Change UnboundedSourceAsSdfWrapperFn to share the cache across instances.

[scwhittle]

Add a utility class to enable sharing across all deserialized instances of a DoFn and use it in UnboundedSourceAsSdfWrapperFn to cache Readers across dofn instances.

fixes #32968

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

• Mention the appropriate issue in your description (for example: addresses #123), if applicable. This will automatically add a link to the pull request in the issue. If you would like the issue to automatically close on merging the pull request, comment fixes #<ISSUE NUMBER> instead.
• Update CHANGES.md with noteworthy changes.
• If this contribution is large, please file an Apache Individual Contributor License Agreement.

See the Contributor Guide for more tips on how to make review process smoother.

To check the build health, please visit https://github.com/apache/beam/blob/master/.test-infra/BUILD_STATUS.md

GitHub Actions Tests Status (on master branch)

See CI.md for more information about GitHub Actions CI or the workflows README to see a list of phrases to trigger workflows.

[scwhittle]

R: @kennknowles

I don't really like the utility class name, let me know if you have some better idea. It seems like this could be a common use within DoFns. For example, @sjvanrossum made me think of this concern in his work on Kafka source in #32986

[github-actions]

Stopping reviewer notifications for this pull request: review requested by someone other than the bot, ceding control. If you'd like to restart, comment assign set of reviewers

[scwhittle]

assign set of reviewers

[scwhittle]

@sjvanrossum I tried to abstract your caching in the KafkaSDF here since there would be similar concerns about a static cache for this dofn.

[github-actions]

Assigning reviewers. If you would like to opt out of this review, comment assign to next reviewer:

R: @m-trieu for label java.

Available commands:

• stop reviewer notifications - opt out of the automated review tooling
• remind me after tests pass - tag the comment author after tests pass
• waiting on author - shift the attention set back to the author (any comment or push by the author will return the attention set to the reviewers)

The PR bot will only process comments in the main thread (not review comments).

[sjvanrossum]

@sjvanrossum I tried to abstract your caching in the KafkaSDF here since there would be similar concerns about a static cache for this dofn.

That looks good to me, thanks for distilling it into a utility class! As to naming, this is essentially a Supplier with additional guarantees about the return value of #get().

Drawing inspiration from Guava's Suppliers #memoize(Supplier), #memoizeWithExpiration(Supplier, long, TimeUnit) and #ofInstance​(T) backed by MemoizingSupplier, ExpiringMemoizingSupplier and SupplierOfInstance respectively, I think MemoizingPerInstantiationSerializableSupplier conveys the intended application of this class.

[scwhittle]

Run Java PreCommit

[scwhittle]

Run Java_IOs_Direct PreCommit

[scwhittle]

Run Java_Pulsar_IO_Direct PreCommit

[kennknowles]

The only pitfall I can think of is if it was used out of context in a DoFn that was used twice without being instantiated separately at each site. But this deep under the hood I think that is a pretty far fetched concern.

[kennknowles]

Fancy! It took me a minute to understand how this works.

[scwhittle]

@kennknowles Could you take another look? I realized that we have to worry now about concurrent cache access (before the DoFn being single-use-at-a-time prevented that) and the get/invalidate as separate methods could be racy. So I looked at dataflow runners/google-cloud-dataflow-java/worker/src/main/java/org/apache/beam/runners/dataflow/worker/ReaderCache.java and changed to use the concurrentmap directly as it does. While there I noticed that there were previously problems due to close blocking so I made this similarly async. http://issues.apache.org/jira/browse/BEAM-11401

[kennknowles]

Interesting. The race is still there, right? The async close could be racing with some other usage of it, no? But it won't block the current thread is the main thing.

[kennknowles]

nit: this doesn't actually change anything, as it will just throw an NPE same as if you didn't use it. So prefer to use Beam SDK's own Preconditions which have checkArgumentNotNull (throws IllegalArgumentException) and checkStateNotNull (throws IllegalStateException)

Out of curiosity, I looked at the docs on RemovalNotification I believe if the value is null here that means it was already garbage collected. Since you aren't using weakValues this will never happen. But if you wanted this code to be agnostic to the cache configuration a bit, you could just check if it is null and return early.

[scwhittle]

Interesting. The race is still there, right? The async close could be racing with some other usage of it, no? But it won't block the current thread is the main thing.

The race was more with the get from the cache and separate invalidate. Since there are simulataneous dofns both could get the same value where we want more of acquire/checkout semantics so that there is just one user. That was addressed using the atomic asMap().remove() instead of the separate get/invalidate methods.

It's true long closes could still be an issue but at least they aren't blocking unrelated processing threads.

[scwhittle]

Run Java_IOs_Direct PreCommit

[scwhittle]

Run Java PreCommit

[scwhittle]

Run Java_Pulsar_IO_Direct PreCommit

[scwhittle]

Run Java_IOs_Direct PreCommit

[scwhittle]

Run Java_Amazon-Web-Services2_IO_Direct PreCommit

[scwhittle]

failures are unrelated