Fixes memory leak in BytesRefLongBlockHash

[ncordon]

Closes #137021.

Repro:

./gradlew ":x-pack:plugin:esql:compute:test" --tests "org.elasticsearch.compute.aggregation.blockhash.BlockHashRandomizedTests.testWithCranky {forcePackedHash=false groups=2 maxValuesPerPosition=1 dups=0 allowedTypes=[Basic[elementType=LONG], Basic[elementType=BYTES_REF]]}" -Dtests.seed=F336120F4E4CDBE -Dtests.locale=en-DG -Dtests.timezone=America/Sitka -Druntime.java=25

The test BlockHashRandomizedTests#testWithCranky was failing in the after phase when we were checking all of the allocated arrays had been released. That test only failed once every 20 times (the frequency CrankyCircuitBreakerService kicks in).

    public void testWithCranky() {
        CircuitBreakerService service = new CrankyCircuitBreakerService();
        CircuitBreaker breaker = service.getBreaker(CircuitBreaker.REQUEST);
        BigArrays bigArrays = new MockBigArrays(PageCacheRecycler.NON_RECYCLING_INSTANCE, service);
        try {
            test(new MockBlockFactory(breaker, bigArrays));
            logger.info("cranky let us finish!");
        } catch (CircuitBreakingException e) {
            logger.info("cranky", e);
            assertThat(e.getMessage(), equalTo(CrankyCircuitBreakerService.ERROR_MESSAGE));
        }
    }

[elasticsearchmachine]

Hi @ncordon, I've created a changelog YAML for you.

[ncordon]

A circuit breaker could kick in here, in the ordinals.build() call and we would not release dict

[nik9000]

Our normal way to do this is to set dict to null after you give it to k1 and then:

Releasables.closeExpectNoException(bytes, dict);

[ncordon]

Gotcha!

[elasticsearchmachine]

Pinging @elastic/es-analytical-engine (Team:Analytics)

[dnhatn]

LGTM. Thanks Nacho!

[elasticsearchmachine]

💔 Backport failed

Status	Branch	Result
❌	9.2	Commit could not be cherrypicked due to conflicts
❌	8.19	Commit could not be cherrypicked due to conflicts
❌	9.1	Commit could not be cherrypicked due to conflicts

You can use sqren/backport to manually backport by running backport --upstream elastic/elasticsearch --pr 137050

[ncordon]

No need to backport this, it seems this bug was introduced only for 9.3