node.py 10 KB
Newer Older
1
from __future__ import absolute_import, unicode_literals
Őry Máté committed
2 3 4
from logging import getLogger

from django.db.models import (
5
    CharField, IntegerField, ForeignKey, BooleanField, ManyToManyField,
6
    FloatField, permalink,
Őry Máté committed
7 8 9 10 11 12 13
)
from django.utils.translation import ugettext_lazy as _

from celery.exceptions import TimeoutError
from model_utils.models import TimeStampedModel
from taggit.managers import TaggableManager

14
from common.models import method_cache, WorkerNotFound
Őry Máté committed
15 16
from firewall.models import Host
from ..tasks import vm_tasks
17
from .common import Trait
Őry Máté committed
18

19
from .activity import node_activity, NodeActivity
20

Gregory Nagy committed
21 22
from monitor.calvin.calvin import Query
from monitor.calvin.calvin import GraphiteHandler
23
from django.utils import timezone
24

Őry Máté committed
25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54
logger = getLogger(__name__)


class Node(TimeStampedModel):

    """A VM host machine, a hypervisor.
    """
    name = CharField(max_length=50, unique=True,
                     verbose_name=_('name'),
                     help_text=_('Human readable name of node.'))
    priority = IntegerField(verbose_name=_('priority'),
                            help_text=_('Node usage priority.'))
    host = ForeignKey(Host, verbose_name=_('host'),
                      help_text=_('Host in firewall.'))
    enabled = BooleanField(verbose_name=_('enabled'), default=False,
                           help_text=_('Indicates whether the node can '
                                       'be used for hosting.'))
    traits = ManyToManyField(Trait, blank=True,
                             help_text=_("Declared traits."),
                             verbose_name=_('traits'))
    tags = TaggableManager(blank=True, verbose_name=_("tags"))
    overcommit = FloatField(default=1.0, verbose_name=_("overcommit ratio"),
                            help_text=_("The ratio of total memory with "
                                        "to without overcommit."))

    class Meta:
        app_label = 'vm'
        db_table = 'vm_node'
        permissions = ()

Őry Máté committed
55 56 57
    def __unicode__(self):
        return self.name

Őry Máté committed
58
    @method_cache(10, 5)
59
    def get_online(self):
60
        """Check if the node is online.
Őry Máté committed
61

62 63 64 65 66 67
        Runs a remote ping task if the worker is running.
        """
        try:
            return self.remote_query(vm_tasks.ping, timeout=1, default=False)
        except WorkerNotFound:
            return False
Őry Máté committed
68

69 70
    online = property(get_online)

Őry Máté committed
71
    @method_cache(300)
72
    def get_num_cores(self):
Őry Máté committed
73 74
        """Number of CPU threads available to the virtual machines.
        """
75

76
        return self.remote_query(vm_tasks.get_core_num, default=0)
77

78 79
    num_cores = property(get_num_cores)

80 81
    @property
    def state(self):
82
        """The state combined of online and enabled attributes.
83
        """
84
        if self.enabled and self.online:
85
            return 'ONLINE'
86
        elif self.enabled and not self.online:
87
            return 'MISSING'
88
        elif not self.enabled and self.online:
89
            return 'DISABLED'
90
        else:
91
            return 'OFFLINE'
92 93 94

    def disable(self, user=None):
        ''' Disable the node.'''
95 96 97 98
        if self.enabled is True:
            with node_activity(code_suffix='disable', node=self, user=user):
                self.enabled = False
                self.save()
99 100 101

    def enable(self, user=None):
        ''' Enable the node. '''
102 103 104 105 106 107
        if self.enabled is not True:
            with node_activity(code_suffix='enable', node=self, user=user):
                self.enabled = True
                self.save()
            self.get_num_cores(invalidate_cache=True)
            self.get_ram_size(invalidate_cache=True)
Őry Máté committed
108 109

    @method_cache(300)
110
    def get_ram_size(self):
Őry Máté committed
111 112
        """Bytes of total memory in the node.
        """
113
        return self.remote_query(vm_tasks.get_ram_size, default=0)
Őry Máté committed
114

115 116
    ram_size = property(get_ram_size)

Őry Máté committed
117 118 119 120 121 122
    @property
    def ram_size_with_overcommit(self):
        """Bytes of total memory including overcommit margin.
        """
        return self.ram_size * self.overcommit

123
    @method_cache(30)
Őry Máté committed
124
    def get_remote_queue_name(self, queue_id):
125 126
        """Return the name of the remote celery queue for this node.

127
        throws Exception if there is no worker on the queue.
128
        Until the cache provide reult there can be dead queues.
129
        """
130

131
        if vm_tasks.check_queue(self.host.hostname, queue_id):
132
            self.node_online()
133 134
            return self.host.hostname + "." + queue_id
        else:
135 136
            if self.enabled is True:
                self.node_offline()
137
            raise WorkerNotFound()
Őry Máté committed
138

139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183
    def node_online(self):
        """Create activity and log entry when node reappears.
        """

        try:
            act = self.activity_log.order_by('-pk')[0]
        except IndexError:
            pass  # no monitoring activity at all
        else:
            logger.debug("The last activity was %s" % act)
            if act.activity_code.endswith("offline"):
                act = NodeActivity.create(code_suffix='monitor_succes_online',
                                          node=self, user=None)
                act.started = timezone.now()
                act.finished = timezone.now()
                act.succeeded = True
                act.save()
                logger.info("Node %s is ONLINE." % self.name)
                self.get_num_cores(invalidate_cache=True)
                self.get_ram_size(invalidate_cache=True)

    def node_offline(self):
        """Called when a node disappears.

        If the node is not already offline, record an activity and a log entry.
        """

        try:
            act = self.activity_log.order_by('-pk')[0]
        except IndexError:
            pass  # no activity at all
        else:
            logger.debug("The last activity was %s" % act)
            if act.activity_code.endswith("offline"):
                return
        act = NodeActivity.create(code_suffix='monitor_failed_offline',
                                  node=self, user=None)
        act.started = timezone.now()
        act.finished = timezone.now()
        act.succeeded = False
        act.save()
        logger.critical("Node %s is OFFLINE%s.", self.name,
                        ", but enabled" if self.enabled else "")
        # TODO: check if we should reschedule any VMs?

184
    def remote_query(self, task, timeout=30, raise_=False, default=None):
Őry Máté committed
185 186
        """Query the given task, and get the result.

187 188 189 190
        If the result is not ready or worker not reachable
        in timeout secs, return default value or raise a
        TimeoutError or WorkerNotFound exception.
        """
Őry Máté committed
191
        try:
192 193
            r = task.apply_async(
                queue=self.get_remote_queue_name('vm'), expires=timeout + 60)
Őry Máté committed
194
            return r.get(timeout=timeout)
195
        except (TimeoutError, WorkerNotFound):
Őry Máté committed
196 197 198 199 200
            if raise_:
                raise
            else:
                return default

201
    def get_monitor_info(self):
202 203 204 205
        collected = {}
        try:
            handler = GraphiteHandler()
        except:
Gregory Nagy committed
206
            response = self.remote_query(vm_tasks.get_node_metrics, 30)
207 208
            collected['cpu.usage'] = response['cpu.usage']
            collected['memory.usage'] = response['memory.usage']
209
            return collected
210
        query = Query()
Gregory Nagy committed
211 212 213
        query.set_target(self.host.hostname + ".circle")
        query.set_format("json")
        query.set_relative_start(5, "minutes")
214
        metrics = ["cpu.usage", "memory.usage"]
215 216
        collected = {}
        for metric in metrics:
Gregory Nagy committed
217
            query.set_metric(metric)
218 219 220 221
            query.generate()
            handler.put(query)
            handler.send()
        for metric in metrics:
Gregory Nagy committed
222
            response = handler.pop()
223 224 225 226 227 228 229
            length = len(response[0]["datapoints"])
            cache = response[0]["datapoints"][length - 1][0]
            if cache is None:
                cache = 0
            collected[metric] = cache
        return collected

230
    @property
231
    def cpu_usage(self):
232
        return float(self.get_monitor_info()["cpu.usage"]) / 100
233

234
    @property
235
    def ram_usage(self):
236
        return float(self.get_monitor_info()["memory.usage"]) / 100
237

238 239 240 241
    @property
    def byte_ram_usage(self):
        return self.ram_usage * self.ram_size

Őry Máté committed
242
    def update_vm_states(self):
243 244 245 246 247
        """Update state of Instances running on this Node.

        Query state of all libvirt domains, and notify Instances by their
        vm_state_changed hook.
        """
Őry Máté committed
248
        domains = {}
249 250 251 252 253
        domain_list = self.remote_query(vm_tasks.list_domains_info, timeout=5)
        if domain_list is None:
            logger.info("Monitoring failed at: %s", self.name)
            return
        for i in domain_list:
Őry Máté committed
254 255 256 257 258 259 260 261
            # [{'name': 'cloud-1234', 'state': 'RUNNING', ...}, ...]
            try:
                id = int(i['name'].split('-')[1])
            except:
                pass  # name format doesn't match
            else:
                domains[id] = i['state']

262 263
        instances = [{'id': i.id, 'state': i.state}
                     for i in self.instance_set.order_by('id').all()]
Őry Máté committed
264 265 266 267 268 269
        for i in instances:
            try:
                d = domains[i['id']]
            except KeyError:
                logger.info('Node %s update: instance %s missing from '
                            'libvirt', self, i['id'])
270 271
                # Set state to STOPPED when instance is missing
                self.instance_set.get(id=i['id']).vm_state_changed('STOPPED')
Őry Máté committed
272 273 274 275 276
            else:
                if d != i['state']:
                    logger.info('Node %s update: instance %s state changed '
                                '(libvirt: %s, db: %s)',
                                self, i['id'], d, i['state'])
277
                    self.instance_set.get(id=i['id']).vm_state_changed(d)
Őry Máté committed
278 279 280 281 282

                del domains[i['id']]
        for i in domains.keys():
            logger.info('Node %s update: domain %s in libvirt but not in db.',
                        self, i)
283 284 285

    @classmethod
    def get_state_count(cls, online, enabled):
286 287
        return len([1 for i in cls.objects.filter(enabled=enabled).all()
                    if i.online == online])
288 289 290 291

    @permalink
    def get_absolute_url(self):
        return ('dashboard.views.node-detail', None, {'pk': self.id})