X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=sfa%2Fopenstack%2Fosaggregate.py;h=de3499556ed0e3a7bed5a846d160f55dd404b876;hb=1cc8e9613cab8b5b22478de369f259e591c54e6d;hp=7eb6e2721e81ee2a48822309b522f0ccd8baaff6;hpb=923e39ee1e69d239f42651b2f1b7243e6547ed65;p=sfa.git diff --git a/sfa/openstack/osaggregate.py b/sfa/openstack/osaggregate.py index 7eb6e272..de349955 100644 --- a/sfa/openstack/osaggregate.py +++ b/sfa/openstack/osaggregate.py @@ -3,11 +3,13 @@ import os import socket import base64 import string -import random +import random +import time from collections import defaultdict from nova.exception import ImageNotFound from nova.api.ec2.cloud import CloudController -from sfa.util.faults import SfaAPIError +from sfa.util.faults import SfaAPIError, SliverDoesNotExist +from sfa.util.sfatime import utcparse, datetime_to_string, datetime_to_epoch from sfa.rspecs.rspec import RSpec from sfa.rspecs.elements.hardware_type import HardwareType from sfa.rspecs.elements.node import Node @@ -16,11 +18,13 @@ from sfa.rspecs.elements.login import Login from sfa.rspecs.elements.disk_image import DiskImage from sfa.rspecs.elements.services import Services from sfa.rspecs.elements.interface import Interface +from sfa.rspecs.elements.fw_rule import FWRule from sfa.util.xrn import Xrn from sfa.planetlab.plxrn import PlXrn from sfa.openstack.osxrn import OSXrn, hrn_to_os_slicename from sfa.rspecs.version_manager import VersionManager from sfa.openstack.security_group import SecurityGroup +from sfa.server.threadmanager import ThreadManager from sfa.util.sfalogging import logger def pubkeys_to_user_data(pubkeys): @@ -33,20 +37,6 @@ def pubkeys_to_user_data(pubkeys): user_data += "\n" return user_data -def instance_to_sliver(instance, slice_xrn=None): - sliver_id = None - if slice_xrn: - xrn = Xrn(slice_xrn, 'slice') - sliver_id = xrn.get_sliver_id(instance.project_id, instance.hostname, instance.id) - - sliver = Sliver({'slice_id': sliver_id, - 'name': instance.name, - 'type': instance.name, - 'cpus': str(instance.vcpus), - 'memory': str(instance.ram), - 'storage': str(instance.disk)}) - return sliver - def image_to_rspec_disk_image(image): img = DiskImage() img['name'] = image['name'] @@ -60,84 +50,201 @@ class OSAggregate: def __init__(self, driver): self.driver = driver - def get_rspec(self, slice_xrn=None, version=None, options={}): - version_manager = VersionManager() - version = version_manager.get_version(version) - if not slice_xrn: - rspec_version = version_manager._get_version(version.type, version.version, 'ad') - nodes = self.get_aggregate_nodes() - else: - rspec_version = version_manager._get_version(version.type, version.version, 'manifest') - nodes = self.get_slice_nodes(slice_xrn) - rspec = RSpec(version=rspec_version, user_options=options) - rspec.version.add_nodes(nodes) - return rspec.toxml() - def get_availability_zones(self): - # essex release zones = self.driver.shell.nova_manager.dns_domains.domains() - if not zones: zones = ['cloud'] else: zones = [zone.name for zone in zones] return zones - def get_slice_nodes(self, slice_xrn): - zones = self.get_availability_zones() - name = hrn_to_os_slicename(slice_xrn) - instances = self.driver.shell.nova_manager.servers.findall(name=name) + def list_resources(self, version=None, options={}): + version_manager = VersionManager() + version = version_manager.get_version(version) + rspec_version = version_manager._get_version(version.type, version.version, 'ad') + rspec = RSpec(version=version, user_options=options) + nodes = self.get_aggregate_nodes() + rspec.version.add_nodes(nodes) + return rspec.toxml() + + def describe(self, urns, version=None, options={}): + # update nova connection + tenant_name = OSXrn(xrn=urns[0], type='slice').get_tenant_name() + self.driver.shell.nova_manager.connect(tenant=tenant_name) + instances = self.get_instances(urns) + # lookup the sliver allocations + sliver_ids = [sliver['sliver_id'] for sliver in slivers] + constraint = SliverAllocation.sliver_id.in_(sliver_ids) + sliver_allocations = dbsession.query(SliverAllocation).filter(constraint) + sliver_allocation_dict = {} + for sliver_allocation in sliver_allocations: + sliver_allocation_dict[sliver_allocation.sliver_id] = sliver_allocation + + geni_slivers = [] rspec_nodes = [] for instance in instances: - rspec_node = Node() - - #TODO: find a way to look up an instances availability zone in essex - #if instance.availability_zone: - # node_xrn = OSXrn(instance.availability_zone, 'node') - #else: - # node_xrn = OSXrn('cloud', 'node') - node_xrn = instance.metadata.get('component_id') - if not node_xrn: - node_xrn = OSXrn('cloud', 'node') - else: - node_xrn = OSXrn(node_xrn, 'node') + rspec_nodes.append(self.instance_to_rspec_node(instance)) + geni_sliver = self.instance_to_geni_sliver(instance, sliver_sllocation_dict) + geni_slivers.append(geni_sliver) + version_manager = VersionManager() + version = version_manager.get_version(version) + rspec_version = version_manager._get_version(version.type, version.version, 'manifest') + rspec = RSpec(version=rspec_version, user_options=options) + rspec.xml.set('expires', datetime_to_string(utcparse(time.time()))) + rspec.version.add_nodes(rspec_nodes) + result = {'geni_urn': Xrn(urns[0]).get_urn(), + 'geni_rspec': rspec.toxml(), + 'geni_slivers': geni_slivers} + + return result - rspec_node['component_id'] = node_xrn.urn - rspec_node['component_name'] = node_xrn.name - rspec_node['component_manager_id'] = Xrn(self.driver.hrn, 'authority+cm').get_urn() - flavor = self.driver.shell.nova_manager.flavors.find(id=instance.flavor['id']) - sliver = instance_to_sliver(flavor) - rspec_node['slivers'] = [sliver] - image = self.driver.shell.image_manager.get_images(id=instance.image['id']) - if isinstance(image, list) and len(image) > 0: - image = image[0] - disk_image = image_to_rspec_disk_image(image) - sliver['disk_image'] = [disk_image] - - # build interfaces - interfaces = [] - addresses = instance.addresses - for private_ip in addresses.get('private', []): - if_xrn = PlXrn(auth=self.driver.hrn, - interface='node%s:eth0' % (instance.hostId)) - interface = Interface({'component_id': if_xrn.urn}) - interface['ips'] = [{'address': private_ip['addr'], - #'netmask': private_ip['network'], - 'type': private_ip['version']}] - interfaces.append(interface) - rspec_node['interfaces'] = interfaces - - # slivers always provide the ssh service - rspec_node['services'] = [] - for public_ip in addresses.get('public', []): - login = Login({'authentication': 'ssh-keys', - 'hostname': public_ip['addr'], - 'port':'22', 'username': 'root'}) - service = Services({'login': login}) - rspec_node['services'].append(service) - rspec_nodes.append(rspec_node) - return rspec_nodes + def get_instances(self, urns): + # parse slice names and sliver ids + names = set() + ids = set() + for urn in urns: + xrn = OSXrn(xrn=urn) + if xrn.type == 'slice': + names.add(xrn.get_slice_name()) + elif xrn.type == 'sliver': + ids.add(xrn.leaf) + + # look up instances + instances = [] + filter = {} + if names: + filter['name'] = names + if ids: + filter['id'] = ids + servers = self.driver.shell.nova_manager.servers.findall(**filter) + instances.extend(servers) + + return instances + def instance_to_rspec_node(self, instance): + # determine node urn + node_xrn = instance.metadata.get('component_id') + if not node_xrn: + node_xrn = OSXrn('cloud', type='node') + else: + node_xrn = OSXrn(xrn=node_xrn, type='node') + + rspec_node = Node() + rspec_node['component_id'] = node_xrn.urn + rspec_node['component_name'] = node_xrn.name + rspec_node['component_manager_id'] = Xrn(self.driver.hrn, 'authority+cm').get_urn() + rspec_node['sliver_id'] = OSXrn(name=instance.name, type='slice', id=instance.id).get_urn() + if instance.metadata.get('client_id'): + rspec_node['client_id'] = instance.metadata.get('client_id') + + # get sliver details + flavor = self.driver.shell.nova_manager.flavors.find(id=instance.flavor['id']) + sliver = self.instance_to_sliver(flavor) + # get firewall rules + fw_rules = [] + group_name = instance.metadata.get('security_groups') + if group_name: + group = self.driver.shell.nova_manager.security_groups.find(name=group_name) + for rule in group.rules: + port_range ="%s:%s" % (rule['from_port'], rule['to_port']) + fw_rule = FWRule({'protocol': rule['ip_protocol'], + 'port_range': port_range, + 'cidr_ip': rule['ip_range']['cidr']}) + fw_rules.append(fw_rule) + sliver['fw_rules'] = fw_rules + rspec_node['slivers'] = [sliver] + + # get disk image + image = self.driver.shell.image_manager.get_images(id=instance.image['id']) + if isinstance(image, list) and len(image) > 0: + image = image[0] + disk_image = image_to_rspec_disk_image(image) + sliver['disk_image'] = [disk_image] + + # get interfaces + rspec_node['services'] = [] + rspec_node['interfaces'] = [] + addresses = instance.addresses + # HACK: public ips are stored in the list of private, but + # this seems wrong. Assume pub ip is the last in the list of + # private ips until openstack bug is fixed. + if addresses.get('private'): + login = Login({'authentication': 'ssh-keys', + 'hostname': addresses.get('private')[-1]['addr'], + 'port':'22', 'username': 'root'}) + service = Services({'login': login}) + rspec_node['services'].append(service) + + for private_ip in addresses.get('private', []): + if_xrn = PlXrn(auth=self.driver.hrn, + interface='node%s' % (instance.hostId)) + if_client_id = Xrn(if_xrn.urn, type='interface', id="eth%s" %if_index).urn + if_sliver_id = Xrn(rspec_node['sliver_id'], type='slice', id="eth%s" %if_index).urn + interface = Interface({'component_id': if_xrn.urn, + 'client_id': if_client_id, + 'sliver_id': if_sliver_id}) + interface['ips'] = [{'address': private_ip['addr'], + #'netmask': private_ip['network'], + 'type': private_ip['version']}] + rspec_node['interfaces'].append(interface) + + # slivers always provide the ssh service + for public_ip in addresses.get('public', []): + login = Login({'authentication': 'ssh-keys', + 'hostname': public_ip['addr'], + 'port':'22', 'username': 'root'}) + service = Services({'login': login}) + rspec_node['services'].append(service) + return rspec_node + + + def instance_to_sliver(self, instance, xrn=None): + if xrn: + sliver_hrn = '%s.%s' % (self.driver.hrn, instance.id) + sliver_id = Xrn(sliver_hrn, type='sliver').urn + + sliver = Sliver({'sliver_id': sliver_id, + 'name': instance.name, + 'type': instance.name, + 'cpus': str(instance.vcpus), + 'memory': str(instance.ram), + 'storage': str(instance.disk)}) + return sliver + + def instance_to_geni_sliver(self, instance, sliver_allocations = {}): + sliver_hrn = '%s.%s' % (self.driver.hrn, instance.id) + sliver_id = Xrn(sliver_hrn, type='sliver').urn + + # set sliver allocation and operational status + sliver_allocation = sliver_allocations[sliver_id] + if sliver_allocation: + allocation_status = sliver_allocation.allocation_state + if allocation_status == 'geni_allocated': + op_status = 'geni_pending_allocation' + elif allocation_status == 'geni_provisioned': + state = instance.state.lower() + if state == 'active': + op_status = 'geni_ready' + elif state == 'building': + op_status = 'geni_notready' + elif state == 'failed': + op_status =' geni_failed' + else: + op_status = 'geni_unknown' + else: + allocation_status = 'geni_unallocated' + # required fields + geni_sliver = {'geni_sliver_urn': sliver_id, + 'geni_expires': None, + 'geni_allocation_status': allocation_status, + 'geni_operational_status': op_status, + 'geni_error': None, + 'plos_created_at': datetime_to_string(utcparse(instance.created)), + 'plos_sliver_type': self.shell.nova_manager.flavors.find(id=instance.flavor['id']).name, + } + + return geni_sliver + def get_aggregate_nodes(self): zones = self.get_availability_zones() # available sliver/instance/vm types @@ -159,18 +266,28 @@ class OSAggregate: HardwareType({'name': 'pc'})] slivers = [] for instance in instances: - sliver = instance_to_sliver(instance) + sliver = self.instance_to_sliver(instance) sliver['disk_image'] = disk_images slivers.append(sliver) - + rspec_node['available'] = 'true' rspec_node['slivers'] = slivers rspec_nodes.append(rspec_node) return rspec_nodes - + def create_tenant(self, tenant_name): + tenants = self.driver.shell.auth_manager.tenants.findall(name=tenant_name) + if not tenants: + self.driver.shell.auth_manager.tenants.create(tenant_name, tenant_name) + tenant = self.driver.shell.auth_manager.tenants.find(name=tenant_name) + else: + tenant = tenants[0] + return tenant + def create_instance_key(self, slice_hrn, user): - key_name = "%s:%s" (slice_name, Xrn(user['urn']).get_hrn()) + slice_name = Xrn(slice_hrn).leaf + user_name = Xrn(user['urn']).leaf + key_name = "%s_%s" % (slice_name, user_name) pubkey = user['keys'][0] key_found = False existing_keys = self.driver.shell.nova_manager.keypairs.findall(name=key_name) @@ -181,7 +298,7 @@ class OSAggregate: key_found = True if not key_found: - self.driver.shll.nova_manager.keypairs.create(key_name, pubkey) + self.driver.shell.nova_manager.keypairs.create(key_name, pubkey) return key_name @@ -203,6 +320,11 @@ class OSAggregate: cidr_ip = rule.get('cidr_ip'), port_range = rule.get('port_range'), icmp_type_code = rule.get('icmp_type_code')) + # Open ICMP by default + security_group.add_rule_to_group(group_name, + protocol = "icmp", + cidr_ip = "0.0.0.0/0", + icmp_type_code = "-1:-1") return group_name def add_rule_to_security_group(self, group_name, **kwds): @@ -214,62 +336,120 @@ class OSAggregate: - def run_instances(self, slicename, rspec, key_name, pubkeys): + def run_instances(self, instance_name, tenant_name, rspec, key_name, pubkeys): #logger.debug('Reserving an instance: image: %s, flavor: ' \ # '%s, key: %s, name: %s' % \ # (image_id, flavor_id, key_name, slicename)) + # make sure a tenant exists for this slice + tenant = self.create_tenant(tenant_name) + + # add the sfa admin user to this tenant and update our nova client connection + # to use these credentials for the rest of this session. This emsures that the instances + # we create will be assigned to the correct tenant. + sfa_admin_user = self.driver.shell.auth_manager.users.find(name=self.driver.shell.auth_manager.opts['OS_USERNAME']) + user_role = self.driver.shell.auth_manager.roles.find(name='user') + admin_role = self.driver.shell.auth_manager.roles.find(name='admin') + self.driver.shell.auth_manager.roles.add_user_role(sfa_admin_user, admin_role, tenant) + self.driver.shell.auth_manager.roles.add_user_role(sfa_admin_user, user_role, tenant) + self.driver.shell.nova_manager.connect(tenant=tenant.name) + authorized_keys = "\n".join(pubkeys) files = {'/root/.ssh/authorized_keys': authorized_keys} rspec = RSpec(rspec) requested_instances = defaultdict(list) + # iterate over clouds/zones/nodes + slivers = [] for node in rspec.version.get_nodes_with_slivers(): instances = node.get('slivers', []) if not instances: continue for instance in instances: - metadata = {} - flavor_id = self.driver.shell.nova_manager.flavors.find(name=instance['name']) - image = instance.get('disk_image') - if image and isinstance(image, list): - image = image[0] - image_id = self.driver.shell.nova_manager.images.find(name=image['name']) - fw_rules = instance.get('fw_rules', []) - group_name = self.create_security_group(slicename, fw_rules) - metadata['security_groups'] = group_name - metadata['component_id'] = node['component_id'] try: - self.driver.shell.nova_manager.servers.create(flavor=flavor_id, + metadata = {} + flavor_id = self.driver.shell.nova_manager.flavors.find(name=instance['name']) + image = instance.get('disk_image') + if image and isinstance(image, list): + image = image[0] + else: + raise InvalidRSpec("Must specify a disk_image for each VM") + image_id = self.driver.shell.nova_manager.images.find(name=image['name']) + fw_rules = instance.get('fw_rules', []) + group_name = self.create_security_group(instance_name, fw_rules) + metadata['security_groups'] = group_name + if node.get('component_id'): + metadata['component_id'] = node['component_id'] + if node.get('client_id'): + metadata['client_id'] = node['client_id'] + server = self.driver.shell.nova_manager.servers.create( + flavor=flavor_id, image=image_id, key_name = key_name, - security_group = group_name, + security_groups = [group_name], files=files, meta=metadata, - name=slicename) + name=instance_name) + slivers.append(server) except Exception, err: logger.log_exc(err) + return slivers + def delete_instance(self, instance): + + def _delete_security_group(inst): + security_group = inst.metadata.get('security_groups', '') + if security_group: + manager = SecurityGroup(self.driver) + timeout = 10.0 # wait a maximum of 10 seconds before forcing the security group delete + start_time = time.time() + instance_deleted = False + while instance_deleted == False and (time.time() - start_time) < timeout: + tmp_inst = self.driver.shell.nova_manager.servers.findall(id=inst.id) + if not tmp_inst: + instance_deleted = True + time.sleep(.5) + manager.delete_security_group(security_group) - def delete_instances(self, instance_name): - instances = self.driver.shell.nova_manager.servers.findall(name=instance_name) + thread_manager = ThreadManager() + tenant = self.driver.shell.auth_manager.tenants.find(id=instance.tenant_id) + self.driver.shell.nova_manager.connect(tenant=tenant.name) + args = {'name': instance.name, + 'id': instance.id} + instances = self.driver.shell.nova_manager.servers.findall(**args) security_group_manager = SecurityGroup(self.driver) for instance in instances: - # deleate this instance's security groups - for security_group in instance.metadata.get('security_groups', []): - # dont delete the default security group - if security_group != 'default': - security_group_manager.delete_security_group(security_group) # destroy instance self.driver.shell.nova_manager.servers.delete(instance) + # deleate this instance's security groups + thread_manager.run(_delete_security_group, instance) return 1 - def stop_instances(self, instance_name): - instances = self.driver.shell.nova_manager.servers.findall(name=instance_name) + def stop_instances(self, instance_name, tenant_name, id=None): + self.driver.shell.nova_manager.connect(tenant=tenant_name) + args = {'name': instance_name} + if id: + args['id'] = id + instances = self.driver.shell.nova_manager.servers.findall(**args) for instance in instances: self.driver.shell.nova_manager.servers.pause(instance) return 1 + def start_instances(self, instance_name, tenant_name, id=None): + self.driver.shell.nova_manager.connect(tenant=tenant_name) + args = {'name': instance_name} + if id: + args['id'] = id + instances = self.driver.shell.nova_manager.servers.findall(**args) + for instance in instances: + self.driver.shell.nova_manager.servers.resume(instance) + return 1 + + def restart_instances(self, instacne_name, tenant_name, id=None): + self.stop_instances(instance_name, tenant_name, id) + self.start_instances(instance_name, tenant_name, id) + return 1 + def update_instances(self, project_name): pass