OAK-11556 : removed usage of Guava's Iterables.getFirst with oak-commons API

[rishabhdaim]

No description provided.

[github-actions]

Commit-Check ✔️

[nfsantos]

This might be introducing a performance regression. Iterables.getFirst does not allocate any object, while StreamUtils.toStream() creates an object internally. And the code path of Iterables.getFirst is much shorter, so potentially faster than the Guava equivalent. This method, createProperty(), is likely performance critical, so I'm worried this could significantly degrade performance of some parts of Oak.
In general, I'm concerned about all these patches removing Guava. We have tests for correctness (and even with those tests, these patches have broken the build several times). But we do not have any performance regression tests, so these patches may be degrading the performance without us realising.

[rishabhdaim]

Thanks, @nfsantos for the review.

I measured the performance of 3 implementations (using JMH) for:

1. Guava's
2. Stream
3. Apache commons

Please find the results below:

Code is:

`@Benchmark
public void guava(Blackhole bh) {
   bh.consume(Iterables.getFirst(itr, bh));
}

@Benchmark
public void apache(Blackhole bh) {
    bh.consume(org.apache.commons.collections4.IterableUtils.first(itr));
}

@Benchmark
public void stream(Blackhole bh) {
    bh.consume(StreamSupport.stream(itr.spliterator(), false).findFirst().orElse(0));
}`

Results

Benchmark (size) Mode Cnt Score Error Units
IterableUtils.apache 1 avgt 4 0.001 ± 0.001 us/op
IterableUtils.apache 100 avgt 4 0.001 ± 0.001 us/op
IterableUtils.apache 10000 avgt 4 0.001 ± 0.001 us/op
IterableUtils.guava 1 avgt 4 0.001 ± 0.001 us/op
IterableUtils.guava 100 avgt 4 0.001 ± 0.001 us/op
IterableUtils.guava 10000 avgt 4 0.001 ± 0.001 us/op
IterableUtils.stream 1 avgt 4 0.008 ± 0.003 us/op
IterableUtils.stream 100 avgt 4 0.007 ± 0.002 us/op
IterableUtils.stream 10000 avgt 4 0.007 ± 0.002 us/op

Although the performance with Stream is a bit slow for getting first, that is not going to have a very drastic impact on overall OAK performance.

But, we do have a better alternative in commons-collections4, I think we can go with that as well.

Will update the PR to use Apache commons-collections4.

[nfsantos]

@rishabhdaim, thanks for micro benchmark results. Could you reformat them so they are easier to read?

If I understood correctly, the longest iteration repeated the operation 10000 times, right? This is not much, it would be better to run for at least a few seconds to give time for the JIT to quick in.

But sounds good to use commons-collections4 if its performance is equivalent to Guava's.

[rishabhdaim]

Added PR #2150 to create Iterables.getFirst() replacement in oak-commons. Once approved/merged, will use that.

[rishabhdaim]

@nfsantos Even after increasing the param size to 1000000, the results remain the same.

[rishabhdaim]

Benchmark (size) Mode Cnt Score Error Units
IterableUtils.apache 1 avgt 4 0.001 ± 0.001 us/op
IterableUtils.apache 100 avgt 4 0.001 ± 0.001 us/op
IterableUtils.apache 10000 avgt 4 0.001 ± 0.001 us/op
IterableUtils.apache 1000000 avgt 4 0.001 ± 0.001 us/op
IterableUtils.guava 1 avgt 4 0.001 ± 0.001 us/op
IterableUtils.guava 100 avgt 4 0.001 ± 0.001 us/op
IterableUtils.guava 10000 avgt 4 0.001 ± 0.001 us/op
IterableUtils.guava 1000000 avgt 4 0.001 ± 0.001 us/op
IterableUtils.stream 1 avgt 4 0.007 ± 0.002 us/op
IterableUtils.stream 100 avgt 4 0.007 ± 0.001 us/op
IterableUtils.stream 10000 avgt 4 0.007 ± 0.001 us/op
IterableUtils.stream 1000000 avgt 4 0.007 ± 0.001 us/op

[rishabhdaim]

review comments addressed in 11a7aca and now using new oak-commons util.

[sonarqubecloud]

Quality Gate passed

Issues
5 New issues
0 Accepted issues

Measures
0 Security Hotspots
50.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud