self, name, dimage, **kwargs)
  34
  35     def getNetworkStatus(self):
  36         """
  37         Helper method to receive information about the virtual networks
  38         this compute instance is connected to.
  39         """
  40         # format list of tuples (name, Ip, MAC, isUp, status)
  41         return [(str(i), i.IP(), i.MAC(), i.isUp(), i.status())
  42                 for i in self.intfList()]
  43
  44     def getStatus(self):
  45         """
  46         Helper method to receive information about this compute instance.
  47         """
  48         status = {}
  49         status["name"] = self.name
  50         status["network"] = self.getNetworkStatus()
  51         status["image"] = self.dimage
  52         status["cpu_quota"] = self.cpu_quota
  53         status["cpu_period"] = self.cpu_period
  54         status["cpu_shares"] = self.cpu_shares
  55         status["cpuset"] = self.cpuset
  56         status["mem_limit"] = self.mem_limit
  57         status["memswap_limit"] = self.memswap_limit
  58         status["state"] = self.dcli.inspect_container(self.dc)["State"]
  59         status["id"] = self.dcli.inspect_container(self.dc)["Id"]
  60         status["datacenter"] = (None if self.datacenter is None
  61                                 else self.datacenter.label)
  62         return status
  63
  64
  65 class Datacenter(object):
  66     """
  67     Represents a logical data center to which compute resources
  68     (Docker containers) can be added at runtime.
  69
  70     Will also implement resource bookkeeping in later versions.
  71     """
  72
  73     DC_COUNTER = 1
  74
  75     def __init__(self, label, metadata={}, resource_log_path=None):
  76         self.net = None  # DCNetwork to which we belong
  77         # each node (DC) has a short internal name used by Mininet
  78         # this is caused by Mininets naming limitations for swtiches etc.
  79         self.name = "dc%d" % Datacenter.DC_COUNTER
  80         Datacenter.DC_COUNTER += 1
  81         # use this for user defined names that can be longer than self.name
  82         self.label = label
  83         # dict to store arbitrary metadata (e.g. latitude and longitude)
  84         self.metadata = metadata
  85         # path to which resource information should be logged (e.g. for experiments). None = no logging
  86         self.resource_log_path = resource_log_path
  87         # first prototype assumes one "bigswitch" per DC
  88         self.switch = None
  89         # keep track of running containers
  90         self.containers = {}
  91         # pointer to assigned resource model
  92         self._resource_model = None
  93
  94     def __repr__(self):
  95         return self.label
  96
  97     def _get_next_dc_dpid(self):
  98         global DCDPID_BASE
  99         DCDPID_BASE += 1
 100         return DCDPID_BASE
 101
 102     def create(self):
 103         """
 104         Each data center is represented by a single switch to which
 105         compute resources can be connected at run time.
 106
 107         TODO: This will be changed in the future to support multiple networks
 108         per data center
 109         """
 110         self.switch = self.net.addSwitch(
 111             "%s.s1" % self.name, dpid=hex(self._get_next_dc_dpid())[2:])
 112         LOG.debug("created data center switch: %s" % str(self.switch))
 113
 114     def start(self):
 115         pass
 116
 117     def startCompute(self, name, image=None, command=None, network=None, flavor_name="tiny"):
 118         """
 119         Create a new container as compute resource and connect it to this
 120         data center.
 121         :param name: name (string)
 122         :param image: image name (string)
 123         :param command: command (string)
 124         :param network: networks list({"ip": "10.0.0.254/8"}, {"ip": "11.0.0.254/24"})
 125         :param flavor_name: name of the flavor for this compute container
 126         :return:
 127         """
 128         assert name is not None
 129         # no duplications
 130         if name in [c.name for c in self.net.getAllContainers()]:
 131             raise Exception("Container with name %s already exists." % name)
 132         # set default parameter
 133         if image is None:
 134             image = "ubuntu"
 135         if network is None:
 136             network = {}  # {"ip": "10.0.0.254/8"}
 137         if isinstance(network, dict):
 138             network = [network]  # if we have only one network, put it in a list
 139         if isinstance(network, list):
 140             if len(network) < 1:
 141                 network.append({})
 142
 143         # allocate in resource resource model and compute resource limits for new container
 144         cpu_limit = mem_limit = disk_limit = -1
 145         cpu_period = cpu_quota = None
 146         if self._resource_model is not None:
 147             # call allocate in resource model to calculate resource limit for this container
 148             (cpu_limit, mem_limit, disk_limit) = alloc = self._resource_model.allocate(name, flavor_name)
 149             LOG.debug("Allocation result: %r" % str(alloc))
 150             # check if we have a cpu_limit given by the used resource model
 151             if cpu_limit > 0:
 152                 # calculate cpu period and quota for CFS
 153                 # (see: https://www.kernel.org/doc/Documentation/scheduler/sched-bwc.txt)
 154                 # TODO consider multi core machines etc! non trivial!
 155                 # Attention minimum cpu_quota is 1ms (micro)
 156                 cpu_period = 100000  # lets consider a fixed period of 100000 microseconds for now
 157                 cpu_quota = cpu_period * cpu_limit  # calculate the fraction of cpu time for this container
 158                 LOG.debug(
 159                     "CPU limit: cpu_quota = cpu_period * cpu_limit = %f * %f = %f" % (cpu_period, cpu_limit, cpu_quota))
 160                 # ATTENTION >= 1000 to avoid a invalid argument system error ... no idea why
 161                 if cpu_quota < 1000:
 162                     cpu_quota = 1000
 163                     LOG.warning("Increased CPU quota for %r to avoid system error." % name)
 164             # check if we have a mem_limit given by the used resource model
 165             if mem_limit > 0:
 166                 LOG.debug(
 167                     "MEM limit: mem_limit = %f MB" % mem_limit)
 168                 # ATTENTION minimum mem_limit per container is 4MB
 169                 if mem_limit < 4:
 170                     mem_limit = 4
 171                     LOG.warning("Increased MEM limit for %r because it was less than 4.0 MB." % name)
 172         # create the container
 173         d = self.net.addDocker(
 174             "%s" % (name),
 175             dimage=image,
 176             dcmd=command,
 177             datacenter=self,
 178             flavor_name=flavor_name,
 179             cpu_period=int(cpu_period) if cpu_limit > 0 else None,  # set cpu limits if needed
 180             cpu_quota=int(cpu_quota) if cpu_limit > 0 else None,
 181             mem_limit="%dm" % int(mem_limit) if mem_limit > 0 else None,  # set mem limits if needed
 182             memswap_limit="%dm" % int(mem_limit) if mem_limit > 0 else None  # lets set swap to mem limit for now
 183         )
 184         # connect all given networks
 185         for nw in network:
 186             # TODO we cannot use TCLink here (see: https://github.com/mpeuster/dockernet/issues/3)
 187             self.net.addLink(d, self.switch, params1=nw, cls=Link)
 188         # do bookkeeping
 189         self.containers[name] = d
 190
 191         # write resource log if a path is given
 192         if self.resource_log_path is not None:
 193             l = dict()
 194             l["t"] = time.time()
 195             l["name"] = name
 196             l["compute"] = d.getStatus()
 197             l["flavor_name"] = flavor_name
 198             l["action"] = "allocate"
 199             l["cpu_limit"] = cpu_limit
 200             l["mem_limit"] = mem_limit
 201             l["disk_limit"] = disk_limit
 202             l["rm_state"] = None if self._resource_model is None else self._resource_model.get_state_dict()
 203             # append to logfile
 204             with open(self.resource_log_path, "a") as f:
 205                 f.write("%s\n" % json.dumps(l))
 206         return d  # we might use UUIDs for naming later on
 207
 208     def stopCompute(self, name):
 209         """
 210         Stop and remove a container from this data center.
 211         """
 212         assert name is not None
 213         if name not in self.containers:
 214             raise Exception("Container with name %s not found." % name)
 215         LOG.debug("Stopping compute instance %r in data center %r" % (name, str(self)))
 216         self.net.removeLink(
 217             link=None, node1=self.containers[name], node2=self.switch)
 218         self.net.removeDocker("%s" % (name))
 219         del self.containers[name]
 220         # call resource model and free resources
 221         if self._resource_model is not None:
 222             self._resource_model.free(name)
 223
 224         # write resource log if a path is given
 225         if self.resource_log_path is not None:
 226             l = dict()
 227             l["t"] = time.time()
 228             l["name"] = name
 229             l["flavor_name"] = None
 230             l["action"] = "free"
 231             l["cpu_limit"] = -1
 232             l["mem_limit"] = -1
 233             l["disk_limit"] = -1
 234             l["rm_state"] = None if self._resource_model is None else self._resource_model.get_state_dict()
 235             # append to logfile
 236             with open(self.resource_log_path, "a") as f:
 237                 f.write("%s\n" % json.dumps(l))
 238         return True
 239
 240     def listCompute(self):
 241         """
 242         Return a list of all running containers assigned to this
 243         data center.
 244         """
 245         return list(self.containers.itervalues())
 246
 247     def getStatus(self):
 248         """
 249         Return a dict with status information about this DC.
 250         """
 251         return {
 252             "label": self.label,
 253             "internalname": self.name,
 254             "switch": self.switch.name,
 255             "n_running_containers": len(self.containers),
 256             "metadata": self.metadata
 257         }
 258
 259     def assignResourceModel(self, rm):
 260         """
 261         Assign a resource model to this DC.
 262         :param rm: a BaseResourceModel object
 263         :return:
 264         """
 265         if self._resource_model is not None:
 266             raise Exception("There is already an resource model assigned to this DC.")
 267         self._resource_model = rm
 268         self.net.rm_registrar.register(self, rm)
 269         LOG.info("Assigned RM: %r to DC: %r" % (rm, self))
 270