apache_beam.testing.synthetic_pipeline module

A set of utilities to write pipelines for performance tests.

This module offers a way to create pipelines using synthetic sources and steps. Exact shape of the pipeline and the behaviour of sources and steps can be controlled through arguments. Please see function ‘parse_args()’ for more details about the arguments.

Shape of the pipeline is primariy controlled through two arguments. Argument ‘steps’ can be used to define a list of steps as a JSON string. Argument ‘barrier’ describes how these steps are separated from each other. Argument ‘barrier’ can be use to build a pipeline as a a series of steps or a tree of steps with a fanin or a fanout of size 2.

Other arguments describe what gets generated by synthetic sources that produce data for the pipeline.

apache_beam.testing.synthetic_pipeline.Generator

alias of apache_beam.testing.synthetic_pipeline._Random

apache_beam.testing.synthetic_pipeline.parse_byte_size(s)

apache_beam.testing.synthetic_pipeline.div_round_up(a, b)

Return ceil(a/b).

apache_beam.testing.synthetic_pipeline.rotate_key(element)

Returns a new key-value pair of the same size but with a different key.

apache_beam.testing.synthetic_pipeline.initial_splitting_zipf(start_position, stop_position, desired_num_bundles, distribution_parameter, num_total_records=None)

Split the given range (defined by start_position, stop_position) into desired_num_bundles using zipf with the given distribution_parameter.

class apache_beam.testing.synthetic_pipeline.SyntheticStep(per_element_delay_sec=0, per_bundle_delay_sec=0, output_records_per_input_record=1, output_filter_ratio=0)

Bases: apache_beam.transforms.core.DoFn

A DoFn of which behavior can be controlled through prespecified parameters.

start_bundle()

finish_bundle()

process(element)

class apache_beam.testing.synthetic_pipeline.NonLiquidShardingOffsetRangeTracker(offset_range)

Bases: apache_beam.io.restriction_trackers.OffsetRestrictionTracker

An OffsetRangeTracker that doesn’t allow splitting.

try_split(split_offset)

checkpoint()

class apache_beam.testing.synthetic_pipeline.SyntheticSDFStepRestrictionProvider(num_records, initial_splitting_num_bundles, initial_splitting_uneven_chunks, disable_liquid_sharding, size_estimate_override)

Bases: apache_beam.transforms.core.RestrictionProvider

A RestrictionProvider for SyntheticSDFStep.

An initial_restriction and split that operate on num_records and ignores source description (element). Splits into initial_splitting_num_bundles. Returns size_estimate_override as restriction size, if set. Otherwise uses element size.

If initial_splitting_uneven_chunks, produces uneven chunks.

initial_restriction(element)

create_tracker(restriction)

split(element, restriction)

restriction_size(element, restriction)

apache_beam.testing.synthetic_pipeline.get_synthetic_sdf_step(per_element_delay_sec=0, per_bundle_delay_sec=0, output_records_per_input_record=1, output_filter_ratio=0, initial_splitting_num_bundles=8, initial_splitting_uneven_chunks=False, disable_liquid_sharding=False, size_estimate_override=None)

A function which returns a SyntheticSDFStep with given parameters.

class apache_beam.testing.synthetic_pipeline.SyntheticSource(input_spec)

Bases: apache_beam.io.iobase.BoundedSource

A custom source of a specified size.

Initiates a synthetic source.

Parameters:input_spec – Input specification of the source. See corresponding option in function ‘parse_args()’ below for more details. Raises:ValueError – if input parameters are invalid.

element_size

estimate_size()

split(desired_bundle_size, start_position=0, stop_position=None)

get_range_tracker(start_position, stop_position)

read(range_tracker)

default_output_coder()

class apache_beam.testing.synthetic_pipeline.SyntheticSDFSourceRestrictionProvider

Bases: apache_beam.transforms.core.RestrictionProvider

A RestrictionProvider for SyntheticSDFAsSource.

In initial_restriction(element) and split(element), element means source description. A typical element is like:

{

‘key_size’: 1, ‘value_size’: 1, ‘initial_splitting_num_bundles’: 8, ‘initial_splitting_desired_bundle_size’: 2, ‘sleep_per_input_record_sec’: 0, ‘initial_splitting’ : ‘const’

}

initial_restriction(element)

create_tracker(restriction)

split(element, restriction)

restriction_size(element, restriction)

class apache_beam.testing.synthetic_pipeline.SyntheticSDFAsSource(*unused_args, **unused_kwargs)

Bases: apache_beam.transforms.core.DoFn

A SDF that generates records like a source.

This SDF accepts a PCollection of record-based source description. A typical description is like:

{

‘key_size’: 1, ‘value_size’: 1, ‘initial_splitting_num_bundles’: 8, ‘initial_splitting_desired_bundle_size’: 2, ‘sleep_per_input_record_sec’: 0, ‘initial_splitting’ : ‘const’

}

A simple pipeline taking this SDF as a source is like:

p | beam.Create([description1, description2,…]) | beam.ParDo(SyntheticSDFAsSource())

Note

The SDF.process() will have different param content between defining a DoFn and runtime. When defining an SDF.process, the restriction_tracker should be a RestrictionProvider. During runtime, the DoFnRunner.process_with_sized_restriction() will feed a ‘RestrictionTracker’ based on a restriction to SDF.process().

process(element, restriction_tracker=RestrictionParam(SyntheticSDFSourceRestrictionProvider))

class apache_beam.testing.synthetic_pipeline.ShuffleBarrier(label=None)

Bases: apache_beam.transforms.ptransform.PTransform

expand(pc)

class apache_beam.testing.synthetic_pipeline.SideInputBarrier(label=None)

Bases: apache_beam.transforms.ptransform.PTransform

expand(pc)

apache_beam.testing.synthetic_pipeline.merge_using_gbk(name, pc1, pc2)

Merges two given PCollections using a CoGroupByKey.

apache_beam.testing.synthetic_pipeline.merge_using_side_input(name, pc1, pc2)

Merges two given PCollections using side inputs.

apache_beam.testing.synthetic_pipeline.expand_using_gbk(name, pc)

Expands a given PCollection into two copies using GroupByKey.

apache_beam.testing.synthetic_pipeline.expand_using_second_output(name, pc)

Expands a given PCollection into two copies using side outputs.

apache_beam.testing.synthetic_pipeline.parse_args(args)

Parses a given set of arguments.

Parameters:args – set of arguments to be passed. Returns:a tuple where first item gives the set of arguments defined and parsed within this method and second item gives the set of unknown arguments.

apache_beam.testing.synthetic_pipeline.run(argv=None, save_main_session=True)

Runs the workflow.

class apache_beam.testing.synthetic_pipeline.StatefulLoadGenerator(input_options, num_keys=100)

Bases: apache_beam.transforms.ptransform.PTransform

A PTransform for generating random data using Timers API.

class GenerateKeys(num_keys, key_size)

Bases: apache_beam.transforms.core.DoFn

process(impulse)

class GenerateLoad(num_records_per_key, value_size, bundle_size=1000)

Bases: apache_beam.transforms.core.DoFn

state_spec = CombiningValueStateSpec(bundles_remaining)

timer_spec = TimerSpec(ts-timer)

process(_element, records_remaining=StateParam(bundles_remaining), timer=TimerParam(ts-timer))

process_timer(key=KeyParam, records_remaining=StateParam(bundles_remaining), timer=TimerParam(ts-timer))

expand(pbegin)