[Notes] [Git][BuildGrid/buildgrid][finn/48-cancellation-leases] Working commit.



Title: GitLab

finn pushed to branch finn/48-cancellation-leases at BuildGrid / buildgrid

Commits:

13 changed files:

Changes:

  • buildgrid/_app/bots/dummy.py
    ... ... @@ -46,4 +46,7 @@ def work_dummy(context, lease):
    46 46
     
    
    47 47
         lease.result.Pack(action_result)
    
    48 48
     
    
    49
    +    # while True:
    
    50
    +    #    pass
    
    51
    +
    
    49 52
         return lease

  • buildgrid/_app/commands/cmd_bot.py
    ... ... @@ -28,8 +28,11 @@ from urllib.parse import urlparse
    28 28
     import click
    
    29 29
     import grpc
    
    30 30
     
    
    31
    -from buildgrid.bot import bot, bot_interface
    
    32
    -from buildgrid.bot.bot_session import BotSession, Device, Worker
    
    31
    +from buildgrid.bot import bot, interface, session
    
    32
    +from buildgrid.bot.hardware.interface import HardwareInterface
    
    33
    +from buildgrid.bot.hardware.device import Device
    
    34
    +from buildgrid.bot.hardware.worker import Worker
    
    35
    +
    
    33 36
     
    
    34 37
     from ..bots import buildbox, dummy, host
    
    35 38
     from ..cli import pass_context
    
    ... ... @@ -123,13 +126,14 @@ def cli(context, parent, update_period, remote, client_key, client_cert, server_
    123 126
         context.logger = logging.getLogger(__name__)
    
    124 127
         context.logger.debug("Starting for remote {}".format(context.remote))
    
    125 128
     
    
    126
    -    interface = bot_interface.BotInterface(context.channel)
    
    129
    +    bot_interface = interface.BotInterface(context.channel)
    
    127 130
     
    
    128 131
         worker = Worker()
    
    129 132
         worker.add_device(Device())
    
    130 133
     
    
    131
    -    bot_session = BotSession(parent, interface)
    
    132
    -    bot_session.add_worker(worker)
    
    134
    +    hardware_interface = HardwareInterface(worker)
    
    135
    +
    
    136
    +    bot_session = session.BotSession(parent, bot_interface, hardware_interface)
    
    133 137
     
    
    134 138
         context.bot_session = bot_session
    
    135 139
     
    
    ... ... @@ -142,8 +146,7 @@ def run_dummy(context):
    142 146
         """
    
    143 147
         try:
    
    144 148
             b = bot.Bot(context.bot_session, context.update_period)
    
    145
    -        b.session(dummy.work_dummy,
    
    146
    -                  context)
    
    149
    +        b.session(dummy.work_dummy, context)
    
    147 150
         except KeyboardInterrupt:
    
    148 151
             pass
    
    149 152
     
    

  • buildgrid/bot/bot.py
    ... ... @@ -17,11 +17,12 @@
    17 17
     Bot
    
    18 18
     ====
    
    19 19
     
    
    20
    -Creates a bot session.
    
    20
    +Creates a bot session and sends updates to the server.
    
    21 21
     """
    
    22 22
     
    
    23 23
     import asyncio
    
    24 24
     import logging
    
    25
    +import sys
    
    25 26
     
    
    26 27
     
    
    27 28
     class Bot:
    
    ... ... @@ -45,6 +46,7 @@ class Bot:
    45 46
                 loop.run_forever()
    
    46 47
             except KeyboardInterrupt:
    
    47 48
                 pass
    
    49
    +
    
    48 50
             finally:
    
    49 51
                 task.cancel()
    
    50 52
                 loop.close()
    
    ... ... @@ -54,5 +56,11 @@ class Bot:
    54 56
             Calls the server periodically to inform the server the client has not died.
    
    55 57
             """
    
    56 58
             while True:
    
    57
    -            self._bot_session.update_bot_session()
    
    59
    +            try:
    
    60
    +                self._bot_session.update_bot_session()
    
    61
    +
    
    62
    +            except Exception as e:
    
    63
    +                self.logger.error("Error: [{}]".format(e))
    
    64
    +                raise
    
    65
    +
    
    58 66
                 await asyncio.sleep(self._update_period)

  • buildgrid/bot/hardware/device.py
    ... ... @@ -13,6 +13,17 @@
    13 13
     # limitations under the License.
    
    14 14
     
    
    15 15
     
    
    16
    +"""
    
    17
    +Device
    
    18
    +======
    
    19
    +
    
    20
    +A device.
    
    21
    +"""
    
    22
    +
    
    23
    +
    
    24
    +import uuid
    
    25
    +from buildgrid._protos.google.devtools.remoteworkers.v1test2 import worker_pb2
    
    26
    +
    
    16 27
     class Device:
    
    17 28
     
    
    18 29
         def __init__(self, properties=None):
    
    ... ... @@ -21,23 +32,18 @@ class Device:
    21 32
             is running a bit and responsible to actually executing commands.
    
    22 33
             All other devices are known as Attatched Devices and must be controlled
    
    23 34
             by the Primary Device.
    
    35
    +
    
    36
    +        properties (list(dict(string : string))) : Properties of device. Keys may
    
    37
    +        repeated.
    
    24 38
             """
    
    25 39
     
    
    26
    -        self.__properties = {}
    
    40
    +        self._properties = {}
    
    41
    +        self.__property_keys = ['os', 'has-docker']
    
    27 42
             self.__name = str(uuid.uuid4())
    
    28 43
     
    
    29 44
             if properties:
    
    30
    -            for k, v in properties.items():
    
    31
    -                if k == 'os':
    
    32
    -                    self.__properties[k] = v
    
    33
    -
    
    34
    -                elif k == 'docker':
    
    35
    -                    if v not in ('True', 'False'):
    
    36
    -                        raise ValueError('Value not supported: [{}]'.format(v))
    
    37
    -                    self.__properties[k] = v
    
    38
    -
    
    39
    -                else:
    
    40
    -                    raise KeyError('Key not supported: [{}]'.format(k))
    
    45
    +            for prop in properties:
    
    46
    +                self._add_property(prop)
    
    41 47
     
    
    42 48
         @property
    
    43 49
         def name(self):
    
    ... ... @@ -45,13 +51,25 @@ class Device:
    45 51
     
    
    46 52
         @property
    
    47 53
         def properties(self):
    
    48
    -        return self.__properties
    
    54
    +        return self._properties
    
    49 55
     
    
    50 56
         def get_pb2(self):
    
    51
    -        device = worker_pb2.Device(handle=self._name)
    
    52
    -        property_message = worker_pb2.Device.Property()
    
    57
    +        device = worker_pb2.Device(handle=self.__name)
    
    53 58
             for k, v in self._properties.items():
    
    54
    -            property_message.key = k
    
    55
    -            property_message.value = v
    
    56
    -            device.properties.extend([property_message])
    
    59
    +            for prop in v:
    
    60
    +                property_message = worker_pb2.Device.Property()
    
    61
    +                property_message.key = k
    
    62
    +                property_message.value = prop
    
    63
    +                device.properties.extend([property_message])
    
    57 64
             return device
    
    65
    +
    
    66
    +    def _add_property(self, key, value):
    
    67
    +        if key in self.__property_keys:
    
    68
    +            prop = self._properties.get(key)
    
    69
    +            if not prop:
    
    70
    +                self._properties[key] = [value]
    
    71
    +            else:
    
    72
    +                prop[key].append(value)
    
    73
    +
    
    74
    +        else:
    
    75
    +            raise KeyError('Key not supported: [{}]'.format(key))

  • buildgrid/bot/hardware/interface.py
    1
    +# Copyright (C) 2018 Bloomberg LP
    
    2
    +#
    
    3
    +# Licensed under the Apache License, Version 2.0 (the "License");
    
    4
    +# you may not use this file except in compliance with the License.
    
    5
    +# You may obtain a copy of the License at
    
    6
    +#
    
    7
    +#  <http://www.apache.org/licenses/LICENSE-2.0>
    
    8
    +#
    
    9
    +# Unless required by applicable law or agreed to in writing, software
    
    10
    +# distributed under the License is distributed on an "AS IS" BASIS,
    
    11
    +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    
    12
    +# See the License for the specific language governing permissions and
    
    13
    +# limitations under the License.
    
    14
    +
    
    15
    +
    
    16
    +"""
    
    17
    +HardwareInterface
    
    18
    +=================
    
    19
    +
    
    20
    +Class to configure hardware and check requirements of leases.
    
    21
    +
    
    22
    +In the future this could also be used to request and display
    
    23
    +the status of hardware.
    
    24
    +"""
    
    25
    +
    
    26
    +
    
    27
    +from buildgrid._exceptions import FailedPreconditionError
    
    28
    +
    
    29
    +
    
    30
    +class HardwareInterface:
    
    31
    +
    
    32
    +    def __init__(self, worker):
    
    33
    +        self._worker = worker
    
    34
    +
    
    35
    +    def configure_hardware(self, requirements):
    
    36
    +        """ Can check if the requirements can be met and also
    
    37
    +        in the future, potentially configure the hardware.
    
    38
    +        """
    
    39
    +        worker = self._worker
    
    40
    +
    
    41
    +        for config_requirement in requirements.configs:
    
    42
    +            if config_requirement.key not in worker.configs:
    
    43
    +                raise FailedPreconditionError("Config not supported: [{}]".format(config_requirement))
    
    44
    +
    
    45
    +    def get_worker_pb2(self):
    
    46
    +        return self._worker.get_pb2()

  • buildgrid/bot/hardware/worker.py
    ... ... @@ -13,39 +13,34 @@
    13 13
     # limitations under the License.
    
    14 14
     
    
    15 15
     
    
    16
    +from buildgrid._protos.google.devtools.remoteworkers.v1test2 import worker_pb2
    
    17
    +
    
    18
    +
    
    16 19
     class Worker:
    
    17 20
     
    
    18 21
         def __init__(self, properties=None, configs=None):
    
    19 22
             self._devices = []
    
    20
    -        self.__configs = {}
    
    21
    -        self.__properties = {}
    
    23
    +        self._configs = {}
    
    24
    +        self._properties = {}
    
    25
    +        self.__property_keys = ['pool']
    
    26
    +        self.__config_keys = ['DockerImage']
    
    22 27
     
    
    23 28
             if properties:
    
    24 29
                 for k, v in properties.items():
    
    25
    -                self.update_config(k, v)
    
    30
    +                if k in self.__property_keys:
    
    31
    +                    self._add_properties(k, v)
    
    26 32
     
    
    27 33
             if configs:
    
    28 34
                 for k, v in configs.items():
    
    29
    -                if k == 'DockerImage':
    
    30
    -                    self.__configs[k] = v
    
    31
    -                else:
    
    32
    -                    raise KeyError('Key not supported: [{}]'.format(k))
    
    35
    +                self._add_config(k, v)
    
    33 36
     
    
    34 37
         @property
    
    35 38
         def configs(self):
    
    36
    -        return self.__configs
    
    39
    +        return self._configs
    
    37 40
     
    
    38 41
         @property
    
    39 42
         def properties(self):
    
    40
    -        return self.__properties
    
    41
    -
    
    42
    -    def update_config(self, key, value):
    
    43
    -        """Update the config of the device.
    
    44
    -        Only certain keys are allowed."""
    
    45
    -        if k == 'pool':
    
    46
    -            self.__properties[k] = v
    
    47
    -        else:
    
    48
    -            raise KeyError('Key not supported: [{}]'.format(k))
    
    43
    +        return self._properties
    
    49 44
     
    
    50 45
         def add_device(self, device):
    
    51 46
             self._devices.append(device)
    
    ... ... @@ -53,16 +48,41 @@ class Worker:
    53 48
         def get_pb2(self):
    
    54 49
             devices = [device.get_pb2() for device in self._devices]
    
    55 50
             worker = worker_pb2.Worker(devices=devices)
    
    56
    -        property_message = worker_pb2.Worker.Property()
    
    57
    -        for k, v in self.__properties.items():
    
    58
    -            property_message.key = k
    
    59
    -            property_message.value = v
    
    60
    -            worker.properties.extend([property_message])
    
    61
    -
    
    62
    -        config_message = worker_pb2.Worker.Config()
    
    63
    -        for k, v in self.__properties.items():
    
    64
    -            property_message.key = k
    
    65
    -            property_message.value = v
    
    66
    -            worker.configs.extend([config_message])
    
    51
    +
    
    52
    +        for k, v in self._properties.items():
    
    53
    +            for prop in v:
    
    54
    +                property_message = worker_pb2.Device.Property()
    
    55
    +                property_message.key = k
    
    56
    +                property_message.value = prop
    
    57
    +                device.properties.extend([property_message])
    
    58
    +
    
    59
    +        for k, v in self._configs.items():
    
    60
    +            for cfg in v:
    
    61
    +                config_message = worker_pb2.Worker.Config()
    
    62
    +                config.key = k
    
    63
    +                config_message.value = cfg
    
    64
    +                worker.configs.extend([config_message])
    
    67 65
     
    
    68 66
             return worker
    
    67
    +
    
    68
    +    def _add_config(self, key, value):
    
    69
    +        if key in self.__config_keys:
    
    70
    +            cfg = self._configs.get(key)
    
    71
    +            if not cfg:
    
    72
    +                self._configs[key] = [value]
    
    73
    +            else:
    
    74
    +                cfg[key].append(value)
    
    75
    +
    
    76
    +        else:
    
    77
    +            raise KeyError('Key not supported: [{}]'.format(key))
    
    78
    +
    
    79
    +    def _add_property(self, key, value):
    
    80
    +        if key in self.__property_keys:
    
    81
    +            prop = self._properties.get(key)
    
    82
    +            if not prop:
    
    83
    +                self._properties[key] = [value]
    
    84
    +            else:
    
    85
    +                prop[key].append(value)
    
    86
    +
    
    87
    +        else:
    
    88
    +            raise KeyError('Key not supported: [{}]'.format(key))

  • buildgrid/bot/interface.py
    ... ... @@ -38,10 +38,17 @@ class BotInterface:
    38 38
         def create_bot_session(self, parent, bot_session):
    
    39 39
             request = bots_pb2.CreateBotSessionRequest(parent=parent,
    
    40 40
                                                        bot_session=bot_session)
    
    41
    -        return self._stub.CreateBotSession(request)
    
    41
    +        try:
    
    42
    +            return self._stub.CreateBotSession(request)
    
    43
    +        except Exception as e:
    
    44
    +            self.logger.error("Error creating bot session: [{}]".format(e))
    
    45
    +            raise
    
    42 46
     
    
    43 47
         def update_bot_session(self, bot_session, update_mask=None):
    
    44 48
             request = bots_pb2.UpdateBotSessionRequest(name=bot_session.name,
    
    45 49
                                                        bot_session=bot_session,
    
    46 50
                                                        update_mask=update_mask)
    
    47
    -        return self._stub.UpdateBotSession(request)
    51
    +        try:
    
    52
    +            return self._stub.UpdateBotSession(request)
    
    53
    +        except Exception as e:
    
    54
    +            self.logger.error("Error updating bot session: [{}]".format(e))

  • buildgrid/bot/session.py
    ... ... @@ -25,18 +25,22 @@ Allows connections
    25 25
     import asyncio
    
    26 26
     import logging
    
    27 27
     import platform
    
    28
    -import uuid
    
    28
    +import pdb
    
    29
    +# pdb.set_trace()
    
    29 30
     
    
    30 31
     import grpc
    
    31 32
     
    
    32 33
     from buildgrid._enums import BotStatus, LeaseState
    
    34
    +from buildgrid._protos.google.devtools.remoteworkers.v1test2 import bots_pb2
    
    33 35
     from buildgrid._protos.google.rpc import code_pb2
    
    34
    -from buildgrid._protos.google.devtools.remoteworkers.v1test2 import bots_pb2, worker_pb2
    
    35 36
     from buildgrid._exceptions import BotError
    
    36 37
     
    
    38
    +from buildgrid._exceptions import FailedPreconditionError
    
    39
    +
    
    40
    +from .tenantmanager import TenantManager
    
    37 41
     
    
    38 42
     class BotSession:
    
    39
    -    def __init__(self, parent, interface, worker):
    
    43
    +    def __init__(self, parent, bots_interface, hardware_interface):
    
    40 44
             """ Unique bot ID within the farm used to identify this bot
    
    41 45
             Needs to be human readable.
    
    42 46
             All prior sessions with bot_id of same ID are invalidated.
    
    ... ... @@ -45,88 +49,82 @@ class BotSession:
    45 49
             """
    
    46 50
     
    
    47 51
             self.logger = logging.getLogger(__name__)
    
    48
    -        self._context = None
    
    49 52
     
    
    50
    -        self._worker = worker
    
    51
    -        self._interface = interface
    
    52
    -        self._leases = {}
    
    53
    -        self._parent = parent
    
    53
    +        self._bots_interface = bots_interface
    
    54
    +        self._hardware_interface = hardware_interface
    
    55
    +
    
    54 56
             self._status = BotStatus.OK.value
    
    57
    +        self._tenant_manager = TenantManager()
    
    55 58
     
    
    59
    +        self.__parent = parent
    
    56 60
             self.__bot_id = '{}.{}'.format(parent, platform.node())
    
    57 61
             self.__name = None
    
    58 62
     
    
    63
    +        # Remove these and add to a worker config in the future
    
    64
    +        self._work = None
    
    65
    +        self._context = None
    
    66
    +
    
    59 67
         @property
    
    60 68
         def bot_id(self):
    
    61 69
             return self.__bot_id
    
    62 70
     
    
    63
    -    def create_bot_session(self, work, context=None):
    
    64
    -        self.logger.debug("Creating bot session")
    
    71
    +    def create_bot_session(self, work, context):
    
    72
    +        # Drop this when properly adding to the work
    
    65 73
             self._work = work
    
    66 74
             self._context = context
    
    67 75
     
    
    68
    -        session = self._interface.create_bot_session(self._parent, self.get_pb2())
    
    76
    +        self.logger.debug("Creating bot session")
    
    77
    +
    
    78
    +        session = self._bots_interface.create_bot_session(self.__parent, self.get_pb2())
    
    69 79
             self.__name = session.name
    
    70 80
     
    
    71 81
             self.logger.info("Created bot session with name: [{}]".format(self.__name))
    
    72 82
     
    
    73 83
             for lease in session.leases:
    
    74
    -            self._update_lease_from_server(lease)
    
    84
    +            self._register_lease(lease)
    
    75 85
     
    
    76 86
         def update_bot_session(self):
    
    77 87
             self.logger.debug("Updating bot session: [{}]".format(self.__bot_id))
    
    78
    -        session = self._interface.update_bot_session(self.get_pb2())
    
    79
    -        for k, v in list(self._leases.items()):
    
    80
    -            if v.state == LeaseState.COMPLETED.value:
    
    81
    -                del self._leases[k]
    
    88
    +
    
    89
    +        session = self._bots_interface.update_bot_session(self.get_pb2())
    
    90
    +        server_ids = []
    
    82 91
     
    
    83 92
             for lease in session.leases:
    
    84
    -            self._update_lease_from_server(lease)
    
    93
    +            server_ids.append(lease.id)
    
    85 94
     
    
    86
    -    def get_pb2(self):
    
    87
    -        leases = list(self._leases.values())
    
    88
    -        if not leases:
    
    89
    -            leases = None
    
    95
    +            lease_state = LeaseState(lease.state)
    
    96
    +            if lease_state == LeaseState.PENDING:
    
    97
    +                self._register_lease(lease)
    
    98
    +
    
    99
    +            elif lease_state == LeaseState.CANCELLED:
    
    100
    +                self._tenant_manager.cancel_tenancy(lease_id)
    
    101
    +
    
    102
    +        closed_lease_ids = [x for x in self._tenant_manager.get_lease_ids() if x not in server_ids]
    
    103
    +        for lease_id in closed_lease_ids:
    
    104
    +            self._tenant_manager.remove_tenant(lease_id)
    
    90 105
     
    
    91
    -        return bots_pb2.BotSession(worker=self._worker.get_pb2(),
    
    106
    +    def get_pb2(self):
    
    107
    +        return bots_pb2.BotSession(worker=self._hardware_interface.get_worker_pb2(),
    
    92 108
                                        status=self._status,
    
    93
    -                                   leases=leases,
    
    109
    +                                   leases=self._tenant_manager.get_leases(),
    
    94 110
                                        bot_id=self.__bot_id,
    
    95 111
                                        name=self.__name)
    
    96 112
     
    
    97
    -    def lease_completed(self, lease):
    
    98
    -        lease.state = LeaseState.COMPLETED.value
    
    99
    -        self._leases[lease.id] = lease
    
    100
    -
    
    101
    -    def _update_lease_from_server(self, lease):
    
    102
    -        """
    
    103
    -        State machine for any recieved updates to the leases.
    
    104
    -        """
    
    105
    -        # TODO: Compare with previous state of lease
    
    106
    -        if lease.state == LeaseState.PENDING.value:
    
    107
    -            lease.state = LeaseState.ACTIVE.value
    
    108
    -            self._leases[lease.id] = lease
    
    109
    -            self.update_bot_session()
    
    110
    -            asyncio.ensure_future(self.create_work(lease))
    
    111
    -
    
    112
    -    async def create_work(self, lease):
    
    113
    -        self.logger.debug("Work created: [{}]".format(lease.id))
    
    114
    -        loop = asyncio.get_event_loop()
    
    115
    -
    
    113
    +    def _register_lease(self, lease):
    
    114
    +        lease_id = lease.id
    
    116 115
             try:
    
    117
    -            lease = await loop.run_in_executor(None, self._work, self._context, lease)
    
    116
    +            self._tenant_manager.create_tenancy(lease)
    
    118 117
     
    
    119
    -        except grpc.RpcError as e:
    
    120
    -            self.logger.error("RPC error thrown: [{}]".format(e))
    
    121
    -            lease.status.CopyFrom(e.code())
    
    118
    +        except KeyError as e:
    
    119
    +            self.logger.debug("Cannot register lease=[{}]. {}".format(lease.id, e))
    
    122 120
     
    
    123
    -        except BotError as e:
    
    124
    -            self.logger.error("Internal bot error thrown: [{}]".format(e))
    
    125
    -            lease.status.code = code_pb2.INTERNAL
    
    121
    +        else:
    
    122
    +            try:
    
    123
    +                self._hardware_interface.configure_hardware(lease.requirements)
    
    126 124
     
    
    127
    -        except Exception as e:
    
    128
    -            self.logger.error("Exception thrown: [{}]".format(e))
    
    129
    -            lease.status.code = code_pb2.INTERNAL
    
    125
    +            except FailedPreconditionError as e:
    
    126
    +                self.logger.error("Failed precondition: [{}]".format(e))
    
    127
    +                self._tenant_manager.complete_lease(lease_id, status=code_pb2.FailedPreconditionError)
    
    130 128
     
    
    131
    -        self.logger.debug("Work complete: [{}]".format(lease.id))
    
    132
    -        self.lease_completed(lease)
    129
    +            else:
    
    130
    +                self._tenant_manager.create_work(lease_id, self._work, self._context)

  • buildgrid/bot/tenant.py
    1
    +# Copyright (C) 2018 Bloomberg LP
    
    2
    +#
    
    3
    +# Licensed under the Apache License, Version 2.0 (the "License");
    
    4
    +# you may not use this file except in compliance with the License.
    
    5
    +# You may obtain a copy of the License at
    
    6
    +#
    
    7
    +#  <http://www.apache.org/licenses/LICENSE-2.0>
    
    8
    +#
    
    9
    +# Unless required by applicable law or agreed to in writing, software
    
    10
    +# distributed under the License is distributed on an "AS IS" BASIS,
    
    11
    +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    
    12
    +# See the License for the specific language governing permissions and
    
    13
    +# limitations under the License.
    
    14
    +
    
    15
    +"""
    
    16
    +Tenant
    
    17
    +======
    
    18
    +
    
    19
    +Handles leased and runs leased work.
    
    20
    +"""
    
    21
    +
    
    22
    +import asyncio
    
    23
    +import logging
    
    24
    +
    
    25
    +from functools import partial
    
    26
    +
    
    27
    +from buildgrid._protos.google.devtools.remoteworkers.v1test2 import bots_pb2
    
    28
    +
    
    29
    +from buildgrid._enums import LeaseState
    
    30
    +
    
    31
    +
    
    32
    +class Tenant:
    
    33
    +
    
    34
    +    def __init__(self, lease):
    
    35
    +
    
    36
    +        if lease.state != LeaseState.PENDING.value:
    
    37
    +            raise ValueError("Lease state not `PENDING`: {}".format(lease.state))
    
    38
    +
    
    39
    +        self.logger = logging.getLogger(__name__)
    
    40
    +        self.lease_finished = False
    
    41
    +
    
    42
    +        self._lease = lease
    
    43
    +
    
    44
    +    @property
    
    45
    +    def lease(self):
    
    46
    +        return self._lease
    
    47
    +
    
    48
    +    def get_lease_state(self):
    
    49
    +        return LeaseState(self._lease.state)
    
    50
    +
    
    51
    +    def update_lease_state(self, state):
    
    52
    +        self._lease.state = state.value
    
    53
    +
    
    54
    +    def update_lease_status(self, status):
    
    55
    +        self._lease.status.CopyFrom(status)
    
    56
    +
    
    57
    +    async def run_work(self, work, context=None, executor=None):
    
    58
    +        self.logger.debug("Work created: [{}]".format(self._lease.id))
    
    59
    +
    
    60
    +        # Ensures if anything happens to the lease during work, we still have a copy.
    
    61
    +        lease = bots_pb2.Lease()
    
    62
    +        lease.CopyFrom(self._lease)
    
    63
    +
    
    64
    +        loop = asyncio.get_event_loop()
    
    65
    +
    
    66
    +        try:
    
    67
    +            lease = await loop.run_in_executor(executor, partial(work, context, self._lease))
    
    68
    +            self._lease.CopyFrom(lease)
    
    69
    +
    
    70
    +        except asyncio.CancelledError as e:
    
    71
    +            self.logger.error("Task cancelled: [{}]".format(e))
    
    72
    +
    
    73
    +        except grpc.RpcError as e:
    
    74
    +            self.logger.error("RPC error thrown: [{}]".format(e))
    
    75
    +            lease.status.CopyFrom(e.code())
    
    76
    +
    
    77
    +        except BotError as e:
    
    78
    +            self.logger.error("Internal bot error thrown: [{}]".format(e))
    
    79
    +            lease.status.code = code_pb2.INTERNAL
    
    80
    +
    
    81
    +        except Exception as e:
    
    82
    +            self.logger.error("Exception thrown: [{}]".format(e))
    
    83
    +            lease.status.code = code_pb2.INTERNAL
    
    84
    +
    
    85
    +        self.logger.debug("Work completed: [{}]".format(lease.id))

  • buildgrid/bot/tenantmanager.py
    1
    +# Copyright (C) 2018 Bloomberg LP
    
    2
    +#
    
    3
    +# Licensed under the Apache License, Version 2.0 (the "License");
    
    4
    +# you may not use this file except in compliance with the License.
    
    5
    +# You may obtain a copy of the License at
    
    6
    +#
    
    7
    +#  <http://www.apache.org/licenses/LICENSE-2.0>
    
    8
    +#
    
    9
    +# Unless required by applicable law or agreed to in writing, software
    
    10
    +# distributed under the License is distributed on an "AS IS" BASIS,
    
    11
    +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    
    12
    +# See the License for the specific language governing permissions and
    
    13
    +# limitations under the License.
    
    14
    +
    
    15
    +
    
    16
    +"""
    
    17
    +TenantManager
    
    18
    +=============
    
    19
    +
    
    20
    +Looks after leases of work.
    
    21
    +"""
    
    22
    +
    
    23
    +
    
    24
    +import asyncio
    
    25
    +import logging
    
    26
    +from functools import partial
    
    27
    +
    
    28
    +import grpc
    
    29
    +
    
    30
    +from buildgrid._enums import LeaseState
    
    31
    +
    
    32
    +from .tenant import Tenant
    
    33
    +
    
    34
    +class TenantManager:
    
    35
    +
    
    36
    +    def __init__(self):
    
    37
    +        self.logger = logging.getLogger(__name__)
    
    38
    +        self._tenants = {}
    
    39
    +        self._tasks = {}
    
    40
    +
    
    41
    +    def create_tenancy(self, lease):
    
    42
    +        lease_id = lease.id
    
    43
    +
    
    44
    +        if lease_id not in self._tenants:
    
    45
    +            tenant = Tenant(lease)
    
    46
    +            self._tenants[lease_id] = tenant
    
    47
    +
    
    48
    +        else:
    
    49
    +            raise KeyError("Lease id already exists: [{}]".format(lease_id))
    
    50
    +
    
    51
    +    def remove_tenant(self, lease_id):
    
    52
    +        state = self.get_lease_state(lease_id)
    
    53
    +        if state == LeaseState.PENDING or state == LeaseState.ACTIVE:
    
    54
    +            self.logger.error("Attempting to remove a lease not finished."
    
    55
    +                              "Bot will not remove lease."
    
    56
    +                              "Lease: [{}]".format(self._tenants[lease_id].lease))
    
    57
    +
    
    58
    +        else:
    
    59
    +            del self._tenants[lease_id]
    
    60
    +            del self._tasks[lease_id]
    
    61
    +
    
    62
    +    def get_leases(self):
    
    63
    +        leases = []
    
    64
    +        for tenant in self._tenants.values():
    
    65
    +            leases.append(tenant.lease)
    
    66
    +
    
    67
    +        if not leases:
    
    68
    +            return None
    
    69
    +
    
    70
    +        return leases
    
    71
    +
    
    72
    +    def get_lease_ids(self):
    
    73
    +        return self._tenants.keys()
    
    74
    +
    
    75
    +    def get_lease_state(self, lease_id):
    
    76
    +        return self._tenants[lease_id].get_lease_state()
    
    77
    +
    
    78
    +    def complete_lease(self, lease_id, status, task=None):
    
    79
    +        if status is not None:
    
    80
    +            self._update_lease_status(lease_id, status)
    
    81
    +
    
    82
    +        if self._tenants[lease_id].get_lease_state() != LeaseState.CANCELLED:
    
    83
    +            self._update_lease_state(lease_id, LeaseState.COMPLETED)
    
    84
    +
    
    85
    +    def create_work(self, lease_id, work, context):
    
    86
    +        self._update_lease_state(lease_id, LeaseState.ACTIVE)
    
    87
    +        tenant = self._tenants[lease_id]
    
    88
    +        task = asyncio.ensure_future(tenant.run_work(work, context))
    
    89
    +
    
    90
    +        task.add_done_callback(partial(self.complete_lease, lease_id, None))
    
    91
    +
    
    92
    +        self._tasks[lease_id] = task
    
    93
    +
    
    94
    +    def cancel_tenancy(self, lease_id):
    
    95
    +        self._update_lease_state(LeaseState.CANCELLED)
    
    96
    +        self._tasks[lease_id].cancel()
    
    97
    +
    
    98
    +    def _update_lease_state(self, lease_id, state):
    
    99
    +        self._tenants[lease_id].update_lease_state(state)
    
    100
    +
    
    101
    +    def _update_lease_status(self, lease_id, status):
    
    102
    +        self._tenants[lease_id].update_lease_status(status)

  • buildgrid/server/bots/instance.py
    ... ... @@ -34,7 +34,7 @@ class BotsInterface:
    34 34
             self.logger = logging.getLogger(__name__)
    
    35 35
     
    
    36 36
             self._bot_ids = {}
    
    37
    -        self._bot_sessions = {}
    
    37
    +        self._assigned_leases = {}
    
    38 38
             self._scheduler = scheduler
    
    39 39
     
    
    40 40
         def register_instance_with_server(self, instance_name, server):
    
    ... ... @@ -59,18 +59,15 @@ class BotsInterface:
    59 59
     
    
    60 60
             # Bot session name, selected by the server
    
    61 61
             name = "{}/{}".format(parent, str(uuid.uuid4()))
    
    62
    -
    
    63 62
             bot_session.name = name
    
    64 63
     
    
    65 64
             self._bot_ids[name] = bot_id
    
    66
    -        self._bot_sessions[name] = bot_session
    
    67 65
             self.logger.info("Created bot session name=[{}] with bot_id=[{}]".format(name, bot_id))
    
    68 66
     
    
    69
    -        # TODO: Send worker capabilities to the scheduler!
    
    70
    -        leases = self._scheduler.request_job_leases({})
    
    71
    -        if leases:
    
    72
    -            bot_session.leases.extend(leases)
    
    67
    +        # We want to keep a copy of lease ids we have assigned
    
    68
    +        self._assigned_leases[name] = set()
    
    73 69
     
    
    70
    +        self._request_leases(bot_session)
    
    74 71
             return bot_session
    
    75 72
     
    
    76 73
         def update_bot_session(self, name, bot_session):
    
    ... ... @@ -79,39 +76,53 @@ class BotsInterface:
    79 76
             """
    
    80 77
             self.logger.debug("Updating bot session name={}".format(name))
    
    81 78
             self._check_bot_ids(bot_session.bot_id, name)
    
    82
    -
    
    83
    -        leases = filter(None, [self._check_lease_state(lease) for lease in bot_session.leases])
    
    79
    +        self._check_assigned_leases(bot_session)
    
    84 80
     
    
    85 81
             for lease in bot_session.leases:
    
    86
    -            lease.Clear()
    
    87
    -
    
    88
    -        bot_session.leases.extend(leases)
    
    82
    +            checked_lease = self._check_lease_state(lease)
    
    83
    +            if not checked_lease:
    
    84
    +                # TODO: Make sure we don't need this
    
    85
    +                try:
    
    86
    +                    self._assigned_leases[name].remove(lease.id)
    
    87
    +                except KeyError:
    
    88
    +                    pass
    
    89
    +                lease.Clear()
    
    90
    +
    
    91
    +        self._request_leases(bot_session)
    
    92
    +        return bot_session
    
    89 93
     
    
    94
    +    def _request_leases(self, bot_session):
    
    90 95
             # TODO: Send worker capabilities to the scheduler!
    
    96
    +        # Only send one lease at a time currently.
    
    91 97
             if not bot_session.leases:
    
    92 98
                 leases = self._scheduler.request_job_leases({})
    
    93 99
                 if leases:
    
    100
    +                for lease in leases:
    
    101
    +                    self._assigned_leases[bot_session.name].add(lease.id)
    
    94 102
                     bot_session.leases.extend(leases)
    
    95 103
     
    
    96
    -        self._bot_sessions[name] = bot_session
    
    97
    -        return bot_session
    
    98
    -
    
    99 104
         def _check_lease_state(self, lease):
    
    105
    +        # careful here
    
    106
    +        # should store bot name in scheduler
    
    107
    +        lease_state = LeaseState(lease.state)
    
    108
    +
    
    109
    +        # Lease has replied with cancelled, remove
    
    110
    +        if lease_state == LeaseState.CANCELLED:
    
    111
    +            return None
    
    100 112
     
    
    101
    -        # Check for cancelled lease
    
    102
    -        if self._scheduler.get_lease_cancelled(lease.id):
    
    113
    +        try:
    
    114
    +            if self._scheduler.get_job_lease_cancelled(lease.id):
    
    115
    +                lease.state.CopyFrom(LeaseState.CANCELLED.value)
    
    116
    +                return lease
    
    117
    +        except KeyError:
    
    118
    +            # Job does not exist, remove from bot.
    
    103 119
                 return None
    
    104 120
     
    
    105
    -        # If not cancelled, update the status
    
    106 121
             self._scheduler.update_job_lease(lease)
    
    107 122
     
    
    108
    -        lease_state = LeaseState(lease.state)
    
    109 123
             if lease_state == LeaseState.COMPLETED:
    
    110 124
                 return None
    
    111 125
     
    
    112
    -        elif lease_state == LeaseState.CANCELLED:
    
    113
    -            return None
    
    114
    -
    
    115 126
             return lease
    
    116 127
     
    
    117 128
         def _check_bot_ids(self, bot_id, name=None):
    
    ... ... @@ -134,6 +145,19 @@ class BotsInterface:
    134 145
                             'Bot id already registered. ID sent: [{}].'
    
    135 146
                             'Id registered: [{}] with name: [{}]'.format(bot_id, _bot_id, _name))
    
    136 147
     
    
    148
    +    def _check_assigned_leases(self, bot_session):
    
    149
    +        session_lease_ids = []
    
    150
    +
    
    151
    +        for lease in bot_session.leases:
    
    152
    +            session_lease_ids.append(lease.id)
    
    153
    +
    
    154
    +        for lease_id in self._assigned_leases[bot_session.name]:
    
    155
    +            if lease_id not in session_lease_ids:
    
    156
    +                self.logger.error("Assigned lease id=[{}],"
    
    157
    +                                  " not found on bot with name=[{}] and id=[{}]."
    
    158
    +                                  " Retrying job".format(lease_id, bot_session.name, bot_session.bot_id))
    
    159
    +                self._scheduler.retry_job(lease_id)
    
    160
    +
    
    137 161
         def _close_bot_session(self, name):
    
    138 162
             """ Before removing the session, close any leases and
    
    139 163
             requeue with high priority.
    
    ... ... @@ -144,10 +168,9 @@ class BotsInterface:
    144 168
                 raise InvalidArgumentError("Bot id does not exist: [{}]".format(name))
    
    145 169
     
    
    146 170
             self.logger.debug("Attempting to close [{}] with name: [{}]".format(bot_id, name))
    
    147
    -        for lease in self._bot_sessions[name].leases:
    
    148
    -            if lease.state != LeaseState.COMPLETED.value:
    
    149
    -                # TODO: Be wary here, may need to handle rejected leases in future
    
    150
    -                self._scheduler.retry_job(lease.id)
    
    171
    +        for lease_id in self._assigned_leases[name]:
    
    172
    +            self._scheduler.retry_job(lease_id)
    
    173
    +        self._assigned_leases.pop(name)
    
    151 174
     
    
    152 175
             self.logger.debug("Closing bot session: [{}]".format(name))
    
    153 176
             self._bot_ids.pop(name)
    

  • buildgrid/server/job.py
    ... ... @@ -19,11 +19,9 @@ import uuid
    19 19
     from google.protobuf import timestamp_pb2
    
    20 20
     
    
    21 21
     from buildgrid._enums import LeaseState, OperationStage
    
    22
    -from buildgrid._exceptions import CancelledError
    
    23 22
     from buildgrid._protos.build.bazel.remote.execution.v2 import remote_execution_pb2
    
    24 23
     from buildgrid._protos.google.devtools.remoteworkers.v1test2 import bots_pb2
    
    25 24
     from buildgrid._protos.google.longrunning import operations_pb2
    
    26
    -from buildgrid._protos.google.rpc import code_pb2
    
    27 25
     
    
    28 26
     
    
    29 27
     class Job:
    
    ... ... @@ -36,18 +34,16 @@ class Job:
    36 34
             self._operation = operations_pb2.Operation()
    
    37 35
             self._lease = None
    
    38 36
     
    
    37
    +        self.__lease_cancelled = False
    
    39 38
             self.__execute_response = None
    
    40 39
             self.__operation_metadata = remote_execution_pb2.ExecuteOperationMetadata()
    
    41
    -
    
    42 40
             self.__queued_timestamp = timestamp_pb2.Timestamp()
    
    43 41
             self.__worker_start_timestamp = timestamp_pb2.Timestamp()
    
    44 42
             self.__worker_completed_timestamp = timestamp_pb2.Timestamp()
    
    45 43
     
    
    46
    -        self.__operation_cancelled = False
    
    47
    -        self.__lease_cancelled = False
    
    48
    -
    
    49 44
             self.__operation_metadata.action_digest.CopyFrom(action_digest)
    
    50 45
             self.__operation_metadata.stage = OperationStage.UNKNOWN.value
    
    46
    +        self.__operation_cancelled = False
    
    51 47
     
    
    52 48
             self._action.CopyFrom(action)
    
    53 49
             self._do_not_cache = self._action.do_not_cache
    
    ... ... @@ -140,9 +136,7 @@ class Job:
    140 136
             Only one :class:`Lease` can be emitted for a given job. This method
    
    141 137
             should only be used once, any furhter calls are ignored.
    
    142 138
             """
    
    143
    -        if self.__operation_cancelled:
    
    144
    -            return None
    
    145
    -        elif self._lease is not None:
    
    139
    +        if self._lease is not None:
    
    146 140
                 return None
    
    147 141
     
    
    148 142
             self._lease = bots_pb2.Lease()
    
    ... ... @@ -183,7 +177,7 @@ class Job:
    183 177
                 action_result = remote_execution_pb2.ActionResult()
    
    184 178
     
    
    185 179
                 # TODO: Make a distinction between build and bot failures!
    
    186
    -            if status.code != code_pb2.OK:
    
    180
    +            if status.code != 0:
    
    187 181
                     self._do_not_cache = True
    
    188 182
     
    
    189 183
                 if result is not None:
    
    ... ... @@ -200,15 +194,6 @@ class Job:
    200 194
                 self.__execute_response.cached_result = False
    
    201 195
                 self.__execute_response.status.CopyFrom(status)
    
    202 196
     
    
    203
    -    def cancel_lease(self):
    
    204
    -        """Triggers a job's :class:Lease cancellation.
    
    205
    -
    
    206
    -        This will not cancel the job's :class:Operation.
    
    207
    -        """
    
    208
    -        self.__lease_cancelled = True
    
    209
    -        if self._lease is not None:
    
    210
    -            self.update_lease_state(LeaseState.CANCELLED)
    
    211
    -
    
    212 197
         def update_operation_stage(self, stage):
    
    213 198
             """Operates a stage transition for the job's :class:Operation.
    
    214 199
     
    
    ... ... @@ -244,6 +229,10 @@ class Job:
    244 229
             if self.__operation_cancelled:
    
    245 230
                 raise CancelledError(self.__execute_response.status.message)
    
    246 231
     
    
    232
    +    def cancel_lease(self):
    
    233
    +        self.__lease_cancelled = True
    
    234
    +        self._update_lease_state(LeaseState.CANCELLED)
    
    235
    +
    
    247 236
         def cancel_operation(self):
    
    248 237
             """Triggers a job's :class:Operation cancellation.
    
    249 238
     
    

  • buildgrid/server/scheduler.py
    ... ... @@ -74,7 +74,8 @@ class Scheduler:
    74 74
                     # TODO: Mark these jobs as done
    
    75 75
                 else:
    
    76 76
                     job.update_operation_stage(OperationStage.QUEUED)
    
    77
    -                self.queue.appendleft(job)
    
    77
    +                job.update_lease_state(LeaseState.PENDING)
    
    78
    +                self.queue.append(job)
    
    78 79
     
    
    79 80
         def list_jobs(self):
    
    80 81
             return self.jobs.values()
    
    ... ... @@ -91,8 +92,12 @@ class Scheduler:
    91 92
                 return []
    
    92 93
     
    
    93 94
             job = self.queue.popleft()
    
    94
    -        # For now, one lease at a time:
    
    95
    -        lease = job.create_lease()
    
    95
    +
    
    96
    +        lease = job.lease
    
    97
    +
    
    98
    +        if not lease:
    
    99
    +            # For now, one lease at a time:
    
    100
    +            lease = job.create_lease()
    
    96 101
     
    
    97 102
             if lease:
    
    98 103
                 return [lease]
    



  • [Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]