Developing I/O connectors for Python
IMPORTANT: Please use Splittable DoFn to develop your new I/O. For more details, please read
the new I/O connector overview.
To connect to a data store that isn’t supported by Beam’s existing I/O connectors, you must create a custom I/O connector that usually consist of a source and a sink. All Beam sources and sinks are composite transforms; however, the implementation of your custom I/O depends on your use case. Before you start, read the new I/O connector overview for an overview of developing a new I/O connector, the available implementation options, and how to choose the right option for your use case.
This guide covers using the Source and FileBasedSink interfaces for Python. The Java SDK offers the same functionality, but uses a slightly different API. See Developing I/O connectors for Java for information specific to the Java SDK.
Basic code requirements
Beam runners use the classes you provide to read and/or write data using
multiple worker instances in parallel. As such, the code you provide for
Source and FileBasedSink subclasses must meet some basic requirements:
Serializability: Your
SourceorFileBasedSinksubclass must be serializable. The service may create multiple instances of yourSourceorFileBasedSinksubclass to be sent to multiple remote workers to facilitate reading or writing in parallel. The way the source and sink objects are serialized is runner specific.Immutability: Your
SourceorFileBasedSinksubclass must be effectively immutable. You should only use mutable state in yourSourceorFileBasedSinksubclass if you are using lazy evaluation of expensive computations that you need to implement the source.Thread-Safety: Your code must be thread-safe. The Beam SDK for Python provides the
RangeTrackerclass to make this easier.Testability: It is critical to exhaustively unit-test all of your
SourceandFileBasedSinksubclasses. A minor implementation error can lead to data corruption or data loss (such as skipping or duplicating records) that can be hard to detect. You can use test harnesses and utility methods available in the source_test_utils module to develop tests for your source.
In addition, see the PTransform style guide for Beam’s transform style guidance.
Implementing the Source interface
To create a new data source for your pipeline, you’ll need to provide the format-specific logic that tells the service how to read data from your input source, and how to split your data source into multiple parts so that multiple worker instances can read your data in parallel.
If you’re creating a data source that reads unbounded data, also provide the logic for managing your source’s watermark and checkpointing.
Supply the logic for your new source by creating the following classes:
- A subclass of
BoundedSourceif you want to read a finite (batch) data set, or a subclass ofUnboundedSourceif you want to read an infinite (streaming) data set. The class describes the data you want to read, including the data’s location and parameters (such as how much data to read). - For a
BoundedSource, a subclass ofRangeTracker.RangeTrackeris a thread-safe object used to manage a range for a given position type. - For an
UnboundedSource, a subclass ofUnboundedReader, which holds the state involved in reading the stream, and a subclass ofCheckpointMark, which records the position that a reader resumes from. - One or more user-facing wrapper composite transforms (
PTransform) that wrap read operations. PTransform wrappers discusses why you should avoid exposing your sources, and walks through how to create a wrapper.
You can find BoundedSource and RangeTracker in the
apache_beam.io.iobase module,
and the unbounded classes in the
apache_beam.io.unbounded_source module.
Implementing the Source subclass
Create a subclass of either BoundedSource or UnboundedSource, depending on
whether your data is a finite batch or an infinite stream. In either case, the
subclass overrides the methods that a runner uses to split the data and to
create a reader for it.
BoundedSource
BoundedSource represents a finite data set from which the service reads, possibly in parallel. BoundedSource contains a set of methods that the service uses to split the data set for reading by multiple remote workers.
To implement a BoundedSource, your subclass must override the following methods:
estimate_size: Services use this method to estimate the total size of your data, in bytes. This estimate is in terms of external storage size, before performing decompression or other processing.split: Service use this method to split your finite data into bundles of a given size.get_range_tracker: Services use this method to get theRangeTrackerfor a given position range, and use the information to report progress and perform dynamic splitting of sources.read: This method returns an iterator that reads data from the source, with respect to the boundaries defined by the givenRangeTrackerobject.
UnboundedSource
UnboundedSource represents an infinite data stream from which the runner may
read, possibly in parallel. UnboundedSource contains a set of methods that
support streaming reads in parallel; these include checkpointing for failure
recovery and watermarking for estimating data completeness in downstream parts
of your pipeline.
UnboundedSource is experimental, and its API may change in
backwards-incompatible ways.
To implement an UnboundedSource, your subclass must override the following
methods:
split: The SDK uses this method to generate a list ofUnboundedSourceobjects that represent the sub-streams to read in parallel. Each sub-source must be independent and must not share mutable state with its siblings, because the runner may read them concurrently on different workers. Return[self]if the source cannot be split. Splitting happens once, before any checkpoint exists.create_reader: Creates the associatedUnboundedReaderfor thisUnboundedSource. Whencheckpoint_markisNone, the reader starts at the beginning of the stream. Otherwise it resumes strictly after the position that the mark encodes and does not re-deliver records that a previous bundle already read.get_checkpoint_mark_coder: Returns theCoderfor this source’sCheckpointMarkinstances. The SDK calls it while encoding and decoding a reader’s position, so it should be side-effect free and should not perform I/O.
Override default_output_coder to return a coder for your record type. The
default is a pickle coder, and a tighter coder also gives the output
PCollection an element type.
UnboundedSource has no per-record deduplication hook. If your data source can
deliver the same record more than once, drop the duplicates with the
Deduplicate or DeduplicatePerKey
transform after the read.
Implementing the RangeTracker and UnboundedReader subclasses
A runner uses these classes to do the actual reading of your data set and to
track a reader’s progress. A BoundedSource reads through its read method,
which claims positions from a RangeTracker. An UnboundedSource reads through
an UnboundedReader, which also reports a watermark and produces the checkpoint
marks that a runner resumes from.
RangeTracker
A RangeTracker is a thread-safe object used to manage the current range and current position of the reader of a BoundedSource and protect concurrent access to them.
To implement a RangeTracker, you should first familiarize yourself with the following definitions:
Position-based sources - A position-based source can be described by a range of positions of an ordered type, and the records read by the source can be described by positions of that type. For example, for a record within a file, the position can be the starting byte offset of the record. The position type for the record in this case is
long.The main requirement for position-based sources is associativity: Reading records in position range ‘[A, B)’ and records in position range ‘[B, C)’ should give the same records as reading records in position range ‘[A, C)’, where ‘A’ <= ‘B’ <= ‘C’. This property ensures that no matter how many arbitrary sub-ranges a range of positions is split into, the total set of records they describe stays the same.
The other important property is how the source’s range relates to positions of records in the source. In many sources each record can be identified by a unique starting position. In this case:
- All records returned by a source ‘[A, B)’ must have starting positions in this range.
- All but the last record should end within this range. The last record may or may not extend past the end of the range.
- Records must not overlap.
Such sources should define “read ‘[A, B)’” as “read from the first record starting at or after ‘A’, up to but not including the first record starting at or after ‘B’”.
Some examples of such sources include reading lines or CSV from a text file, reading keys and values from a database, etc.
The concept of split points allows to extend the definitions for dealing with sources where some records cannot be identified by a unique starting position.
Split points - A split point describes a record that is the first one returned when reading the range from and including position A up to infinity (i.e. [A, infinity)).
Some sources may have records that are not directly addressable. For example, imagine a file format consisting of a sequence of compressed blocks. Each block can be assigned an offset, but records within the block cannot be directly addressed without decompressing the block. Let us refer to this hypothetical format as CBF (Compressed Blocks Format).
Many such formats can still satisfy the associativity property. For example, in CBF, reading [A, B) can mean “read all the records in all blocks whose starting offset is in [A, B)”.
To support such complex formats, Beam introduces the notion of split points. A record is a split point if there exists a position A such that the record is the first one to be returned when reading the range [A, infinity). In CBF, the only split points would be the first records in each block.
Split points allow us to define the meaning of a record’s position and a source’s range in the following cases:
- For a record that is at a split point, its position is defined to be the largest A such that reading a source with the range [A, infinity) returns this record.
- Positions of other records are only required to be non-decreasing.
- Reading the source [A, B) must return records starting from the first split point at or after A, up to but not including the first split point at or after B. In particular, this means that the first record returned by a source MUST always be a split point.
- Positions of split points must be unique.
As a result, for any decomposition of the full range of the source into position ranges, the total set of records will be the full set of records in the source, and each record will be read exactly once.
Consumed positions - Consumed positions refer to records that have been read.
As the source is being read, and records read from it are being passed to the downstream transforms in the pipeline, we say that positions in the source are being consumed. When a reader has read a record (or promised to a caller that a record will be returned), positions up to and including the record’s start position are considered consumed.
Dynamic splitting can happen only at unconsumed positions. If the reader just returned a record at offset 42 in a file, dynamic splitting can happen only at offset 43 or beyond. Otherwise, that record could be read twice (by the current reader and the reader of the new task).
RangeTracker methods
To implement a RangeTracker, your subclass must override the following methods:
start_position: Returns the starting position of the current range, inclusive.stop_position: Returns the ending position of the current range, exclusive.try_claim: This method is used to determine if a record at a split point is within the range. This method should modify the internal state of theRangeTrackerby updating the last-consumed position to the given startingpositionof the record being read by the source. The method returns true if the given position falls within the current range.set_current_position: This method updates the last-consumed position to the given starting position of a record being read by a source. You can invoke this method for records that do not start at split points, and this should modify the internal state of theRangeTracker. If the record starts at a split point, you must invoketry_claiminstead of this method.position_at_fraction: Given a fraction within the range [0.0, 1.0), this method will return the position at the given fraction compared to the position range [self.start_position,self.stop_position).try_split: This method attempts to split the current range into two parts around a suggested position. It is allowed to split at a different position, but in most cases it will split at the suggested position.
This method splits the current range [self.start_position, self.stop_position) into a “primary” part [self.start_position, split_position), and a “residual” part [split_position, self.stop_position), assuming that split_position has not been consumed yet.
If split_position has already been consumed, the method returns None. Otherwise, it updates the current range to be the primary and returns a tuple (split_position, split_fraction). split_fraction should be the fraction of size of range [self.start_position, split_position) compared to the original (before split) range [self.start_position, self.stop_position).
fraction_consumed: Returns the approximate fraction of consumed positions in the source.
Note: Methods of class iobase.RangeTracker may be invoked by multiple threads, hence this class must be made thread-safe, for example, by using a single lock object.
UnboundedReader
An UnboundedReader holds the state involved in reading one UnboundedSource,
such as connections and buffers. start is called exactly once, then advance
is called repeatedly; whenever either returns True, the current record is
available through get_current and get_current_timestamp.
To implement an UnboundedReader, your subclass must override the following
methods:
start: Initializes the reader, positions it at the first record, and returns whether one is available. This is a good place for expensive initialization.advance: Advances to the next record and returns whether one is available. AFalsereturn means that no data is available right now, which differs from the end of the stream: a reader signals a permanent end by reporting a watermark ofMAX_TIMESTAMP. This method should not block; returnFalsewhen no data is currently available instead of waiting for more.get_current: Returns the record at the current position, last read bystartoradvance.get_current_timestamp: Returns the event-time timestamp of the current record, which becomes the timestamp of the output element.get_watermark: Returns a watermark, the approximate lower bound on the timestamps of the records that this reader produces in the future. The runner uses the watermark as an estimate of data completeness in windowing and triggers. The watermark is treated as monotonic.get_checkpoint_mark: Returns aCheckpointMarkthat records how far the reader has read. It is called only at a bundle boundary, and the mark is passed back tocreate_readerto resume.close: Releases the reader’s resources. The default is a no-op.
Checkpoint marks
A CheckpointMark is a durable, serializable position in the stream. The runner
persists it with the coder from get_checkpoint_mark_coder and hands it to
create_reader when a bundle resumes or when a worker recovers from a failure.
Override finalize_checkpoint to acknowledge or commit the consumed records
upstream, for example to ack the messages on a queue. It is called after the
runner has durably committed the work covered by that mark. Finalization is best
effort: a mark may never be finalized, and a retried bundle may re-cut a mark
over an overlapping span, so acknowledge by absolute position and keep the method
idempotent.
Convenience Source base classes
The Beam SDK for Python contains some convenient abstract base classes to help you easily create new sources.
FileBasedSource
FileBasedSource is a framework for developing sources for new file types. You can derive your BoundedSource class from the FileBasedSource class.
To create a source for a new file type, you need to create a sub-class of FileBasedSource. Sub-classes of FileBasedSource must implement the method FileBasedSource.read_records().
See AvroSource for an example implementation of FileBasedSource.
Reading from a new Source
The following example, CountingSource, demonstrates an implementation of BoundedSource and uses the SDK-provided RangeTracker called OffsetRangeTracker.
class CountingSource(iobase.BoundedSource):
def __init__(self, count):
self.records_read = Metrics.counter(self.__class__, 'recordsRead')
self._count = count
def estimate_size(self):
return self._count
def get_range_tracker(self, start_position, stop_position):
if start_position is None:
start_position = 0
if stop_position is None:
stop_position = self._count
return OffsetRangeTracker(start_position, stop_position)
def read(self, range_tracker):
for i in range(range_tracker.start_position(),
range_tracker.stop_position()):
if not range_tracker.try_claim(i):
return
self.records_read.inc()
yield i
def split(self, desired_bundle_size, start_position=None, stop_position=None):
if start_position is None:
start_position = 0
if stop_position is None:
stop_position = self._count
bundle_start = start_position
while bundle_start < stop_position:
bundle_stop = min(stop_position, bundle_start + desired_bundle_size)
yield iobase.SourceBundle(
weight=(bundle_stop - bundle_start),
source=self,
start_position=bundle_start,
stop_position=bundle_stop)
bundle_start = bundle_stopTo read data from the source in your pipeline, use the Read transform:
Note: When you create a source that end-users are going to use, we
recommended that you do not expose the code for the source itself as
demonstrated in the example above. Use a wrapping PTransform instead.
PTransform wrappers discusses why you should avoid
exposing your sources, and walks through how to create a wrapper.
Reading from an UnboundedSource
The following example, QueueSource, reads from a partitioned message queue.
my_queue stands in for the client library of the system you read from.
import apache_beam as beam
from apache_beam.io.unbounded_source import CheckpointMark
from apache_beam.io.unbounded_source import UnboundedReader
from apache_beam.io.unbounded_source import UnboundedSource
from apache_beam.utils.timestamp import Timestamp
class QueueCheckpointMark(CheckpointMark):
def __init__(self, offset):
self.offset = offset
def finalize_checkpoint(self):
# Acknowledging an absolute offset is safe to repeat.
my_queue.ack_through(self.offset)
class QueueReader(UnboundedReader):
def __init__(self, partition, offset):
self._partition = partition
self._offset = offset
self._message = None
def start(self):
return self.advance()
def advance(self):
message = my_queue.poll(self._partition, self._offset)
if message is None:
return False
self._offset = message.offset
self._message = message
return True
def get_current(self):
return self._message.body
def get_current_timestamp(self):
return Timestamp(micros=self._message.event_time_micros)
def get_watermark(self):
return Timestamp(micros=my_queue.oldest_pending_micros(self._partition))
def get_checkpoint_mark(self):
return QueueCheckpointMark(self._offset)
def close(self):
my_queue.disconnect(self._partition)
class QueueSource(UnboundedSource):
def __init__(self, topic, partition=None):
self._topic = topic
self._partition = partition
def split(self, desired_num_splits, options=None):
if self._partition is not None:
return [self]
return [
QueueSource(self._topic, partition)
for partition in my_queue.partitions(self._topic)
]
def create_reader(self, options, checkpoint_mark):
offset = None if checkpoint_mark is None else checkpoint_mark.offset
return QueueReader(self._partition, offset)
def get_checkpoint_mark_coder(self):
return beam.coders.PickleCoder()
def default_output_coder(self):
return beam.coders.BytesCoder()To read data from the source in your pipeline, use the Read transform, which
dispatches an UnboundedSource automatically:
A bundle ends when the reader runs out of data, and also once the reader has
emitted max_records_per_bundle records or spent max_read_time_seconds in the
bundle, so the runner commits the checkpoint and runs finalization before the
read resumes. An idle reader is polled again after poll_interval seconds.
Apply ReadFromUnboundedSource directly to change these defaults:
Note: As with a bounded source, we recommend that you do not expose the code
for the source itself to end-users. Use a wrapping PTransform instead.
PTransform wrappers discusses why you should avoid
exposing your sources, and walks through how to create a wrapper.
Using the FileBasedSink abstraction
If your data source uses files, you can implement the FileBasedSink
abstraction to create a file-based sink. For other sinks, use ParDo,
GroupByKey, and other transforms offered by the Beam SDK for Python. See the
developing I/O connectors overview
for more details.
When using the FileBasedSink interface, you must provide the format-specific
logic that tells the runner how to write bounded data from your pipeline’s
PCollections to an output sink. The runner writes bundles of data in parallel
using multiple workers.
Supply the logic for your file-based sink by implementing the following classes:
A subclass of the abstract base class
FileBasedSink.FileBasedSinkdescribes a location or resource that your pipeline can write to in parallel. To avoid exposing your sink to end-users, use the_prefix when creating yourFileBasedSinksubclass.A user-facing wrapper
PTransformthat, as part of the logic, callsWriteand passes yourFileBasedSinkas a parameter. A user should not need to callWritedirectly.
The FileBasedSink abstract base class implements code that is common to Beam
sinks that interact with files, including:
- Setting file headers and footers
- Sequential record writing
- Setting the output MIME type
FileBasedSink and its subclasses support writing files to any Beam-supported
FileSystem implementations. See the following Beam-provided FileBasedSink
implementation for an example:
PTransform wrappers
When you create a source or sink that end-users will use, avoid exposing your
source or sink code. To avoid exposing your sources and sinks to end-users, your
new classes should use the _ prefix. Then, implement a user-facing
wrapper PTransform.`By exposing your source or sink as a transform, your
implementation is hidden and can be arbitrarily complex or simple. The greatest
benefit of not exposing implementation details is that later on, you can add
additional functionality without breaking the existing implementation for users.
For example, if your users’ pipelines read from your source using
beam.io.Read and you want to insert a reshard into the pipeline, all
users would need to add the reshard themselves (using the GroupByKey
transform). To solve this, we recommended that you expose the source as a
composite PTransform that performs both the read operation and the reshard.
See Beam’s PTransform style guide
for additional information about wrapping with a PTransform.
The following examples change the source and sink from the above sections so
that they are not exposed to end-users. For the source, rename CountingSource
to _CountingSource. Then, create the wrapper PTransform, called
ReadFromCountingSource:
Finally, read from the source:
For the sink, rename SimpleKVSink to _SimpleKVSink. Then, create the wrapper PTransform, called WriteToKVSink:
class WriteToKVSink(PTransform):
def __init__(self, simplekv, url, final_table_name):
self._simplekv = simplekv
super().__init__()
self._url = url
self._final_table_name = final_table_name
def expand(self, pcoll):
return pcoll | iobase.Write(
_SimpleKVSink(self._simplekv, self._url, self._final_table_name))Finally, write to the sink:
Last updated on 2026/08/23
Have you found everything you were looking for?
Was it all useful and clear? Is there anything that you would like to change? Let us know!

