IGNITE-20052 Release all locks locally on self primary replica expira…

[vldpyatkov]

…tion

https://issues.apache.org/jira/browse/IGNITE-20052

[Cyrill]

listeners

[vldpyatkov]

Yes, rewrited the javadoc.

[Cyrill]

is it okay that we ignore the previous notification? What are the cases when it's possible?
Cannot we have smth like this?
imagine we have a few listeners, A, B and C
and the execution is:
notify(A, old)
notify(A, new)
notify(B, old)
notify(B, new)
notify(C, new)
notify(C, old)

[vldpyatkov]

It could happens in the case when a lease expired, but the lease does not accepeted. I think again about it and desided to no trigger the expire event in the case.
Currently, we can change this warning message to just checkoing assertion.

[Cyrill]

Iteration over 'txCleanupReadyFutures.keySet()' may be replaced with 'entrySet()' iteration.

But strictly speaking, the processing of txCleanupReadyFutures is not atomic here.
If we allow to execute cleanup on non-primary replicas and cleanup this map (as stated in IGNITE-18617), this code might start to crash -
the keyset iterator will return a txId, but txCleanupReadyFutures.get(txId) might return null

[vldpyatkov]

Rewrote the place to the Map#compute invocation. I hope it allows us to not think about CME.

[Cyrill]

Thank you.
What do you think if we change the code a bit?

 txCleanupReadyFutures.compute(txId, (id, txOps) -> {
                    if (txOps == null || isFinalState(txOps.state)) {
                        return null;
                    }

                    if (!txOps.futures.isEmpty()) {
                        CompletableFuture<?>[] txFuts = txOps.futures.values().stream()
                                .flatMap(Collection::stream)
                                .toArray(CompletableFuture[]::new);

                        futs.add(allOf(txFuts).whenComplete((unused, throwable) -> releaseTxLocks(txId)));
                    }

                    return txOps;
                });

[vldpyatkov]

Ok.