Apache Beam - Start sliding window from first element - google-cloud-dataflow

Question

I am trying to develop dataflow pipeline using sliding window on bounded and streaming dataset using Apache Beam Python sdk. The pipeline is as follows:
Reading data
Assigning timestamps
Windowing using SlidingWindows() of 3 size and 1 period
Grouping the elements in the windows
Printing the output
The sample Data:
data = [{'serverID': 'server_1', 'CPU_Utilization': 0, 'timestamp': 1},
{'serverID': 'server_1', 'CPU_Utilization': 1, 'timestamp': 2},
{'serverID': 'server_1', 'CPU_Utilization': 2, 'timestamp': 3},
{'serverID': 'server_1', 'CPU_Utilization': 3, 'timestamp': 4}]
The beam pipeline:
import apache_beam as beam
from apache_beam.options.pipeline_options import PipelineOptions
from apache_beam.options.pipeline_options import SetupOptions
pipeline_options = PipelineOptions()
pipeline_options.view_as(SetupOptions).save_main_session = True
p = beam.Pipeline(options=pipeline_options)
with beam.Pipeline(options=pipeline_options) as p:
events = (p | 'Create Events' >> beam.Create(data) \
| 'Add Timestamps' >> beam.Map(lambda x: beam.window.TimestampedValue(x,
x['timestamp'])) \
| 'PairWithOne' >> beam.Map(lambda x: (None, x)) \
|'Sliding Window'>>beam.WindowInto(beam.window.SlidingWindows(3,1))
| 'Group by key' >> beam.GroupByKey()
|beam.Map(print))
The output I got:
(None, [{'serverID': 'server_1', 'CPU_Utilization': 0, 'timestamp': 1}, {'serverID':
'server_1', 'CPU_Utilization': 1, 'timestamp': 2}, {'serverID': 'server_1',
'CPU_Utilization': 2, 'timestamp': 3}])
(None, [{'serverID': 'server_1', 'CPU_Utilization': 0, 'timestamp': 1}, {'serverID':
'server_1', 'CPU_Utilization': 1, 'timestamp': 2}])
(None, [{'serverID': 'server_1', 'CPU_Utilization': 0, 'timestamp': 1}])
(None, [{'serverID': 'server_1', 'CPU_Utilization': 1, 'timestamp': 2}, {'serverID':
'server_1', 'CPU_Utilization': 2, 'timestamp': 3}, {'serverID': 'server_1',
'CPU_Utilization': 3, 'timestamp': 4}])
(None, [{'serverID': 'server_1', 'CPU_Utilization': 2, 'timestamp': 3}, {'serverID':
'server_1', 'CPU_Utilization': 3, 'timestamp': 4}])
(None, [{'serverID': 'server_1', 'CPU_Utilization': 3, 'timestamp': 4}])
The expected output (i.e to discard data when starting window timestamp is less than the timestamp of 1st row or element in data):
(None, [{'serverID': 'server_1', 'CPU_Utilization': 0, 'timestamp': 1}, {'serverID':
'server_1', 'CPU_Utilization': 1, 'timestamp': 2}, {'serverID': 'server_1',
'CPU_Utilization': 2, 'timestamp': 3}])
(None, [{'serverID': 'server_1', 'CPU_Utilization': 1, 'timestamp': 2}, {'serverID':
'server_1', 'CPU_Utilization': 2, 'timestamp': 3}, {'serverID': 'server_1',
'CPU_Utilization': 3, 'timestamp': 4}])
(None, [{'serverID': 'server_1', 'CPU_Utilization': 2, 'timestamp': 3}, {'serverID':
'server_1', 'CPU_Utilization': 3, 'timestamp': 4}])
(None, [{'serverID': 'server_1', 'CPU_Utilization': 3, 'timestamp': 4}])
I have also tried AfterCount(n) trigger but this trigger does not consider the data when number of data points is less than n.
Any help on this would be really appreciated.

Answer 1

It's probably working as intended. Could you try validating the output by writing it into a file instead of using print? It could be that there are multiple panes fired and printed but the accumulation_mode is discarding.
Also, you can try using Beam Notebooks when debugging streaming pipelines. Use something like show(pcoll, include_window_info=True) to visualize the event time, windows and pane info of a PCollection instead of printing them.