GNOME.org
 

[Notes] [Git][BuildGrid/buildgrid][mablanch/132-gather-state-metrics] 19 commits: buildgrid/utils.py: New `get_hash_type` function.

Title: GitLab

Martin Blanchard pushed to branch mablanch/132-gather-state-metrics at BuildGrid / buildgrid

Commits:

21 changed files:

Changes:

  • .pylintrc
    ... ... @@ -185,6 +185,7 @@ ignore-on-opaque-inference=yes
    185 185 
     # for classes with dynamically set attributes). This supports the use of
    186 186 
     # qualified names.
    187 187 
     ignored-classes=google.protobuf.any_pb2.Any,
    188 
    +                google.protobuf.duration_pb2.Duration,
    188 189 
                     google.protobuf.timestamp_pb2.Timestamp
    189 190
    190 191 
     # List of module names for which member attributes should not be checked
    ... ... @@ -460,6 +461,7 @@ known-third-party=boto3,
    460 461 
                       enchant,
    461 462 
                       google,
    462 463 
                       grpc,
    464 
    +                  janus,
    463 465 
                       moto,
    464 466 
                       yaml
    465 467

  • buildgrid/_app/commands/cmd_capabilities.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 
    +import sys
    17 
    +from urllib.parse import urlparse
    18 
    +
    19 
    +import click
    20 
    +import grpc
    21 
    +
    22 
    +from buildgrid.client.capabilities import CapabilitiesInterface
    23 
    +
    24 
    +from ..cli import pass_context
    25 
    +
    26 
    +
    27 
    +@click.command(name='capabilities', short_help="Capabilities service.")
    28 
    +@click.option('--remote', type=click.STRING, default='http://localhost:50051', show_default=True,
    29 
    +              help="Remote execution server's URL (port defaults to 50051 if no specified).")
    30 
    +@click.option('--client-key', type=click.Path(exists=True, dir_okay=False), default=None,
    31 
    +              help="Private client key for TLS (PEM-encoded)")
    32 
    +@click.option('--client-cert', type=click.Path(exists=True, dir_okay=False), default=None,
    33 
    +              help="Public client certificate for TLS (PEM-encoded)")
    34 
    +@click.option('--server-cert', type=click.Path(exists=True, dir_okay=False), default=None,
    35 
    +              help="Public server certificate for TLS (PEM-encoded)")
    36 
    +@click.option('--instance-name', type=click.STRING, default='main', show_default=True,
    37 
    +              help="Targeted farm instance name.")
    38 
    +@pass_context
    39 
    +def cli(context, remote, instance_name, client_key, client_cert, server_cert):
    40 
    +    click.echo("Getting capabilities...")
    41 
    +    url = urlparse(remote)
    42 
    +
    43 
    +    remote = '{}:{}'.format(url.hostname, url.port or 50051)
    44 
    +    instance_name = instance_name
    45 
    +
    46 
    +    if url.scheme == 'http':
    47 
    +        channel = grpc.insecure_channel(remote)
    48 
    +    else:
    49 
    +        credentials = context.load_client_credentials(client_key, client_cert, server_cert)
    50 
    +        if not credentials:
    51 
    +            click.echo("ERROR: no TLS keys were specified and no defaults could be found.", err=True)
    52 
    +            sys.exit(-1)
    53 
    +
    54 
    +        channel = grpc.secure_channel(remote, credentials)
    55 
    +
    56 
    +    interface = CapabilitiesInterface(channel)
    57 
    +    response = interface.get_capabilities(instance_name)
    58 
    +    click.echo(response)

  • buildgrid/client/capabilities.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 
    +import logging
    17 
    +import grpc
    18 
    +
    19 
    +from buildgrid._protos.build.bazel.remote.execution.v2 import remote_execution_pb2, remote_execution_pb2_grpc
    20 
    +
    21 
    +
    22 
    +class CapabilitiesInterface:
    23 
    +    """Interface for calls the the Capabilities Service."""
    24 
    +
    25 
    +    def __init__(self, channel):
    26 
    +        """Initialises an instance of the capabilities service.
    27 
    +
    28 
    +        Args:
    29 
    +            channel (grpc.Channel): A gRPC channel to the CAS endpoint.
    30 
    +        """
    31 
    +        self.__logger = logging.getLogger(__name__)
    32 
    +        self.__stub = remote_execution_pb2_grpc.CapabilitiesStub(channel)
    33 
    +
    34 
    +    def get_capabilities(self, instance_name):
    35 
    +        """Returns the capabilities or the server to the user.
    36 
    +
    37 
    +        Args:
    38 
    +            instance_name (str): The name of the instance."""
    39 
    +
    40 
    +        request = remote_execution_pb2.GetCapabilitiesRequest(instance_name=instance_name)
    41 
    +        try:
    42 
    +            return self.__stub.GetCapabilities(request)
    43 
    +
    44 
    +        except grpc.RpcError as e:
    45 
    +            self.__logger.error(e)
    46 
    +            raise

  • buildgrid/server/bots/instance.py
    ... ... @@ -37,6 +37,10 @@ class BotsInterface:
    37 37 
             self._assigned_leases = {}
    38 38 
             self._scheduler = scheduler
    39 39
    40 
    +    @property
    41 
    +    def scheduler(self):
    42 
    +        return self._scheduler
    43 
    +
    40 44 
         def register_instance_with_server(self, instance_name, server):
    41 45 
             server.add_bots_interface(self, instance_name)
    42 46

  • buildgrid/server/bots/service.py
    ... ... @@ -23,8 +23,9 @@ import logging
    23 23
    24 24 
     import grpc
    25 25
    26 
    -from google.protobuf.empty_pb2 import Empty
    26 
    +from google.protobuf import empty_pb2, timestamp_pb2
    27 27
    28 
    +from buildgrid._enums import BotStatus
    28 29 
     from buildgrid._exceptions import InvalidArgumentError, OutOfSyncError
    29 30 
     from buildgrid._protos.google.devtools.remoteworkers.v1test2 import bots_pb2
    30 31 
     from buildgrid._protos.google.devtools.remoteworkers.v1test2 import bots_pb2_grpc
    ... ... @@ -32,24 +33,86 @@ from buildgrid._protos.google.devtools.remoteworkers.v1test2 import bots_pb2_grp
    32 33
    33 34 
     class BotsService(bots_pb2_grpc.BotsServicer):
    34 35
    35 
    -    def __init__(self, server):
    36 
    +    def __init__(self, server, monitor=False):
    36 37 
             self.__logger = logging.getLogger(__name__)
    37 38
    39 
    +        self.__bots_by_status = None
    40 
    +        self.__bots_by_instance = None
    41 
    +        self.__bots = None
    42 
    +
    38 43 
             self._instances = {}
    39 44
    40 45 
             bots_pb2_grpc.add_BotsServicer_to_server(self, server)
    41 46
    42 
    -    def add_instance(self, name, instance):
    43 
    -        self._instances[name] = instance
    47 
    +        self._is_instrumented = monitor
    48 
    +
    49 
    +        if self._is_instrumented:
    50 
    +            self.__bots_by_status = {}
    51 
    +            self.__bots_by_instance = {}
    52 
    +            self.__bots = {}
    53 
    +
    54 
    +            self.__bots_by_status[BotStatus.OK] = set()
    55 
    +            self.__bots_by_status[BotStatus.UNHEALTHY] = set()
    56 
    +
    57 
    +    # --- Public API ---
    58 
    +
    59 
    +    def add_instance(self, instance_name, instance):
    60 
    +        """Registers a new servicer instance.
    61 
    +
    62 
    +        Args:
    63 
    +            instance_name (str): The new instance's name.
    64 
    +            instance (BotsInterface): The new instance itself.
    65 
    +        """
    66 
    +        self._instances[instance_name] = instance
    67 
    +
    68 
    +        if self._is_instrumented:
    69 
    +            self.__bots_by_instance[instance_name] = set()
    70 
    +
    71 
    +    def get_scheduler(self, instance_name):
    72 
    +        """Retrieves a reference to the scheduler for an instance.
    73 
    +
    74 
    +        Args:
    75 
    +            instance_name (str): The name of the instance to query.
    76 
    +
    77 
    +        Returns:
    78 
    +            Scheduler: A reference to the scheduler for `instance_name`.
    79 
    +
    80 
    +        Raises:
    81 
    +            InvalidArgumentError: If no instance named `instance_name` exists.
    82 
    +        """
    83 
    +        instance = self._get_instance(instance_name)
    84 
    +
    85 
    +        return instance.scheduler
    86 
    +
    87 
    +    # --- Public API: Servicer ---
    44 88
    45 89 
         def CreateBotSession(self, request, context):
    90 
    +        """Handles CreateBotSessionRequest messages.
    91 
    +
    92 
    +        Args:
    93 
    +            request (CreateBotSessionRequest): The incoming RPC request.
    94 
    +            context (grpc.ServicerContext): Context for the RPC call.
    95 
    +        """
    46 96 
             self.__logger.debug("CreateBotSession request from [%s]", context.peer())
    47 97
    98 
    +        instance_name = request.parent
    99 
    +        bot_status = BotStatus(request.bot_session.status)
    100 
    +        bot_id = request.bot_session.bot_id
    101 
    +
    48 102 
             try:
    49 
    -            parent = request.parent
    50 
    -            instance = self._get_instance(request.parent)
    51 
    -            return instance.create_bot_session(parent,
    52 
    -                                               request.bot_session)
    103 
    +            instance = self._get_instance(instance_name)
    104 
    +            bot_session = instance.create_bot_session(instance_name,
    105 
    +                                                      request.bot_session)
    106 
    +            now = timestamp_pb2.Timestamp()
    107 
    +            now.GetCurrentTime()
    108 
    +
    109 
    +            if self._is_instrumented:
    110 
    +                self.__bots[bot_id] = now
    111 
    +                self.__bots_by_instance[instance_name].add(bot_id)
    112 
    +                if bot_status in self.__bots_by_status:
    113 
    +                    self.__bots_by_status[bot_status].add(bot_id)
    114 
    +
    115 
    +            return bot_session
    53 116
    54 117 
             except InvalidArgumentError as e:
    55 118 
                 self.__logger.error(e)
    ... ... @@ -59,17 +122,41 @@ class BotsService(bots_pb2_grpc.BotsServicer):
    59 122 
             return bots_pb2.BotSession()
    60 123
    61 124 
         def UpdateBotSession(self, request, context):
    125 
    +        """Handles UpdateBotSessionRequest messages.
    126 
    +
    127 
    +        Args:
    128 
    +            request (UpdateBotSessionRequest): The incoming RPC request.
    129 
    +            context (grpc.ServicerContext): Context for the RPC call.
    130 
    +        """
    62 131 
             self.__logger.debug("UpdateBotSession request from [%s]", context.peer())
    63 132
    133 
    +        names = request.name.split("/")
    134 
    +        bot_status = BotStatus(request.bot_session.status)
    135 
    +        bot_id = request.bot_session.bot_id
    136 
    +
    64 137 
             try:
    65 
    -            names = request.name.split("/")
    66 
    -            # Operation name should be in format:
    67 
    -            # {instance/name}/{uuid}
    68 
    -            instance_name = ''.join(names[0:-1])
    138 
    +            instance_name = '/'.join(names[:-1])
    69 139
    70 140 
                 instance = self._get_instance(instance_name)
    71 
    -            return instance.update_bot_session(request.name,
    72 
    -                                               request.bot_session)
    141 
    +            bot_session = instance.update_bot_session(request.name,
    142 
    +                                                      request.bot_session)
    143 
    +
    144 
    +            if self._is_instrumented:
    145 
    +                self.__bots[bot_id].GetCurrentTime()
    146 
    +                if bot_id not in self.__bots_by_status[bot_status]:
    147 
    +                    if bot_status == BotStatus.OK:
    148 
    +                        self.__bots_by_status[BotStatus.OK].add(bot_id)
    149 
    +                        self.__bots_by_status[BotStatus.UNHEALTHY].discard(bot_id)
    150 
    +
    151 
    +                    elif bot_status == BotStatus.UNHEALTHY:
    152 
    +                        self.__bots_by_status[BotStatus.OK].discard(bot_id)
    153 
    +                        self.__bots_by_status[BotStatus.UNHEALTHY].add(bot_id)
    154 
    +
    155 
    +                    else:
    156 
    +                        self.__bots_by_instance[instance_name].remove(bot_id)
    157 
    +                        del self.__bots[bot_id]
    158 
    +
    159 
    +            return bot_session
    73 160
    74 161 
             except InvalidArgumentError as e:
    75 162 
                 self.__logger.error(e)
    ... ... @@ -89,10 +176,47 @@ class BotsService(bots_pb2_grpc.BotsServicer):
    89 176 
             return bots_pb2.BotSession()
    90 177
    91 178 
         def PostBotEventTemp(self, request, context):
    179 
    +        """Handles PostBotEventTempRequest messages.
    180 
    +
    181 
    +        Args:
    182 
    +            request (PostBotEventTempRequest): The incoming RPC request.
    183 
    +            context (grpc.ServicerContext): Context for the RPC call.
    184 
    +        """
    92 185 
             self.__logger.debug("PostBotEventTemp request from [%s]", context.peer())
    93 186
    94 187 
             context.set_code(grpc.StatusCode.UNIMPLEMENTED)
    95 
    -        return Empty()
    188 
    +
    189 
    +        return empty_pb2.Empty()
    190 
    +
    191 
    +    # --- Public API: Monitoring ---
    192 
    +
    193 
    +    @property
    194 
    +    def is_instrumented(self):
    195 
    +        return self._is_instrumented
    196 
    +
    197 
    +    def query_n_bots(self):
    198 
    +        if self.__bots is not None:
    199 
    +            return len(self.__bots)
    200 
    +
    201 
    +        return 0
    202 
    +
    203 
    +    def query_n_bots_for_instance(self, instance_name):
    204 
    +        try:
    205 
    +            if self.__bots_by_instance is not None:
    206 
    +                return len(self.__bots_by_instance[instance_name])
    207 
    +        except KeyError:
    208 
    +            pass
    209 
    +        return 0
    210 
    +
    211 
    +    def query_n_bots_for_status(self, bot_status):
    212 
    +        try:
    213 
    +            if self.__bots_by_status is not None:
    214 
    +                return len(self.__bots_by_status[bot_status])
    215 
    +        except KeyError:
    216 
    +            pass
    217 
    +        return 0
    218 
    +
    219 
    +    # --- Private API ---
    96 220
    97 221 
         def _get_instance(self, name):
    98 222 
             try:

  • buildgrid/server/capabilities/__init__.py

  • buildgrid/server/capabilities/instance.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 
    +import logging
    17 
    +
    18 
    +from buildgrid._protos.build.bazel.remote.execution.v2 import remote_execution_pb2
    19 
    +
    20 
    +
    21 
    +class CapabilitiesInstance:
    22 
    +
    23 
    +    def __init__(self, cas_instance=None, action_cache_instance=None, execution_instance=None):
    24 
    +        self.__logger = logging.getLogger(__name__)
    25 
    +        self.__cas_instance = cas_instance
    26 
    +        self.__action_cache_instance = action_cache_instance
    27 
    +        self.__execution_instance = execution_instance
    28 
    +
    29 
    +    def register_instance_with_server(self, instance_name, server):
    30 
    +        server.add_capabilities_instance(self, instance_name)
    31 
    +
    32 
    +    def add_cas_instance(self, cas_instance):
    33 
    +        self.__cas_instance = cas_instance
    34 
    +
    35 
    +    def add_action_cache_instance(self, action_cache_instance):
    36 
    +        self.__action_cache_instance = action_cache_instance
    37 
    +
    38 
    +    def add_execution_instance(self, execution_instance):
    39 
    +        self.__execution_instance = execution_instance
    40 
    +
    41 
    +    def get_capabilities(self):
    42 
    +        server_capabilities = remote_execution_pb2.ServerCapabilities()
    43 
    +        server_capabilities.cache_capabilities.CopyFrom(self._get_cache_capabilities())
    44 
    +        server_capabilities.execution_capabilities.CopyFrom(self._get_capabilities_execution())
    45 
    +        # TODO
    46 
    +        # When API is stable, fill out SemVer values
    47 
    +        # server_capabilities.deprecated_api_version =
    48 
    +        # server_capabilities.low_api_version =
    49 
    +        # server_capabilities.low_api_version =
    50 
    +        # server_capabilities.hig_api_version =
    51 
    +        return server_capabilities
    52 
    +
    53 
    +    def _get_cache_capabilities(self):
    54 
    +        capabilities = remote_execution_pb2.CacheCapabilities()
    55 
    +        action_cache_update_capabilities = remote_execution_pb2.ActionCacheUpdateCapabilities()
    56 
    +
    57 
    +        if self.__cas_instance:
    58 
    +            capabilities.digest_function.extend([self.__cas_instance.hash_type()])
    59 
    +            capabilities.max_batch_total_size_bytes = self.__cas_instance.max_batch_total_size_bytes()
    60 
    +            capabilities.symlink_absolute_path_strategy = self.__cas_instance.symlink_absolute_path_strategy()
    61 
    +            # TODO: execution priority #102
    62 
    +            # capabilities.cache_priority_capabilities =
    63 
    +
    64 
    +        if self.__action_cache_instance:
    65 
    +            action_cache_update_capabilities.update_enabled = self.__action_cache_instance.allow_updates
    66 
    +
    67 
    +        capabilities.action_cache_update_capabilities.CopyFrom(action_cache_update_capabilities)
    68 
    +        return capabilities
    69 
    +
    70 
    +    def _get_capabilities_execution(self):
    71 
    +        capabilities = remote_execution_pb2.ExecutionCapabilities()
    72 
    +        if self.__execution_instance:
    73 
    +            capabilities.exec_enabled = True
    74 
    +            capabilities.digest_function = self.__execution_instance.hash_type()
    75 
    +            # TODO: execution priority #102
    76 
    +            # capabilities.execution_priority =
    77 
    +
    78 
    +        else:
    79 
    +            capabilities.exec_enabled = False
    80 
    +
    81 
    +        return capabilities

  • buildgrid/server/capabilities/service.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 
    +import logging
    17 
    +
    18 
    +import grpc
    19 
    +
    20 
    +from buildgrid._exceptions import InvalidArgumentError
    21 
    +from buildgrid._protos.build.bazel.remote.execution.v2 import remote_execution_pb2, remote_execution_pb2_grpc
    22 
    +
    23 
    +
    24 
    +class CapabilitiesService(remote_execution_pb2_grpc.CapabilitiesServicer):
    25 
    +
    26 
    +    def __init__(self, server):
    27 
    +        self.__logger = logging.getLogger(__name__)
    28 
    +        self.__instances = {}
    29 
    +        remote_execution_pb2_grpc.add_CapabilitiesServicer_to_server(self, server)
    30 
    +
    31 
    +    def add_instance(self, name, instance):
    32 
    +        self.__instances[name] = instance
    33 
    +
    34 
    +    def add_cas_instance(self, name, instance):
    35 
    +        self.__instances[name].add_cas_instance(instance)
    36 
    +
    37 
    +    def add_action_cache_instance(self, name, instance):
    38 
    +        self.__instances[name].add_action_cache_instance(instance)
    39 
    +
    40 
    +    def add_execution_instance(self, name, instance):
    41 
    +        self.__instances[name].add_execution_instance(instance)
    42 
    +
    43 
    +    def GetCapabilities(self, request, context):
    44 
    +        try:
    45 
    +            instance = self._get_instance(request.instance_name)
    46 
    +            return instance.get_capabilities()
    47 
    +
    48 
    +        except InvalidArgumentError as e:
    49 
    +            self.__logger.error(e)
    50 
    +            context.set_details(str(e))
    51 
    +            context.set_code(grpc.StatusCode.INVALID_ARGUMENT)
    52 
    +
    53 
    +        return remote_execution_pb2.ServerCapabilities()
    54 
    +
    55 
    +    def _get_instance(self, name):
    56 
    +        try:
    57 
    +            return self.__instances[name]
    58 
    +
    59 
    +        except KeyError:
    60 
    +            raise InvalidArgumentError("Instance doesn't exist on server: [{}]".format(name))

  • buildgrid/server/cas/instance.py
    ... ... @@ -25,6 +25,7 @@ from buildgrid._exceptions import InvalidArgumentError, NotFoundError, OutOfRang
    25 25 
     from buildgrid._protos.google.bytestream import bytestream_pb2
    26 26 
     from buildgrid._protos.build.bazel.remote.execution.v2 import remote_execution_pb2 as re_pb2
    27 27 
     from buildgrid.settings import HASH, HASH_LENGTH
    28 
    +from buildgrid.utils import get_hash_type
    28 29
    29 30
    30 31 
     class ContentAddressableStorageInstance:
    ... ... @@ -37,6 +38,19 @@ class ContentAddressableStorageInstance:
    37 38 
         def register_instance_with_server(self, instance_name, server):
    38 39 
             server.add_cas_instance(self, instance_name)
    39 40
    41 
    +    def hash_type(self):
    42 
    +        return get_hash_type()
    43 
    +
    44 
    +    def max_batch_total_size_bytes(self):
    45 
    +        # TODO: link with max size
    46 
    +        # Should be added from settings in MR !119
    47 
    +        return 2000000
    48 
    +
    49 
    +    def symlink_absolute_path_strategy(self):
    50 
    +        # Currently this strategy is hardcoded into BuildGrid
    51 
    +        # With no setting to reference
    52 
    +        return re_pb2.CacheCapabilities().DISALLOWED
    53 
    +
    40 54 
         def find_missing_blobs(self, blob_digests):
    41 55 
             storage = self._storage
    42 56 
             return re_pb2.FindMissingBlobsResponse(

  • buildgrid/server/execution/instance.py
    ... ... @@ -25,6 +25,7 @@ from buildgrid._exceptions import FailedPreconditionError, InvalidArgumentError
    25 25 
     from buildgrid._protos.build.bazel.remote.execution.v2.remote_execution_pb2 import Action
    26 26
    27 27 
     from ..job import Job
    28 
    +from ...utils import get_hash_type
    28 29
    29 30
    30 31 
     class ExecutionInstance:
    ... ... @@ -35,9 +36,16 @@ class ExecutionInstance:
    35 36 
             self._storage = storage
    36 37 
             self._scheduler = scheduler
    37 38
    39 
    +    @property
    40 
    +    def scheduler(self):
    41 
    +        return self._scheduler
    42 
    +
    38 43 
         def register_instance_with_server(self, instance_name, server):
    39 44 
             server.add_execution_instance(self, instance_name)
    40 45
    46 
    +    def hash_type(self):
    47 
    +        return get_hash_type()
    48 
    +
    41 49 
         def execute(self, action_digest, skip_cache_lookup, message_queue=None):
    42 50 
             """ Sends a job for execution.
    43 51 
             Queues an action and creates an Operation instance to be associated with

  • buildgrid/server/execution/service.py
    ... ... @@ -33,30 +33,84 @@ from buildgrid._protos.google.longrunning import operations_pb2
    33 33
    34 34 
     class ExecutionService(remote_execution_pb2_grpc.ExecutionServicer):
    35 35
    36 
    -    def __init__(self, server):
    36 
    +    def __init__(self, server, monitor=False):
    37 37 
             self.__logger = logging.getLogger(__name__)
    38 38
    39 
    +        self.__peers_by_instance = None
    40 
    +        self.__peers = None
    41 
    +
    39 42 
             self._instances = {}
    43 
    +
    40 44 
             remote_execution_pb2_grpc.add_ExecutionServicer_to_server(self, server)
    41 45
    42 
    -    def add_instance(self, name, instance):
    43 
    -        self._instances[name] = instance
    46 
    +        self._is_instrumented = monitor
    47 
    +
    48 
    +        if self._is_instrumented:
    49 
    +            self.__peers_by_instance = {}
    50 
    +            self.__peers = {}
    51 
    +
    52 
    +    # --- Public API ---
    53 
    +
    54 
    +    def add_instance(self, instance_name, instance):
    55 
    +        """Registers a new servicer instance.
    56 
    +
    57 
    +        Args:
    58 
    +            instance_name (str): The new instance's name.
    59 
    +            instance (ExecutionInstance): The new instance itself.
    60 
    +        """
    61 
    +        self._instances[instance_name] = instance
    62 
    +
    63 
    +        if self._is_instrumented:
    64 
    +            self.__peers_by_instance[instance_name] = set()
    65 
    +
    66 
    +    def get_scheduler(self, instance_name):
    67 
    +        """Retrieves a reference to the scheduler for an instance.
    68 
    +
    69 
    +        Args:
    70 
    +            instance_name (str): The name of the instance to query.
    71 
    +
    72 
    +        Returns:
    73 
    +            Scheduler: A reference to the scheduler for `instance_name`.
    74 
    +
    75 
    +        Raises:
    76 
    +            InvalidArgumentError: If no instance named `instance_name` exists.
    77 
    +        """
    78 
    +        instance = self._get_instance(instance_name)
    79 
    +
    80 
    +        return instance.scheduler
    81 
    +
    82 
    +    # --- Public API: Servicer ---
    44 83
    45 84 
         def Execute(self, request, context):
    85 
    +        """Handles ExecuteRequest messages.
    86 
    +
    87 
    +        Args:
    88 
    +            request (ExecuteRequest): The incoming RPC request.
    89 
    +            context (grpc.ServicerContext): Context for the RPC call.
    90 
    +        """
    46 91 
             self.__logger.debug("Execute request from [%s]", context.peer())
    47 92
    93 
    +        instance_name = request.instance_name
    94 
    +        message_queue = queue.Queue()
    95 
    +        peer = context.peer()
    96 
    +
    48 97 
             try:
    49 
    -            message_queue = queue.Queue()
    50 
    -            instance = self._get_instance(request.instance_name)
    98 
    +            instance = self._get_instance(instance_name)
    51 99 
                 operation = instance.execute(request.action_digest,
    52 100 
                                              request.skip_cache_lookup,
    53 101 
                                              message_queue)
    54 102
    55 
    -            context.add_callback(partial(instance.unregister_message_client,
    56 
    -                                         operation.name, message_queue))
    103 
    +            context.add_callback(partial(self._rpc_termination_callback,
    104 
    +                                         peer, instance_name, operation.name, message_queue))
    57 105
    58 
    -            instanced_op_name = "{}/{}".format(request.instance_name,
    59 
    -                                               operation.name)
    106 
    +            if self._is_instrumented:
    107 
    +                if peer not in self.__peers:
    108 
    +                    self.__peers_by_instance[instance_name].add(peer)
    109 
    +                    self.__peers[peer] = 1
    110 
    +                else:
    111 
    +                    self.__peers[peer] += 1
    112 
    +
    113 
    +            instanced_op_name = "{}/{}".format(instance_name, operation.name)
    60 114
    61 115 
                 self.__logger.info("Operation name: [%s]", instanced_op_name)
    62 116
    ... ... @@ -86,23 +140,33 @@ class ExecutionService(remote_execution_pb2_grpc.ExecutionServicer):
    86 140 
                 yield operations_pb2.Operation()
    87 141
    88 142 
         def WaitExecution(self, request, context):
    89 
    -        self.__logger.debug("WaitExecution request from [%s]", context.peer())
    143 
    +        """Handles WaitExecutionRequest messages.
    90 144
    91 
    -        try:
    92 
    -            names = request.name.split("/")
    145 
    +        Args:
    146 
    +            request (WaitExecutionRequest): The incoming RPC request.
    147 
    +            context (grpc.ServicerContext): Context for the RPC call.
    148 
    +        """
    149 
    +        self.__logger.debug("WaitExecution request from [%s]", context.peer())
    93 150
    94 
    -            # Operation name should be in format:
    95 
    -            # {instance/name}/{operation_id}
    96 
    -            instance_name = ''.join(names[0:-1])
    151 
    +        names = request.name.split('/')
    152 
    +        instance_name = '/'.join(names[:-1])
    153 
    +        operation_name = names[-1]
    154 
    +        message_queue = queue.Queue()
    155 
    +        peer = context.peer()
    97 156
    98 
    -            message_queue = queue.Queue()
    99 
    -            operation_name = names[-1]
    157 
    +        try:
    100 158 
                 instance = self._get_instance(instance_name)
    101 159
    102 160 
                 instance.register_message_client(operation_name, message_queue)
    161 
    +            context.add_callback(partial(self._rpc_termination_callback,
    162 
    +                                         peer, instance_name, operation_name, message_queue))
    103 163
    104 
    -            context.add_callback(partial(instance.unregister_message_client,
    105 
    -                                         operation_name, message_queue))
    164 
    +            if self._is_instrumented:
    165 
    +                if peer not in self.__peers:
    166 
    +                    self.__peers_by_instance[instance_name].add(peer)
    167 
    +                    self.__peers[peer] = 1
    168 
    +                else:
    169 
    +                    self.__peers[peer] += 1
    106 170
    107 171 
                 for operation in instance.stream_operation_updates(message_queue,
    108 172 
                                                                    operation_name):
    ... ... @@ -123,6 +187,39 @@ class ExecutionService(remote_execution_pb2_grpc.ExecutionServicer):
    123 187 
                 context.set_code(grpc.StatusCode.CANCELLED)
    124 188 
                 yield operations_pb2.Operation()
    125 189
    190 
    +    # --- Public API: Monitoring ---
    191 
    +
    192 
    +    @property
    193 
    +    def is_instrumented(self):
    194 
    +        return self._is_instrumented
    195 
    +
    196 
    +    def query_n_clients(self):
    197 
    +        if self.__peers is not None:
    198 
    +            return len(self.__peers)
    199 
    +        return 0
    200 
    +
    201 
    +    def query_n_clients_for_instance(self, instance_name):
    202 
    +        try:
    203 
    +            if self.__peers_by_instance is not None:
    204 
    +                return len(self.__peers_by_instance[instance_name])
    205 
    +        except KeyError:
    206 
    +            pass
    207 
    +        return 0
    208 
    +
    209 
    +    # --- Private API ---
    210 
    +
    211 
    +    def _rpc_termination_callback(self, peer, instance_name, job_name, message_queue):
    212 
    +        instance = self._get_instance(instance_name)
    213 
    +
    214 
    +        instance.unregister_message_client(job_name, message_queue)
    215 
    +
    216 
    +        if self._is_instrumented:
    217 
    +            if self.__peers[peer] > 1:
    218 
    +                self.__peers[peer] -= 1
    219 
    +            else:
    220 
    +                self.__peers_by_instance[instance_name].remove(peer)
    221 
    +                del self.__peers[peer]
    222 
    +
    126 223 
         def _get_instance(self, name):
    127 224 
             try:
    128 225 
                 return self._instances[name]

  • buildgrid/server/instance.py
    ... ... @@ -15,12 +15,16 @@
    15 15
    16 16 
     import asyncio
    17 17 
     from concurrent import futures
    18 
    +from datetime import timedelta
    18 19 
     import logging
    19 20 
     import os
    20 21 
     import signal
    22 
    +import time
    21 23
    22 24 
     import grpc
    23 25
    26 
    +from buildgrid._enums import BotStatus, MetricRecordDomain, MetricRecordType
    27 
    +from buildgrid._protos.buildgrid.v2 import monitoring_pb2
    24 28 
     from buildgrid.server.actioncache.service import ActionCacheService
    25 29 
     from buildgrid.server.bots.service import BotsService
    26 30 
     from buildgrid.server.cas.service import ByteStreamService, ContentAddressableStorageService
    ... ... @@ -28,6 +32,9 @@ from buildgrid.server.execution.service import ExecutionService
    28 32 
     from buildgrid.server._monitoring import MonitoringBus, MonitoringOutputType, MonitoringOutputFormat
    29 33 
     from buildgrid.server.operations.service import OperationsService
    30 34 
     from buildgrid.server.referencestorage.service import ReferenceStorageService
    35 
    +from buildgrid.server.capabilities.instance import CapabilitiesInstance
    36 
    +from buildgrid.server.capabilities.service import CapabilitiesService
    37 
    +from buildgrid.settings import MONITORING_PERIOD
    31 38
    32 39
    33 40 
     class BuildGridServer:
    ... ... @@ -55,6 +62,11 @@ class BuildGridServer:
    55 62 
             self.__main_loop = asyncio.get_event_loop()
    56 63 
             self.__monitoring_bus = None
    57 64
    65 
    +        self.__state_monitoring_task = None
    66 
    +
    67 
    +        # We always want a capabilities service
    68 
    +        self._capabilities_service = CapabilitiesService(self.__grpc_server)
    69 
    +
    58 70 
             self._execution_service = None
    59 71 
             self._bots_service = None
    60 72 
             self._operations_service = None
    ... ... @@ -63,6 +75,9 @@ class BuildGridServer:
    63 75 
             self._cas_service = None
    64 76 
             self._bytestream_service = None
    65 77
    78 
    +        self._schedulers = {}
    79 
    +        self._instances = set()
    80 
    +
    66 81 
             self._is_instrumented = monitor
    67 82
    68 83 
             if self._is_instrumented:
    ... ... @@ -79,6 +94,10 @@ class BuildGridServer:
    79 94 
             if self._is_instrumented:
    80 95 
                 self.__monitoring_bus.start()
    81 96
    97 
    +            self.__state_monitoring_task = asyncio.ensure_future(
    98 
    +                self._state_monitoring_worker(period=MONITORING_PERIOD),
    99 
    +                loop=self.__main_loop)
    100 
    +
    82 101 
             self.__main_loop.add_signal_handler(signal.SIGTERM, self.stop)
    83 102
    84 103 
             self.__main_loop.run_forever()
    ... ... @@ -86,6 +105,9 @@ class BuildGridServer:
    86 105 
         def stop(self):
    87 106 
             """Stops the BuildGrid server."""
    88 107 
             if self._is_instrumented:
    108 
    +            if self.__state_monitoring_task is not None:
    109 
    +                self.__state_monitoring_task.cancel()
    110 
    +
    89 111 
                 self.__monitoring_bus.stop()
    90 112
    91 113 
             self.__main_loop.stop()
    ... ... @@ -125,9 +147,14 @@ class BuildGridServer:
    125 147 
                 instance_name (str): Instance name.
    126 148 
             """
    127 149 
             if self._execution_service is None:
    128 
    -            self._execution_service = ExecutionService(self.__grpc_server)
    150 
    +            self._execution_service = ExecutionService(
    151 
    +                self.__grpc_server, monitor=self._is_instrumented)
    129 152
    130 153 
             self._execution_service.add_instance(instance_name, instance)
    154 
    +        self._add_capabilities_instance(instance_name, execution_instance=instance)
    155 
    +
    156 
    +        self._schedulers[instance_name] = instance.scheduler
    157 
    +        self._instances.add(instance_name)
    131 158
    132 159 
         def add_bots_interface(self, instance, instance_name):
    133 160 
             """Adds a :obj:`BotsInterface` to the service.
    ... ... @@ -139,10 +166,13 @@ class BuildGridServer:
    139 166 
                 instance_name (str): Instance name.
    140 167 
             """
    141 168 
             if self._bots_service is None:
    142 
    -            self._bots_service = BotsService(self.__grpc_server)
    169 
    +            self._bots_service = BotsService(
    170 
    +                self.__grpc_server, monitor=self._is_instrumented)
    143 171
    144 172 
             self._bots_service.add_instance(instance_name, instance)
    145 173
    174 
    +        self._instances.add(instance_name)
    175 
    +
    146 176 
         def add_operations_instance(self, instance, instance_name):
    147 177 
             """Adds an :obj:`OperationsInstance` to the service.
    148 178
    ... ... @@ -184,9 +214,10 @@ class BuildGridServer:
    184 214 
                 self._action_cache_service = ActionCacheService(self.__grpc_server)
    185 215
    186 216 
             self._action_cache_service.add_instance(instance_name, instance)
    217 
    +        self._add_capabilities_instance(instance_name, action_cache_instance=instance)
    187 218
    188 219 
         def add_cas_instance(self, instance, instance_name):
    189 
    -        """Stores a :obj:`ContentAddressableStorageInstance` to the service.
    220 
    +        """Adds a :obj:`ContentAddressableStorageInstance` to the service.
    190 221
    191 222 
             If no service exists, it creates one.
    192 223
    ... ... @@ -198,9 +229,10 @@ class BuildGridServer:
    198 229 
                 self._cas_service = ContentAddressableStorageService(self.__grpc_server)
    199 230
    200 231 
             self._cas_service.add_instance(instance_name, instance)
    232 
    +        self._add_capabilities_instance(instance_name, cas_instance=instance)
    201 233
    202 234 
         def add_bytestream_instance(self, instance, instance_name):
    203 
    -        """Stores a :obj:`ByteStreamInstance` to the service.
    235 
    +        """Adds a :obj:`ByteStreamInstance` to the service.
    204 236
    205 237 
             If no service exists, it creates one.
    206 238
    ... ... @@ -218,3 +250,180 @@ class BuildGridServer:
    218 250 
         @property
    219 251 
         def is_instrumented(self):
    220 252 
             return self._is_instrumented
    253 
    +
    254 
    +    # --- Private API ---
    255 
    +
    256 
    +    def _add_capabilities_instance(self, instance_name,
    257 
    +                                   cas_instance=None,
    258 
    +                                   action_cache_instance=None,
    259 
    +                                   execution_instance=None):
    260 
    +        """Adds a :obj:`CapabilitiesInstance` to the service.
    261 
    +
    262 
    +        Args:
    263 
    +            instance (:obj:`CapabilitiesInstance`): Instance to add.
    264 
    +            instance_name (str): Instance name.
    265 
    +        """
    266 
    +
    267 
    +        try:
    268 
    +            if cas_instance:
    269 
    +                self._capabilities_service.add_cas_instance(instance_name, cas_instance)
    270 
    +            if action_cache_instance:
    271 
    +                self._capabilities_service.add_action_cache_instance(instance_name, action_cache_instance)
    272 
    +            if execution_instance:
    273 
    +                self._capabilities_service.add_execution_instance(instance_name, execution_instance)
    274 
    +
    275 
    +        except KeyError:
    276 
    +            capabilities_instance = CapabilitiesInstance(cas_instance,
    277 
    +                                                         action_cache_instance,
    278 
    +                                                         execution_instance)
    279 
    +            self._capabilities_service.add_instance(instance_name, capabilities_instance)
    280 
    +
    281 
    +    async def _state_monitoring_worker(self, period=1.0):
    282 
    +        """Periodically publishes state metrics to the monitoring bus."""
    283 
    +        async def __state_monitoring_worker():
    284 
    +            # Emit total clients count record:
    285 
    +            _, record = self._query_n_clients()
    286 
    +            await self.__monitoring_bus.send_record(record)
    287 
    +
    288 
    +            # Emit total bots count record:
    289 
    +            _, record = self._query_n_bots()
    290 
    +            await self.__monitoring_bus.send_record(record)
    291 
    +
    292 
    +            queue_times = []
    293 
    +            # Emits records by instance:
    294 
    +            for instance_name in self._instances:
    295 
    +                # Emit instance clients count record:
    296 
    +                _, record = self._query_n_clients_for_instance(instance_name)
    297 
    +                await self.__monitoring_bus.send_record(record)
    298 
    +
    299 
    +                # Emit instance bots count record:
    300 
    +                _, record = self._query_n_bots_for_instance(instance_name)
    301 
    +                await self.__monitoring_bus.send_record(record)
    302 
    +
    303 
    +                # Emit instance average queue time record:
    304 
    +                queue_time, record = self._query_am_queue_time_for_instance(instance_name)
    305 
    +                await self.__monitoring_bus.send_record(record)
    306 
    +                if queue_time:
    307 
    +                    queue_times.append(queue_time)
    308 
    +
    309 
    +            # Emits records by bot status:
    310 
    +            for bot_status in [BotStatus.OK, BotStatus.UNHEALTHY]:
    311 
    +                # Emit status bots count record:
    312 
    +                _, record = self._query_n_bots_for_status(bot_status)
    313 
    +                await self.__monitoring_bus.send_record(record)
    314 
    +
    315 
    +            # Emit overall average queue time record:
    316 
    +            if queue_times:
    317 
    +                am_queue_time = sum(queue_times, timedelta()) / len(queue_times)
    318 
    +            else:
    319 
    +                am_queue_time = timedelta()
    320 
    +            record = self._forge_timer_metric_record(
    321 
    +                MetricRecordDomain.STATE,
    322 
    +                'average-queue-time',
    323 
    +                am_queue_time)
    324 
    +
    325 
    +            await self.__monitoring_bus.send_record(record)
    326 
    +
    327 
    +        try:
    328 
    +            while True:
    329 
    +                start = time.time()
    330 
    +                await __state_monitoring_worker()
    331 
    +
    332 
    +                end = time.time()
    333 
    +                await asyncio.sleep(period - (end - start))
    334 
    +
    335 
    +        except asyncio.CancelledError:
    336 
    +            pass
    337 
    +
    338 
    +    def _forge_counter_metric_record(self, domain, name, count, metadata=None):
    339 
    +        counter_record = monitoring_pb2.MetricRecord()
    340 
    +
    341 
    +        counter_record.creation_timestamp.GetCurrentTime()
    342 
    +        counter_record.domain = domain.value
    343 
    +        counter_record.type = MetricRecordType.COUNTER.value
    344 
    +        counter_record.name = name
    345 
    +        counter_record.count = count
    346 
    +        if metadata is not None:
    347 
    +            counter_record.metadata.update(metadata)
    348 
    +
    349 
    +        return counter_record
    350 
    +
    351 
    +    def _forge_timer_metric_record(self, domain, name, duration, metadata=None):
    352 
    +        timer_record = monitoring_pb2.MetricRecord()
    353 
    +
    354 
    +        timer_record.creation_timestamp.GetCurrentTime()
    355 
    +        timer_record.domain = domain.value
    356 
    +        timer_record.type = MetricRecordType.TIMER.value
    357 
    +        timer_record.name = name
    358 
    +        timer_record.duration.FromTimedelta(duration)
    359 
    +        if metadata is not None:
    360 
    +            timer_record.metadata.update(metadata)
    361 
    +
    362 
    +        return timer_record
    363 
    +
    364 
    +    def _forge_gauge_metric_record(self, domain, name, value, metadata=None):
    365 
    +        gauge_record = monitoring_pb2.MetricRecord()
    366 
    +
    367 
    +        gauge_record.creation_timestamp.GetCurrentTime()
    368 
    +        gauge_record.domain = domain.value
    369 
    +        gauge_record.type = MetricRecordType.GAUGE.value
    370 
    +        gauge_record.name = name
    371 
    +        gauge_record.value = value
    372 
    +        if metadata is not None:
    373 
    +            gauge_record.metadata.update(metadata)
    374 
    +
    375 
    +        return gauge_record
    376 
    +
    377 
    +    # --- Private API: Monitoring ---
    378 
    +
    379 
    +    def _query_n_clients(self):
    380 
    +        """Queries the number of clients connected."""
    381 
    +        n_clients = self._execution_service.query_n_clients()
    382 
    +        gauge_record = self._forge_gauge_metric_record(
    383 
    +            MetricRecordDomain.STATE, 'clients-count', n_clients)
    384 
    +
    385 
    +        return n_clients, gauge_record
    386 
    +
    387 
    +    def _query_n_clients_for_instance(self, instance_name):
    388 
    +        """Queries the number of clients connected for a given instance"""
    389 
    +        n_clients = self._execution_service.query_n_clients_for_instance(instance_name)
    390 
    +        gauge_record = self._forge_gauge_metric_record(
    391 
    +            MetricRecordDomain.STATE, 'clients-count', n_clients,
    392 
    +            metadata={'instance-name': instance_name or 'void'})
    393 
    +
    394 
    +        return n_clients, gauge_record
    395 
    +
    396 
    +    def _query_n_bots(self):
    397 
    +        """Queries the number of bots connected."""
    398 
    +        n_bots = self._bots_service.query_n_bots()
    399 
    +        gauge_record = self._forge_gauge_metric_record(
    400 
    +            MetricRecordDomain.STATE, 'bots-count', n_bots)
    401 
    +
    402 
    +        return n_bots, gauge_record
    403 
    +
    404 
    +    def _query_n_bots_for_instance(self, instance_name):
    405 
    +        """Queries the number of bots connected for a given instance."""
    406 
    +        n_bots = self._bots_service.query_n_bots_for_instance(instance_name)
    407 
    +        gauge_record = self._forge_gauge_metric_record(
    408 
    +            MetricRecordDomain.STATE, 'bots-count', n_bots,
    409 
    +            metadata={'instance-name': instance_name or 'void'})
    410 
    +
    411 
    +        return n_bots, gauge_record
    412 
    +
    413 
    +    def _query_n_bots_for_status(self, bot_status):
    414 
    +        """Queries the number of bots connected for a given health status."""
    415 
    +        n_bots = self._bots_service.query_n_bots_for_status(bot_status)
    416 
    +        gauge_record = self._forge_gauge_metric_record(
    417 
    +            MetricRecordDomain.STATE, 'bots-count', n_bots,
    418 
    +            metadata={'bot-status': bot_status.name})
    419 
    +
    420 
    +        return n_bots, gauge_record
    421 
    +
    422 
    +    def _query_am_queue_time_for_instance(self, instance_name):
    423 
    +        """Queries the average job's queue time for a given instance."""
    424 
    +        am_queue_time = self._schedulers[instance_name].query_am_queue_time()
    425 
    +        timer_record = self._forge_timer_metric_record(
    426 
    +            MetricRecordDomain.STATE, 'average-queue-time', am_queue_time,
    427 
    +            metadata={'instance-name': instance_name or 'void'})
    428 
    +
    429 
    +        return am_queue_time, timer_record

  • buildgrid/server/job.py
    ... ... @@ -13,10 +13,11 @@
    13 13 
     # limitations under the License.
    14 14
    15 15
    16 
    +from datetime import datetime
    16 17 
     import logging
    17 18 
     import uuid
    18 19
    19 
    -from google.protobuf import timestamp_pb2
    20 
    +from google.protobuf import duration_pb2, timestamp_pb2
    20 21
    21 22 
     from buildgrid._enums import LeaseState, OperationStage
    22 23 
     from buildgrid._exceptions import CancelledError
    ... ... @@ -40,6 +41,7 @@ class Job:
    40 41 
             self.__operation_metadata = remote_execution_pb2.ExecuteOperationMetadata()
    41 42
    42 43 
             self.__queued_timestamp = timestamp_pb2.Timestamp()
    44 
    +        self.__queued_time_duration = duration_pb2.Duration()
    43 45 
             self.__worker_start_timestamp = timestamp_pb2.Timestamp()
    44 46 
             self.__worker_completed_timestamp = timestamp_pb2.Timestamp()
    45 47
    ... ... @@ -56,6 +58,8 @@ class Job:
    56 58 
             self._operation.done = False
    57 59 
             self._n_tries = 0
    58 60
    61 
    +    # --- Public API ---
    62 
    +
    59 63 
         @property
    60 64 
         def name(self):
    61 65 
             return self._name
    ... ... @@ -193,7 +197,7 @@ class Job:
    193 197 
                     result.Unpack(action_result)
    194 198
    195 199 
                 action_metadata = action_result.execution_metadata
    196 
    -            action_metadata.queued_timestamp.CopyFrom(self.__worker_start_timestamp)
    200 
    +            action_metadata.queued_timestamp.CopyFrom(self.__queued_timestamp)
    197 201 
                 action_metadata.worker_start_timestamp.CopyFrom(self.__worker_start_timestamp)
    198 202 
                 action_metadata.worker_completed_timestamp.CopyFrom(self.__worker_completed_timestamp)
    199 203
    ... ... @@ -227,6 +231,10 @@ class Job:
    227 231 
                     self.__queued_timestamp.GetCurrentTime()
    228 232 
                 self._n_tries += 1
    229 233
    234 
    +        elif self.__operation_metadata.stage == OperationStage.EXECUTING.value:
    235 
    +            queue_in, queue_out = self.__queued_timestamp.ToDatetime(), datetime.now()
    236 
    +            self.__queued_time_duration.FromTimedelta(queue_out - queue_in)
    237 
    +
    230 238 
             elif self.__operation_metadata.stage == OperationStage.COMPLETED.value:
    231 239 
                 if self.__execute_response is not None:
    232 240 
                     self._operation.response.Pack(self.__execute_response)
    ... ... @@ -260,3 +268,11 @@ class Job:
    260 268 
             self.__execute_response.status.message = "Operation cancelled by client."
    261 269
    262 270 
             self.update_operation_stage(OperationStage.COMPLETED)
    271 
    +
    272 
    +    # --- Public API: Monitoring ---
    273 
    +
    274 
    +    def query_queue_time(self):
    275 
    +        return self.__queued_time_duration.ToTimedelta()
    276 
    +
    277 
    +    def query_n_retries(self):
    278 
    +        return self._n_tries - 1 if self._n_tries > 0 else 0

  • buildgrid/server/operations/instance.py
    ... ... @@ -32,6 +32,10 @@ class OperationsInstance:
    32 32
    33 33 
             self._scheduler = scheduler
    34 34
    35 
    +    @property
    36 
    +    def scheduler(self):
    37 
    +        return self._scheduler
    38 
    +
    35 39 
         def register_instance_with_server(self, instance_name, server):
    36 40 
             server.add_operations_instance(self, instance_name)
    37 41

  • buildgrid/server/operations/service.py
    ... ... @@ -38,8 +38,18 @@ class OperationsService(operations_pb2_grpc.OperationsServicer):
    38 38
    39 39 
             operations_pb2_grpc.add_OperationsServicer_to_server(self, server)
    40 40
    41 
    -    def add_instance(self, name, instance):
    42 
    -        self._instances[name] = instance
    41 
    +    # --- Public API ---
    42 
    +
    43 
    +    def add_instance(self, instance_name, instance):
    44 
    +        """Registers a new servicer instance.
    45 
    +
    46 
    +        Args:
    47 
    +            instance_name (str): The new instance's name.
    48 
    +            instance (OperationsInstance): The new instance itself.
    49 
    +        """
    50 
    +        self._instances[instance_name] = instance
    51 
    +
    52 
    +    # --- Public API: Servicer ---
    43 53
    44 54 
         def GetOperation(self, request, context):
    45 55 
             self.__logger.debug("GetOperation request from [%s]", context.peer())
    ... ... @@ -127,6 +137,8 @@ class OperationsService(operations_pb2_grpc.OperationsServicer):
    127 137
    128 138 
             return Empty()
    129 139
    140 
    +    # --- Private API ---
    141 
    +
    130 142 
         def _parse_instance_name(self, name):
    131 143 
             """ If the instance name is not blank, 'name' will have the form
    132 144 
             {instance_name}/{operation_uuid}. Otherwise, it will just be

  • buildgrid/server/scheduler.py
    ... ... @@ -20,33 +20,70 @@ Schedules jobs.
    20 20 
     """
    21 21
    22 22 
     from collections import deque
    23 
    +from datetime import timedelta
    23 24 
     import logging
    24 25
    26 
    +from buildgrid._enums import LeaseState, OperationStage
    25 27 
     from buildgrid._exceptions import NotFoundError
    26 28
    27 
    -from .job import OperationStage, LeaseState
    28 
    -
    29 29
    30 30 
     class Scheduler:
    31 31
    32 32 
         MAX_N_TRIES = 5
    33 33
    34 
    -    def __init__(self, action_cache=None):
    34 
    +    def __init__(self, action_cache=None, monitor=False):
    35 35 
             self.__logger = logging.getLogger(__name__)
    36 36
    37 
    +        self.__operations_by_stage = None
    38 
    +        self.__leases_by_state = None
    39 
    +        self.__queue_time_average = None
    40 
    +        self.__retries_count = 0
    41 
    +
    37 42 
             self._action_cache = action_cache
    38 43 
             self.jobs = {}
    39 44 
             self.queue = deque()
    40 45
    46 
    +        self._is_instrumented = monitor
    47 
    +
    48 
    +        if self._is_instrumented:
    49 
    +            self.__operations_by_stage = {}
    50 
    +            self.__leases_by_state = {}
    51 
    +            self.__queue_time_average = 0, timedelta()
    52 
    +
    53 
    +            self.__operations_by_stage[OperationStage.CACHE_CHECK] = set()
    54 
    +            self.__operations_by_stage[OperationStage.QUEUED] = set()
    55 
    +            self.__operations_by_stage[OperationStage.EXECUTING] = set()
    56 
    +            self.__operations_by_stage[OperationStage.COMPLETED] = set()
    57 
    +
    58 
    +            self.__leases_by_state[LeaseState.PENDING] = set()
    59 
    +            self.__leases_by_state[LeaseState.ACTIVE] = set()
    60 
    +            self.__leases_by_state[LeaseState.COMPLETED] = set()
    61 
    +
    62 
    +    # --- Public API ---
    63 
    +
    41 64 
         def register_client(self, job_name, queue):
    42 
    -        self.jobs[job_name].register_client(queue)
    65 
    +        job = self.jobs[job_name]
    66 
    +
    67 
    +        job.register_client(queue)
    43 68
    44 69 
         def unregister_client(self, job_name, queue):
    45 
    -        self.jobs[job_name].unregister_client(queue)
    70 
    +        job = self.jobs[job_name]
    46 71
    47 
    -        if not self.jobs[job_name].n_clients and self.jobs[job_name].operation.done:
    72 
    +        job.unregister_client(queue)
    73 
    +
    74 
    +        if not job.n_clients and job.operation.done:
    48 75 
                 del self.jobs[job_name]
    49 76
    77 
    +            if self._is_instrumented:
    78 
    +                self.__operations_by_stage[OperationStage.CACHE_CHECK].discard(job_name)
    79 
    +                self.__operations_by_stage[OperationStage.QUEUED].discard(job_name)
    80 
    +                self.__operations_by_stage[OperationStage.EXECUTING].discard(job_name)
    81 
    +                self.__operations_by_stage[OperationStage.COMPLETED].discard(job_name)
    82 
    +
    83 
    +                self.__leases_by_state[LeaseState.PENDING].discard(job_name)
    84 
    +                self.__leases_by_state[LeaseState.ACTIVE].discard(job_name)
    85 
    +                self.__leases_by_state[LeaseState.COMPLETED].discard(job_name)
    86 
    +
    50 87 
         def queue_job(self, job, skip_cache_lookup=False):
    51 88 
             self.jobs[job.name] = job
    52 89
    ... ... @@ -62,23 +99,30 @@ class Scheduler:
    62 99 
                     job.set_cached_result(action_result)
    63 100 
                     operation_stage = OperationStage.COMPLETED
    64 101
    102 
    +                if self._is_instrumented:
    103 
    +                    self.__retries_count += 1
    104 
    +
    65 105 
             else:
    66 106 
                 operation_stage = OperationStage.QUEUED
    67 107 
                 self.queue.append(job)
    68 108
    69 
    -        job.update_operation_stage(operation_stage)
    109 
    +        self._update_job_operation_stage(job.name, operation_stage)
    70 110
    71 111 
         def retry_job(self, job_name):
    72 
    -        if job_name in self.jobs:
    73 
    -            job = self.jobs[job_name]
    74 
    -            if job.n_tries >= self.MAX_N_TRIES:
    75 
    -                # TODO: Decide what to do with these jobs
    76 
    -                job.update_operation_stage(OperationStage.COMPLETED)
    77 
    -                # TODO: Mark these jobs as done
    78 
    -            else:
    79 
    -                job.update_operation_stage(OperationStage.QUEUED)
    80 
    -                job.update_lease_state(LeaseState.PENDING)
    81 
    -                self.queue.append(job)
    112 
    +        job = self.jobs[job_name]
    113 
    +
    114 
    +        operation_stage = None
    115 
    +        if job.n_tries >= self.MAX_N_TRIES:
    116 
    +            # TODO: Decide what to do with these jobs
    117 
    +            operation_stage = OperationStage.COMPLETED
    118 
    +            # TODO: Mark these jobs as done
    119 
    +
    120 
    +        else:
    121 
    +            operation_stage = OperationStage.QUEUED
    122 
    +            job.update_lease_state(LeaseState.PENDING)
    123 
    +            self.queue.append(job)
    124 
    +
    125 
    +        self._update_job_operation_stage(job_name, operation_stage)
    82 126
    83 127 
         def list_jobs(self):
    84 128 
             return self.jobs.values()
    ... ... @@ -118,17 +162,27 @@ class Scheduler:
    118 162 
                 lease_result (google.protobuf.Any): the lease execution result, only
    119 163 
                     required if `lease_state` is `COMPLETED`.
    120 164 
             """
    121 
    -
    122 165 
             job = self.jobs[lease.id]
    123 166 
             lease_state = LeaseState(lease.state)
    124 167
    168 
    +        operation_stage = None
    125 169 
             if lease_state == LeaseState.PENDING:
    126 170 
                 job.update_lease_state(LeaseState.PENDING)
    127 
    -            job.update_operation_stage(OperationStage.QUEUED)
    171 
    +            operation_stage = OperationStage.QUEUED
    172 
    +
    173 
    +            if self._is_instrumented:
    174 
    +                self.__leases_by_state[LeaseState.PENDING].add(lease.id)
    175 
    +                self.__leases_by_state[LeaseState.ACTIVE].discard(lease.id)
    176 
    +                self.__leases_by_state[LeaseState.COMPLETED].discard(lease.id)
    128 177
    129 178 
             elif lease_state == LeaseState.ACTIVE:
    130 179 
                 job.update_lease_state(LeaseState.ACTIVE)
    131 
    -            job.update_operation_stage(OperationStage.EXECUTING)
    180 
    +            operation_stage = OperationStage.EXECUTING
    181 
    +
    182 
    +            if self._is_instrumented:
    183 
    +                self.__leases_by_state[LeaseState.PENDING].discard(lease.id)
    184 
    +                self.__leases_by_state[LeaseState.ACTIVE].add(lease.id)
    185 
    +                self.__leases_by_state[LeaseState.COMPLETED].discard(lease.id)
    132 186
    133 187 
             elif lease_state == LeaseState.COMPLETED:
    134 188 
                 job.update_lease_state(LeaseState.COMPLETED,
    ... ... @@ -137,7 +191,14 @@ class Scheduler:
    137 191 
                 if self._action_cache is not None and not job.do_not_cache:
    138 192 
                     self._action_cache.update_action_result(job.action_digest, job.action_result)
    139 193
    140 
    -            job.update_operation_stage(OperationStage.COMPLETED)
    194 
    +            operation_stage = OperationStage.COMPLETED
    195 
    +
    196 
    +            if self._is_instrumented:
    197 
    +                self.__leases_by_state[LeaseState.PENDING].discard(lease.id)
    198 
    +                self.__leases_by_state[LeaseState.ACTIVE].discard(lease.id)
    199 
    +                self.__leases_by_state[LeaseState.COMPLETED].add(lease.id)
    200 
    +
    201 
    +        self._update_job_operation_stage(lease.id, operation_stage)
    141 202
    142 203 
         def get_job_lease(self, job_name):
    143 204 
             """Returns the lease associated to job, if any have been emitted yet."""
    ... ... @@ -160,3 +221,101 @@ class Scheduler:
    160 221 
                 job_name (str): name of the job holding the operation to cancel.
    161 222 
             """
    162 223 
             self.jobs[job_name].cancel_operation()
    224 
    +
    225 
    +    # --- Public API: Monitoring ---
    226 
    +
    227 
    +    @property
    228 
    +    def is_instrumented(self):
    229 
    +        return self._is_instrumented
    230 
    +
    231 
    +    def query_n_jobs(self):
    232 
    +        return len(self.jobs)
    233 
    +
    234 
    +    def query_n_operations(self):
    235 
    +        # For now n_operations == n_jobs:
    236 
    +        return len(self.jobs)
    237 
    +
    238 
    +    def query_n_operations_by_stage(self, operation_stage):
    239 
    +        try:
    240 
    +            if self.__operations_by_stage is not None:
    241 
    +                return len(self.__operations_by_stage[operation_stage])
    242 
    +        except KeyError:
    243 
    +            pass
    244 
    +        return 0
    245 
    +
    246 
    +    def query_n_leases(self):
    247 
    +        return len(self.jobs)
    248 
    +
    249 
    +    def query_n_leases_by_state(self, lease_state):
    250 
    +        try:
    251 
    +            if self.__leases_by_state is not None:
    252 
    +                return len(self.__leases_by_state[lease_state])
    253 
    +        except KeyError:
    254 
    +            pass
    255 
    +        return 0
    256 
    +
    257 
    +    def query_n_retries(self):
    258 
    +        return self.__retries_count
    259 
    +
    260 
    +    def query_am_queue_time(self):
    261 
    +        if self.__queue_time_average is not None:
    262 
    +            return self.__queue_time_average[1]
    263 
    +        return timedelta()
    264 
    +
    265 
    +    # --- Private API ---
    266 
    +
    267 
    +    def _update_job_operation_stage(self, job_name, operation_stage):
    268 
    +        """Requests a stage transition for the job's :class:Operations.
    269 
    +
    270 
    +        Args:
    271 
    +            job_name (str): name of the job to query.
    272 
    +            operation_stage (OperationStage): the stage to transition to.
    273 
    +        """
    274 
    +        job = self.jobs[job_name]
    275 
    +
    276 
    +        if operation_stage == OperationStage.CACHE_CHECK:
    277 
    +            job.update_operation_stage(OperationStage.CACHE_CHECK)
    278 
    +
    279 
    +            if self._is_instrumented:
    280 
    +                self.__operations_by_stage[OperationStage.CACHE_CHECK].add(job_name)
    281 
    +                self.__operations_by_stage[OperationStage.QUEUED].discard(job_name)
    282 
    +                self.__operations_by_stage[OperationStage.EXECUTING].discard(job_name)
    283 
    +                self.__operations_by_stage[OperationStage.COMPLETED].discard(job_name)
    284 
    +
    285 
    +        elif operation_stage == OperationStage.QUEUED:
    286 
    +            job.update_operation_stage(OperationStage.QUEUED)
    287 
    +
    288 
    +            if self._is_instrumented:
    289 
    +                self.__operations_by_stage[OperationStage.CACHE_CHECK].discard(job_name)
    290 
    +                self.__operations_by_stage[OperationStage.QUEUED].add(job_name)
    291 
    +                self.__operations_by_stage[OperationStage.EXECUTING].discard(job_name)
    292 
    +                self.__operations_by_stage[OperationStage.COMPLETED].discard(job_name)
    293 
    +
    294 
    +        elif operation_stage == OperationStage.EXECUTING:
    295 
    +            job.update_operation_stage(OperationStage.EXECUTING)
    296 
    +
    297 
    +            if self._is_instrumented:
    298 
    +                self.__operations_by_stage[OperationStage.CACHE_CHECK].discard(job_name)
    299 
    +                self.__operations_by_stage[OperationStage.QUEUED].discard(job_name)
    300 
    +                self.__operations_by_stage[OperationStage.EXECUTING].add(job_name)
    301 
    +                self.__operations_by_stage[OperationStage.COMPLETED].discard(job_name)
    302 
    +
    303 
    +        elif operation_stage == OperationStage.COMPLETED:
    304 
    +            job.update_operation_stage(OperationStage.COMPLETED)
    305 
    +
    306 
    +            if self._is_instrumented:
    307 
    +                self.__operations_by_stage[OperationStage.CACHE_CHECK].discard(job_name)
    308 
    +                self.__operations_by_stage[OperationStage.QUEUED].discard(job_name)
    309 
    +                self.__operations_by_stage[OperationStage.EXECUTING].discard(job_name)
    310 
    +                self.__operations_by_stage[OperationStage.COMPLETED].add(job_name)
    311 
    +
    312 
    +                average_order, average_time = self.__queue_time_average
    313 
    +
    314 
    +                average_order += 1
    315 
    +                if average_order <= 1:
    316 
    +                    average_time = job.query_queue_time()
    317 
    +                else:
    318 
    +                    queue_time = job.query_queue_time()
    319 
    +                    average_time = average_time + ((queue_time - average_time) / average_order)
    320 
    +
    321 
    +                self.__queue_time_average = average_order, average_time

  • buildgrid/settings.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 
    +
    1 16 
     import hashlib
    2 17
    3 18
    4 
    -# The hash function that CAS uses
    19 
    +# Hash function used for computing digests:
    5 20 
     HASH = hashlib.sha256
    21 
    +
    22 
    +# Lenght in bytes of a hash string returned by HASH:
    6 23 
     HASH_LENGTH = HASH().digest_size * 2
    24 
    +
    25 
    +# Period, in seconds, for the monitoring cycle:
    26 
    +MONITORING_PERIOD = 5.0

  • buildgrid/utils.py
    ... ... @@ -30,6 +30,14 @@ def get_hostname():
    30 30 
         return socket.gethostname()
    31 31
    32 32
    33 
    +def get_hash_type():
    34 
    +    """Returns the hash type."""
    35 
    +    hash_name = HASH().name
    36 
    +    if hash_name == "sha256":
    37 
    +        return remote_execution_pb2.SHA256
    38 
    +    return remote_execution_pb2.UNKNOWN
    39 
    +
    40 
    +
    33 41 
     def create_digest(bytes_to_digest):
    34 42 
         """Computes the :obj:`Digest` of a piece of data.
    35 43

  • setup.py
    ... ... @@ -112,13 +112,15 @@ setup(
    112 112 
         license="Apache License, Version 2.0",
    113 113 
         description="A remote execution service",
    114 114 
         packages=find_packages(),
    115 
    +    python_requires='>= 3.5.3',  # janus requirement
    115 116 
         install_requires=[
    116 
    -        'protobuf',
    117 
    -        'grpcio',
    118 
    -        'Click',
    119 
    -        'PyYAML',
    120 117 
             'boto3 < 1.8.0',
    121 118 
             'botocore < 1.11.0',
    119 
    +        'click',
    120 
    +        'grpcio',
    121 
    +        'janus',
    122 
    +        'protobuf',
    123 
    +        'pyyaml',
    122 124 
         ],
    123 125 
         entry_points={
    124 126 
             'console_scripts': [

  • tests/integration/capabilities_service.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 
    +# pylint: disable=redefined-outer-name
    16 
    +
    17 
    +
    18 
    +import grpc
    19 
    +import pytest
    20 
    +
    21 
    +from buildgrid._protos.build.bazel.remote.execution.v2 import remote_execution_pb2
    22 
    +from buildgrid.client.capabilities import CapabilitiesInterface
    23 
    +from buildgrid.server.controller import ExecutionController
    24 
    +from buildgrid.server.actioncache.storage import ActionCache
    25 
    +from buildgrid.server.cas.instance import ContentAddressableStorageInstance
    26 
    +from buildgrid.server.cas.storage.lru_memory_cache import LRUMemoryCache
    27 
    +
    28 
    +from ..utils.utils import run_in_subprocess
    29 
    +from ..utils.capabilities import serve_capabilities_service
    30 
    +
    31 
    +
    32 
    +INSTANCES = ['', 'instance']
    33 
    +
    34 
    +
    35 
    +# Use subprocess to avoid creation of gRPC threads in main process
    36 
    +# See https://github.com/grpc/grpc/blob/master/doc/fork_support.md
    37 
    +# Multiprocessing uses pickle which protobufs don't work with
    38 
    +# Workaround wrapper to send messages as strings
    39 
    +class ServerInterface:
    40 
    +
    41 
    +    def __init__(self, remote):
    42 
    +        self.__remote = remote
    43 
    +
    44 
    +    def get_capabilities(self, instance_name):
    45 
    +
    46 
    +        def __get_capabilities(queue, remote, instance_name):
    47 
    +            interface = CapabilitiesInterface(grpc.insecure_channel(remote))
    48 
    +
    49 
    +            result = interface.get_capabilities(instance_name)
    50 
    +            queue.put(result.SerializeToString())
    51 
    +
    52 
    +        result = run_in_subprocess(__get_capabilities,
    53 
    +                                   self.__remote, instance_name)
    54 
    +
    55 
    +        capabilities = remote_execution_pb2.ServerCapabilities()
    56 
    +        capabilities.ParseFromString(result)
    57 
    +        return capabilities
    58 
    +
    59 
    +
    60 
    +@pytest.mark.parametrize('instance', INSTANCES)
    61 
    +def test_execution_not_available_capabilities(instance):
    62 
    +    with serve_capabilities_service([instance]) as server:
    63 
    +        server_interface = ServerInterface(server.remote)
    64 
    +        response = server_interface.get_capabilities(instance)
    65 
    +
    66 
    +        assert not response.execution_capabilities.exec_enabled
    67 
    +
    68 
    +
    69 
    +@pytest.mark.parametrize('instance', INSTANCES)
    70 
    +def test_execution_available_capabilities(instance):
    71 
    +    controller = ExecutionController()
    72 
    +
    73 
    +    with serve_capabilities_service([instance],
    74 
    +                                    execution_instance=controller.execution_instance) as server:
    75 
    +        server_interface = ServerInterface(server.remote)
    76 
    +        response = server_interface.get_capabilities(instance)
    77 
    +
    78 
    +        assert response.execution_capabilities.exec_enabled
    79 
    +        assert response.execution_capabilities.digest_function
    80 
    +
    81 
    +
    82 
    +@pytest.mark.parametrize('instance', INSTANCES)
    83 
    +def test_action_cache_allow_updates_capabilities(instance):
    84 
    +    storage = LRUMemoryCache(limit=256)
    85 
    +    action_cache = ActionCache(storage, max_cached_refs=256, allow_updates=True)
    86 
    +
    87 
    +    with serve_capabilities_service([instance],
    88 
    +                                    action_cache_instance=action_cache) as server:
    89 
    +        server_interface = ServerInterface(server.remote)
    90 
    +        response = server_interface.get_capabilities(instance)
    91 
    +
    92 
    +        assert response.cache_capabilities.action_cache_update_capabilities.update_enabled
    93 
    +
    94 
    +
    95 
    +@pytest.mark.parametrize('instance', INSTANCES)
    96 
    +def test_action_cache_not_allow_updates_capabilities(instance):
    97 
    +    storage = LRUMemoryCache(limit=256)
    98 
    +    action_cache = ActionCache(storage, max_cached_refs=256, allow_updates=False)
    99 
    +
    100 
    +    with serve_capabilities_service([instance],
    101 
    +                                    action_cache_instance=action_cache) as server:
    102 
    +        server_interface = ServerInterface(server.remote)
    103 
    +        response = server_interface.get_capabilities(instance)
    104 
    +
    105 
    +        assert not response.cache_capabilities.action_cache_update_capabilities.update_enabled
    106 
    +
    107 
    +
    108 
    +@pytest.mark.parametrize('instance', INSTANCES)
    109 
    +def test_cas_capabilities(instance):
    110 
    +    cas = ContentAddressableStorageInstance(None)
    111 
    +
    112 
    +    with serve_capabilities_service([instance],
    113 
    +                                    cas_instance=cas) as server:
    114 
    +        server_interface = ServerInterface(server.remote)
    115 
    +        response = server_interface.get_capabilities(instance)
    116 
    +
    117 
    +        assert len(response.cache_capabilities.digest_function) == 1
    118 
    +        assert response.cache_capabilities.digest_function[0]
    119 
    +        assert response.cache_capabilities.symlink_absolute_path_strategy
    120 
    +        assert response.cache_capabilities.max_batch_total_size_bytes

  • tests/utils/capabilities.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 
    +from concurrent import futures
    17 
    +from contextlib import contextmanager
    18 
    +import multiprocessing
    19 
    +import os
    20 
    +import signal
    21 
    +
    22 
    +import grpc
    23 
    +import pytest_cov
    24 
    +
    25 
    +from buildgrid.server.capabilities.service import CapabilitiesService
    26 
    +from buildgrid.server.capabilities.instance import CapabilitiesInstance
    27 
    +
    28 
    +
    29 
    +@contextmanager
    30 
    +def serve_capabilities_service(instances,
    31 
    +                               cas_instance=None,
    32 
    +                               action_cache_instance=None,
    33 
    +                               execution_instance=None):
    34 
    +    server = Server(instances,
    35 
    +                    cas_instance,
    36 
    +                    action_cache_instance,
    37 
    +                    execution_instance)
    38 
    +    try:
    39 
    +        yield server
    40 
    +    finally:
    41 
    +        server.quit()
    42 
    +
    43 
    +
    44 
    +class Server:
    45 
    +
    46 
    +    def __init__(self, instances,
    47 
    +                 cas_instance=None,
    48 
    +                 action_cache_instance=None,
    49 
    +                 execution_instance=None):
    50 
    +        self.instances = instances
    51 
    +
    52 
    +        self.__queue = multiprocessing.Queue()
    53 
    +        self.__process = multiprocessing.Process(
    54 
    +            target=Server.serve,
    55 
    +            args=(self.__queue, self.instances, cas_instance, action_cache_instance, execution_instance))
    56 
    +        self.__process.start()
    57 
    +
    58 
    +        self.port = self.__queue.get(timeout=1)
    59 
    +        self.remote = 'localhost:{}'.format(self.port)
    60 
    +
    61 
    +    @staticmethod
    62 
    +    def serve(queue, instances, cas_instance, action_cache_instance, execution_instance):
    63 
    +        pytest_cov.embed.cleanup_on_sigterm()
    64 
    +
    65 
    +        # Use max_workers default from Python 3.5+
    66 
    +        max_workers = (os.cpu_count() or 1) * 5
    67 
    +        server = grpc.server(futures.ThreadPoolExecutor(max_workers))
    68 
    +        port = server.add_insecure_port('localhost:0')
    69 
    +
    70 
    +        capabilities_service = CapabilitiesService(server)
    71 
    +        for name in instances:
    72 
    +            capabilities_instance = CapabilitiesInstance(cas_instance, action_cache_instance, execution_instance)
    73 
    +            capabilities_service.add_instance(name, capabilities_instance)
    74 
    +
    75 
    +        server.start()
    76 
    +        queue.put(port)
    77 
    +        signal.pause()
    78 
    +
    79 
    +    def quit(self):
    80 
    +        if self.__process:
    81 
    +            self.__process.terminate()
    82 
    +            self.__process.join()

    • —
    View it on GitLab.
    You're receiving this email because of your account on gitlab.com. If you'd like to receive fewer emails, you can adjust your notification settings.

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