[BEAM-8088] Track PCollection boundedness in python sdk

[chadrik]

As far as I can tell Python does not care about boundedness of PCollections even in streaming mode, but external transforms do. In my ongoing effort to get PubsubIO external transforms working I discovered that I could not generate an unbounded write using the expansion service from python: it always came back as a bounded write. 

My pipeline looks like this:

(
pipe|external.pubsub.ReadFromPubSub(subscription=subscription, with_attributes=True)
|external.pubsub.WriteToPubSub(topic=OUTPUT_TOPIC, with_attributes=True)
)

The PCollections returned from the external Read are Unbounded, as expected, but python is responsible for creating the intermediate PCollection, which is always Bounded, and thus external Write generated by Java is always Bounded.

If I'm on the right track here I'll make some tests. I've manually tested it enough to confirm that it gets my external xform tests working.

---

Thank you for your contribution! Follow this checklist to help us incorporate your contribution quickly and easily:

• Choose reviewer(s) and mention them in a comment (R: @username).
• Format the pull request title like [BEAM-XXX] Fixes bug in ApproximateQuantiles, where you replace BEAM-XXX with the appropriate JIRA issue, if applicable. This will automatically link the pull request to the issue.
• If this contribution is large, please file an Apache Individual Contributor License Agreement.

Post-Commit Tests Status (on master branch)

Lang	SDK	Apex	Dataflow	Flink	Gearpump	Samza	Spark
Go		---	---		---	---
Java
Python		---			---	---
XLang	---	---	---		---	---	---

Pre-Commit Tests Status (on master branch)

---	Java	Python	Go	Website
Non-portable
Portable	---		---	---

See .test-infra/jenkins/README for trigger phrase, status and link of all Jenkins jobs.

[chadrik]

R: @robertwb
R: @mxm

[mxm]

Just to clarify, this is to have an unbounded PCollection during the expansion of the WriteToPubSub write?

[chadrik]

Just to clarify, this is to have an unbounded PCollection during the expansion of the WriteToPubSub write?

Correct.

PubSubIO.Write.expand looks like this:

@OverridepublicPDoneexpand(PCollection<T> input) {
if (getTopicProvider() == null) {
thrownewIllegalStateException("need to set the topic of a PubsubIO.Write transform");
}
switch (input.isBounded()) {
caseBOUNDED:
input.apply(
ParDo.of(
newPubsubBoundedWriter(
MoreObjects.firstNonNull(getMaxBatchSize(), MAX_PUBLISH_BATCH_SIZE),
MoreObjects.firstNonNull(
getMaxBatchBytesSize(), MAX_PUBLISH_BATCH_BYTE_SIZE_DEFAULT))));
returnPDone.in(input.getPipeline());
caseUNBOUNDED:
returninput
.apply(MapElements.via(getFormatFn()))
.apply(
newPubsubUnboundedSink(
Optional.ofNullable(getPubsubClientFactory()).orElse(FACTORY),
NestedValueProvider.of(getTopicProvider(), newTopicPathTranslator()),
getTimestampAttribute(),
getIdAttribute(),
100/* numShards */,
MoreObjects.firstNonNull(
getMaxBatchSize(), PubsubUnboundedSink.DEFAULT_PUBLISH_BATCH_SIZE),
MoreObjects.firstNonNull(
getMaxBatchBytesSize(),
PubsubUnboundedSink.DEFAULT_PUBLISH_BATCH_BYTES)));
}
thrownewRuntimeException(); // cases are exhaustive.
}

The ReadFromPubsub was expanded by Java, but Python was not properly handling the unbounded PCollections that it returned, so when WriteToPubsub was subsequently expanded, it always produced a PubsubBoundedWriter.

[mxm]

That seems fair. The boundedness is part of the Proto and thus should be available also from within Python.

Should we add a test for this in pipeline_test.py?

[chadrik]

Should we add a test for this in pipeline_test.py?

Done!

[mxm]

Thanks!