Commit e4fb8842 authored by Raoul Hidalgo Charman's avatar Raoul Hidalgo Charman
Browse files

Smarter bot calls 

Bots now have a timeout option, which the server looks at to determine how
long they wait on a job becoming available. As soon as a job becomes available
server responds with a lease. If it gets to the timeout then the server will
tell the bot there is no work and it will then update again and wait.
parent 589ffa40

buildgrid/_app/commands/cmd_bot.py

+6 −6
Original line number Diff line number Diff line
@@ -33,6 +33,7 @@ from buildgrid.bot.bot_session import BotSession, Device, Worker 


from ..bots import buildbox, dummy, host

from ..cli import pass_context

from ...settings import INTERVAL_BUFFER





@click.group(name='bot', short_help="Create and register bot clients.")
@@ -52,7 +53,7 @@ from ..cli import pass_context 
              help="Public CAS client certificate for TLS (PEM-encoded)")

@click.option('--cas-server-cert', type=click.Path(exists=True, dir_okay=False), default=None,

              help="Public CAS server certificate for TLS (PEM-encoded)")

@click.option('--update-period', type=click.FLOAT, default=0.5, show_default=True,

@click.option('--update-period', type=click.FLOAT, default=30, show_default=True,

              help="Time period for bot updates to the server in seconds.")

@click.option('--parent', type=click.STRING, default='main', show_default=True,

              help="Targeted farm resource.")
@@ -64,7 +65,6 @@ def cli(context, parent, update_period, remote, client_key, client_cert, server_ 


    context.remote = '{}:{}'.format(url.hostname, url.port or 50051)

    context.remote_url = remote

    context.update_period = update_period

    context.parent = parent



    if url.scheme == 'http':
@@ -123,7 +123,7 @@ def cli(context, parent, update_period, remote, client_key, client_cert, server_ 
    context.logger = logging.getLogger(__name__)

    context.logger.debug("Starting for remote {}".format(context.remote))



    interface = bot_interface.BotInterface(context.channel)

    interface = bot_interface.BotInterface(context.channel, update_period)



    worker = Worker()

    worker.add_device(Device())
@@ -141,7 +141,7 @@ def run_dummy(context): 
    Creates a session, accepts leases, does fake work and updates the server.

    """

    try:

        b = bot.Bot(context.bot_session, context.update_period)

        b = bot.Bot(context.bot_session)

        b.session(dummy.work_dummy,

                  context)

    except KeyboardInterrupt:
@@ -156,7 +156,7 @@ def run_host_tools(context): 
    result back to CAS.

    """

    try:

        b = bot.Bot(context.bot_session, context.update_period)

        b = bot.Bot(context.bot_session)

        b.session(host.work_host_tools,

                  context)

    except KeyboardInterrupt:
@@ -177,7 +177,7 @@ def run_buildbox(context, local_cas, fuse_dir): 
    context.fuse_dir = fuse_dir



    try:

        b = bot.Bot(context.bot_session, context.update_period)

        b = bot.Bot(context.bot_session)

        b.session(buildbox.work_buildbox,

                  context)

    except KeyboardInterrupt:

buildgrid/bot/bot.py

+1 −3
Original line number Diff line number Diff line
@@ -29,11 +29,10 @@ class Bot: 
    Creates a local BotSession.

    """



    def __init__(self, bot_session, update_period=1):

    def __init__(self, bot_session):

        self.logger = logging.getLogger(__name__)



        self._bot_session = bot_session

        self._update_period = update_period



    def session(self, work, context):

        loop = asyncio.get_event_loop()
@@ -55,4 +54,3 @@ class Bot: 
        """

        while True:

            self._bot_session.update_bot_session()
 
            await asyncio.sleep(self._update_period)

buildgrid/bot/bot_interface.py

+3 −2
Original line number Diff line number Diff line
@@ -30,10 +30,11 @@ class BotInterface: 
    Interface handles calls to the server.

    """



    def __init__(self, channel):

    def __init__(self, channel, interval):

        self.logger = logging.getLogger(__name__)

        self.logger.info(channel)

        self._stub = bots_pb2_grpc.BotsStub(channel)

        self._interval = interval



    def create_bot_session(self, parent, bot_session):

        request = bots_pb2.CreateBotSessionRequest(parent=parent,
@@ -44,4 +45,4 @@ class BotInterface: 
        request = bots_pb2.UpdateBotSessionRequest(name=bot_session.name,

                                                   bot_session=bot_session,

                                                   update_mask=update_mask)

        return self._stub.UpdateBotSession(request)
 
        return self._stub.UpdateBotSession(request, timeout=self._interval)

buildgrid/server/bots/instance.py

+5 −2
Original line number Diff line number Diff line
@@ -26,6 +26,7 @@ import uuid 
from buildgrid._exceptions import InvalidArgumentError, OutOfSyncError



from ..job import LeaseState

from ...settings import INTERVAL_BUFFER





class BotsInterface:
@@ -73,7 +74,7 @@ class BotsInterface: 


        return bot_session



    def update_bot_session(self, name, bot_session):

    def update_bot_session(self, name, bot_session, deadline=None):

        """ Client updates the server. Any changes in state to the Lease should be

        registered server side. Assigns available leases with work.

        """
@@ -87,7 +88,9 @@ class BotsInterface: 


        # TODO: Send worker capabilities to the scheduler!

        if not bot_session.leases:

            leases = self._scheduler.request_job_leases({})

            leases = self._scheduler.request_job_leases(

                {}, block=True if deadline else None,

                timeout=deadline - INTERVAL_BUFFER if deadline else None)

            if leases:

                bot_session.leases.extend(leases)

buildgrid/server/bots/service.py

+3 −1
Original line number Diff line number Diff line
@@ -64,8 +64,10 @@ class BotsService(bots_pb2_grpc.BotsServicer): 
            instance_name = ''.join(names[0:-1])



            instance = self._get_instance(instance_name)

            # server side context time_remaining is maxint - unix time

            return instance.update_bot_session(request.name,

                                               request.bot_session)

                                               request.bot_session,

                                               context.time_remaining())



        except InvalidArgumentError as e:

            self.logger.error(e)