You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@beam.apache.org by "Maximilian Michels (Jira)" <ji...@apache.org> on 2020/09/02 08:41:00 UTC

[jira] [Created] (BEAM-10848) Gauge metrics error when setting timers

Maximilian Michels created BEAM-10848:
-----------------------------------------

             Summary: Gauge metrics error when setting timers
                 Key: BEAM-10848
                 URL: https://issues.apache.org/jira/browse/BEAM-10848
             Project: Beam
          Issue Type: Bug
          Components: sdk-py-harness
            Reporter: Maximilian Michels


Gauges are affected by setting timers leading to {{None}} values:

{noformat}
ERROR:apache_beam.runners.worker.sdk_worker:Error processing instruction 147. Original traceback is
Traceback (most recent call last):
  File "/Users/max/Consulting/Lyft/dev/streamperfbench/beamperfk8s/venv/lib/python3.7/site-packages/apache_beam/runners/worker/sdk_worker.py", line 253, in _execute
    response = task()
  File "/Users/max/Consulting/Lyft/dev/streamperfbench/beamperfk8s/venv/lib/python3.7/site-packages/apache_beam/runners/worker/sdk_worker.py", line 310, in <lambda>
    lambda: self.create_worker().do_instruction(request), request)
  File "/Users/max/Consulting/Lyft/dev/streamperfbench/beamperfk8s/venv/lib/python3.7/site-packages/apache_beam/runners/worker/sdk_worker.py", line 480, in do_instruction
    getattr(request, request_type), request.instruction_id)
  File "/Users/max/Consulting/Lyft/dev/streamperfbench/beamperfk8s/venv/lib/python3.7/site-packages/apache_beam/runners/worker/sdk_worker.py", line 516, in process_bundle
    monitoring_infos = bundle_processor.monitoring_infos()
  File "/Users/max/Consulting/Lyft/dev/streamperfbench/beamperfk8s/venv/lib/python3.7/site-packages/apache_beam/runners/worker/bundle_processor.py", line 1107, in monitoring_infos
    op.monitoring_infos(transform_id, dict(tag_to_pcollection_id)))
  File "apache_beam/runners/worker/operations.py", line 340, in apache_beam.runners.worker.operations.Operation.monitoring_infos
  File "apache_beam/runners/worker/operations.py", line 347, in apache_beam.runners.worker.operations.Operation.monitoring_infos
  File "apache_beam/runners/worker/operations.py", line 386, in apache_beam.runners.worker.operations.Operation.user_monitoring_infos
  File "apache_beam/metrics/execution.py", line 261, in apache_beam.metrics.execution.MetricsContainer.to_runner_api_monitoring_infos
  File "apache_beam/metrics/cells.py", line 222, in apache_beam.metrics.cells.GaugeCell.to_runner_api_monitoring_info
  File "/Users/max/Consulting/Lyft/dev/streamperfbench/beamperfk8s/venv/lib/python3.7/site-packages/apache_beam/metrics/monitoring_infos.py", line 222, in int64_user_gauge
    payload = _encode_gauge(coder, timestamp, value)
  File "/Users/max/Consulting/Lyft/dev/streamperfbench/beamperfk8s/venv/lib/python3.7/site-packages/apache_beam/metrics/monitoring_infos.py", line 397, in _encode_gauge
    coder.get_impl().encode_to_stream(value, stream, True)
  File "apache_beam/coders/coder_impl.py", line 690, in apache_beam.coders.coder_impl.VarIntCoderImpl.encode_to_stream
  File "apache_beam/coders/coder_impl.py", line 692, in apache_beam.coders.coder_impl.VarIntCoderImpl.encode_to_stream
TypeError: an integer is required
{noformat}

The application has the following structure:

{code:python}
class StatefulOperation(beam.DoFn):
  def __init__(self, state_size_per_key_bytes, use_processing_timer=False):
    self.state_size_per_key_bytes = state_size_per_key_bytes
    self.str_coder = StrUtf8Coder().get_impl()
    self.bytes_gauge = Metrics.gauge('synthetic', 'state_bytes')
    self.elements_gauge = Metrics.gauge('synthetic', 'state_elements')
    self.use_processing_timer = use_processing_timer

  state_spec = userstate.BagStateSpec('state', StrUtf8Coder())

  state_spec2 = userstate.CombiningValueStateSpec('state_size_bytes', combine_fn=sum)

  state_spec3 = userstate.CombiningValueStateSpec('num_state_entries', combine_fn=sum)

  event_timer_spec = userstate.TimerSpec('event_timer', beam.TimeDomain.WATERMARK)
  processing_timer_spec = userstate.TimerSpec('proc_timer', beam.TimeDomain.REAL_TIME)

  def process(self,
              element,
              timestamp=beam.DoFn.TimestampParam,
              state=beam.DoFn.StateParam(state_spec),
              state_num_bytes=beam.DoFn.StateParam(state_spec2),
              state_num_entries=beam.DoFn.StateParam(state_spec3),
              event_timer=beam.DoFn.TimerParam(event_timer_spec),
              processing_timer=beam.DoFn.TimerParam(processing_timer_spec)):
    # Append stringified element to state until the threshold has been reached
    # The cleanup timer will then clean up and the process will repeat.
    if state_num_bytes.read() <= self.state_size_per_key_bytes:
      state_element = str(element)
      state.add(state_element)
      bytes_added = len(self.str_coder.encode_nested(state_element))
      state_num_bytes.add(bytes_added)
      state_num_entries.add(1)
      timer = processing_timer if self.use_processing_timer else event_timer
      # Set a timer which will clear the state if it grows too large
      timer.set(timestamp.micros // 1000000 + 5)
    # Metrics
    # TODO Unfortunately buggy with timers, needs to be fixed in Beam:
    #self.bytes_gauge.set(state_num_bytes.read())
    #self.elements_gauge.set(state_num_entries.read())
    yield element

  @userstate.on_timer(event_timer_spec)
  def on_event_timer(self,
                     key=beam.DoFn.KeyParam,
                     state=beam.DoFn.StateParam(state_spec),
                     state_num_bytes=beam.DoFn.StateParam(state_spec2),
                     state_num_entries=beam.DoFn.StateParam(state_spec3)):
    return self.timer_callback(state, state_num_bytes, state_num_entries)

  @userstate.on_timer(processing_timer_spec)
  def on_processing_timer(self,
                          state=beam.DoFn.StateParam(state_spec),
                          state_num_bytes=beam.DoFn.StateParam(state_spec2),
                          state_num_entries=beam.DoFn.StateParam(state_spec3)):
    return self.timer_callback(state, state_num_bytes, state_num_entries)

  def timer_callback(self, state, state_num_bytes, state_num_entries):
    count = 0
    for _ in state.read():
      count += 1
    state_count = state_num_entries.read()
    if count != state_count:
      raise Exception("Actual number of entries (%s) did not match expected (%s)" % (count, state_count))
    # Reset state bags
    state.clear()
    state_num_bytes.clear()
    state_num_entries.clear()
    # Reset metrics
    # TODO Unfortunately buggy with timers, needs to be fixed in Beam:
    #self.bytes_gauge.set(0)
    #self.elements_gauge.set(0)
{code}



--
This message was sent by Atlassian Jira
(v8.3.4#803005)