Source code for apache_beam.testing.benchmarks.nexmark.nexmark_launcher

#
# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements.  See the NOTICE file distributed with
# this work for additional information regarding copyright ownership.
# The ASF licenses this file to You under the Apache License, Version 2.0
# (the "License"); you may not use this file except in compliance with
# the License.  You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#

"""Nexmark launcher.

The Nexmark suite is a series of queries (streaming pipelines) performed
on a simulation of auction events. The launcher orchestrates the generation
and parsing of streaming events and the running of queries.

Model
  - Person: Author of an auction or a bid.
  - Auction: Item under auction.
  - Bid: A bid for an item under auction.

Events
 - Create Person
 - Create Auction
 - Create Bid

Queries
  - Query0: Pass through (send and receive auction events).

Usage
  - DirectRunner
      python nexmark_launcher.py \
          --query/q <query number> \
          --project <project id> \
          --loglevel=DEBUG (optional) \
          --wait_until_finish_duration <time_in_ms> \
          --streaming

  - DataflowRunner
      python nexmark_launcher.py \
          --query/q <query number> \
          --project <project id> \
          --region <GCE region> \
          --loglevel=DEBUG (optional) \
          --wait_until_finish_duration <time_in_ms> \
          --streaming \
          --sdk_location <apache_beam tar.gz> \
          --staging_location=gs://... \
          --temp_location=gs://

"""

# pytype: skip-file

import argparse
import logging
import time
import uuid

import apache_beam as beam
from apache_beam.options.pipeline_options import GoogleCloudOptions
from apache_beam.options.pipeline_options import PipelineOptions
from apache_beam.options.pipeline_options import SetupOptions
from apache_beam.options.pipeline_options import StandardOptions
from apache_beam.runners import PipelineState
from apache_beam.testing.benchmarks.nexmark import nexmark_util
from apache_beam.testing.benchmarks.nexmark.monitor import Monitor
from apache_beam.testing.benchmarks.nexmark.monitor import MonitorSuffix
from apache_beam.testing.benchmarks.nexmark.nexmark_perf import NexmarkPerf
from apache_beam.testing.benchmarks.nexmark.queries import query0
from apache_beam.testing.benchmarks.nexmark.queries import query1
from apache_beam.testing.benchmarks.nexmark.queries import query2
from apache_beam.testing.benchmarks.nexmark.queries import query3
from apache_beam.testing.benchmarks.nexmark.queries import query4
from apache_beam.testing.benchmarks.nexmark.queries import query5
from apache_beam.testing.benchmarks.nexmark.queries import query6
from apache_beam.testing.benchmarks.nexmark.queries import query7
from apache_beam.testing.benchmarks.nexmark.queries import query8
from apache_beam.testing.benchmarks.nexmark.queries import query9
from apache_beam.testing.benchmarks.nexmark.queries import query10
from apache_beam.testing.benchmarks.nexmark.queries import query11
from apache_beam.testing.benchmarks.nexmark.queries import query12
from apache_beam.transforms import window


[docs]class NexmarkLauncher(object): # how long after some result is seen and no activity seen do we cancel job DONE_DELAY = 5 * 60 # delay in seconds between sample perf data PERF_DELAY = 20 # delay before cancelling the job when pipeline appears to be stuck TERMINATE_DELAY = 1 * 60 * 60 # delay before warning when pipeline appears to be stuck WARNING_DELAY = 10 * 60 def __init__(self): self.parse_args() self.manage_resources = self.args.manage_resources self.uuid = str(uuid.uuid4()) if self.manage_resources else '' self.topic_name = ( self.args.topic_name + self.uuid if self.args.topic_name else None) self.subscription_name = ( self.args.subscription_name + self.uuid if self.args.subscription_name else None) self.pubsub_mode = self.args.pubsub_mode if self.manage_resources: from google.cloud import pubsub self.cleanup() publish_client = pubsub.Client(project=self.project) topic = publish_client.topic(self.topic_name) logging.info('creating topic %s', self.topic_name) topic.create() sub = topic.subscription(self.subscription_name) logging.info('creating sub %s', self.topic_name) sub.create()
[docs] def parse_args(self): parser = argparse.ArgumentParser() parser.add_argument( '--query', '-q', type=int, action='append', required=True, choices=[i for i in range(13)], help='Query to run') parser.add_argument( '--subscription_name', type=str, help='Pub/Sub subscription to read from') parser.add_argument( '--topic_name', type=str, help='Pub/Sub topic to read from') parser.add_argument( '--loglevel', choices=['DEBUG', 'INFO', 'WARNING', 'ERROR', 'CRITICAL'], default='INFO', help='Set logging level to debug') parser.add_argument( '--input', type=str, help='Path to the data file containing nexmark events.') parser.add_argument( '--num_events', type=int, default=100000, help='number of events expected to process') parser.add_argument( '--manage_resources', default=False, action='store_true', help='If true, manage the creation and cleanup of topics and ' 'subscriptions.') parser.add_argument( '--pubsub_mode', type=str, default='SUBSCRIBE_ONLY', choices=['PUBLISH_ONLY', 'SUBSCRIBE_ONLY', 'COMBINED'], help='Pubsub mode used in the pipeline.') self.args, self.pipeline_args = parser.parse_known_args() logging.basicConfig( level=getattr(logging, self.args.loglevel, None), format='(%(threadName)-10s) %(message)s') self.pipeline_options = PipelineOptions(self.pipeline_args) logging.debug('args, pipeline_args: %s, %s', self.args, self.pipeline_args) # Usage with Dataflow requires a project to be supplied. self.project = self.pipeline_options.view_as(GoogleCloudOptions).project self.streaming = self.pipeline_options.view_as(StandardOptions).streaming if self.streaming: if self.args.subscription_name is None or self.project is None: raise ValueError( 'argument --subscription_name and --project ' + 'are required when running in streaming mode') else: if self.args.input is None: raise ValueError( 'argument --input is required when running in batch mode') # We use the save_main_session option because one or more DoFn's in this # workflow rely on global context (e.g., a module imported at module level). self.pipeline_options.view_as(SetupOptions).save_main_session = True
[docs] def generate_events(self): from google.cloud import pubsub publish_client = pubsub.Client(project=self.project) topic = publish_client.topic(self.topic_name) logging.info('Generating auction events to topic %s', topic.name) if self.args.input.startswith('gs://'): from apache_beam.io.gcp.gcsfilesystem import GCSFileSystem fs = GCSFileSystem(self.pipeline_options) with fs.open(self.args.input) as infile: for line in infile: topic.publish(line) else: with open(self.args.input) as infile: for line in infile: topic.publish(line) logging.info('Finished event generation.')
[docs] def read_from_file(self): return ( self.pipeline | 'reading_from_file' >> beam.io.ReadFromText(self.args.input) | 'deserialization' >> beam.ParDo(nexmark_util.ParseJsonEventFn()) | 'timestamping' >> beam.Map(lambda e: window.TimestampedValue(e, e.date_time)))
[docs] def read_from_pubsub(self): # Read from PubSub into a PCollection. if self.subscription_name: raw_events = self.pipeline | 'ReadPubSub_sub' >> beam.io.ReadFromPubSub( subscription=self.subscription_name, with_attributes=True, timestamp_attribute='timestamp') else: raw_events = self.pipeline | 'ReadPubSub_topic' >> beam.io.ReadFromPubSub( topic=self.topic_name, with_attributes=True, timestamp_attribute='timestamp') events = ( raw_events | 'pubsub_unwrap' >> beam.Map(lambda m: m.data) | 'deserialization' >> beam.ParDo(nexmark_util.ParseJsonEventFn())) return events
[docs] def run_query(self, query, query_args, pipeline_options, query_errors): try: self.pipeline = beam.Pipeline(options=self.pipeline_options) nexmark_util.setup_coder() event_monitor = Monitor('.events', 'event') result_monitor = Monitor('.results', 'result') if self.streaming: if self.pubsub_mode != 'SUBSCRIBE_ONLY': self.generate_events() if self.pubsub_mode == 'PUBLISH_ONLY': return events = self.read_from_pubsub() else: events = self.read_from_file() events = events | 'event_monitor' >> beam.ParDo(event_monitor.doFn) output = query.load(events, query_args, pipeline_options) output | 'result_monitor' >> beam.ParDo(result_monitor.doFn) # pylint: disable=expression-not-assigned result = self.pipeline.run() if not self.streaming: result.wait_until_finish() perf = self.monitor(result, event_monitor, result_monitor) self.log_performance(perf) except Exception as exc: query_errors.append(str(exc)) raise
[docs] def monitor(self, job, event_monitor, result_monitor): """ keep monitoring the performance and progress of running job and cancel the job if the job is stuck or seems to have finished running Returns: the final performance if it is measured """ logging.info('starting to monitor the job') last_active_ms = -1 perf = None cancel_job = False waiting_for_shutdown = False while True: now = int(time.time() * 1000) # current time in ms logging.debug('now is %d', now) curr_perf = NexmarkLauncher.get_performance( job, event_monitor, result_monitor) if perf is None or curr_perf.has_progress(perf): last_active_ms = now # only judge if the job should be cancelled if it is streaming job and # has not been shut down already if self.streaming and not waiting_for_shutdown: quiet_duration = (now - last_active_ms) // 1000 if (curr_perf.event_count >= self.args.num_events and curr_perf.result_count >= 0 and quiet_duration > self.DONE_DELAY): # we think the job is finished if expected input count has been seen # and no new results have been produced for a while logging.info('streaming query appears to have finished executing') waiting_for_shutdown = True cancel_job = True elif quiet_duration > self.TERMINATE_DELAY: logging.error( 'streaming query have been stuck for %d seconds', quiet_duration) logging.error('canceling streaming job') waiting_for_shutdown = True cancel_job = True elif quiet_duration > self.WARNING_DELAY: logging.warning( 'streaming query have been stuck for %d seconds', quiet_duration) if cancel_job: job.cancel() perf = curr_perf stopped = PipelineState.is_terminal(job.state) if stopped: break if not waiting_for_shutdown: if last_active_ms == now: logging.info('activity seen, new performance data extracted') else: logging.info('no activity seen') else: logging.info('waiting for shutdown') time.sleep(self.PERF_DELAY) return perf
[docs] @staticmethod def log_performance(perf): # type: (NexmarkPerf) -> None logging.info( 'input event count: %d, output event count: %d' % (perf.event_count, perf.result_count)) logging.info( 'query run took %.1f seconds and processed %.1f events per second' % (perf.runtime_sec, perf.event_per_sec))
[docs] @staticmethod def get_performance(result, event_monitor, result_monitor): event_count = nexmark_util.get_counter_metric( result, event_monitor.namespace, event_monitor.name_prefix + MonitorSuffix.ELEMENT_COUNTER) event_start = nexmark_util.get_start_time_metric( result, event_monitor.namespace, event_monitor.name_prefix + MonitorSuffix.EVENT_TIME) event_end = nexmark_util.get_end_time_metric( result, event_monitor.namespace, event_monitor.name_prefix + MonitorSuffix.EVENT_TIME) result_count = nexmark_util.get_counter_metric( result, result_monitor.namespace, result_monitor.name_prefix + MonitorSuffix.ELEMENT_COUNTER) result_end = nexmark_util.get_end_time_metric( result, result_monitor.namespace, result_monitor.name_prefix + MonitorSuffix.EVENT_TIME) perf = NexmarkPerf() perf.event_count = event_count perf.result_count = result_count effective_end = max(event_end, result_end) if effective_end >= 0 and event_start >= 0: perf.runtime_sec = (effective_end - event_start) / 1000 if event_count >= 0 and perf.runtime_sec > 0: perf.event_per_sec = event_count / perf.runtime_sec return perf
[docs] def cleanup(self): if self.manage_resources: from google.cloud import pubsub publish_client = pubsub.Client(project=self.project) topic = publish_client.topic(self.topic_name) if topic.exists(): logging.info('deleting topic %s', self.topic_name) topic.delete() sub = topic.subscription(self.subscription_name) if sub.exists(): logging.info('deleting sub %s', self.topic_name) sub.delete()
[docs] def run(self): queries = { 0: query0, 1: query1, 2: query2, 3: query3, 4: query4, 5: query5, 6: query6, 7: query7, 8: query8, 9: query9, 10: query10, 11: query11, 12: query12 } # TODO(mariagh): Move to a config file. query_args = { 'auction_skip': 123, 'window_size_sec': 10, 'window_period_sec': 5, 'fanout': 5, 'num_max_workers': 5, 'max_log_events': 100000, 'occasional_delay_sec': 3, 'max_auction_waiting_time': 600 } query_errors = [] for i in self.args.query: logging.info('Running query %d', i) self.run_query( queries[i], query_args, self.pipeline_options, query_errors=query_errors) if query_errors: logging.error('Query failed with %s', ', '.join(query_errors)) else: logging.info('Queries run: %s', self.args.query)
if __name__ == '__main__': launcher = NexmarkLauncher() launcher.run() launcher.cleanup()