[DocDB] Single Shard waiters don't make progress when the blocking subtransaction is rolled back

[basavaraj29]

Jira Link: DB-9051

Description

begin TRANSACTION ISOLATION LEVEL READ COMMITTED;
savepoint a;
select * from test where k=1 for update;                            
                                                       update test set v=v+1 where k=1; // ends up getting blocked
rollback to a;
                                                       // doesn't make progress, why?
commit;
                                                       // goes through now

The single shard txn on the right can ideally make progress just after the rollback of the sub-transaction on the left, but it doesn't.
If we have a distributed transaction on the right, it goes through just after the rollback and doesn't wait for the left txn to commit.

upon checking, we override the below in case of OperationConflictResolverContext,

bool CheckConflictWithPending(const TransactionConflictData& transaction_data) override {
  return false;
}

which seems to give us a wrong impression that the wait for is still valid.

for the distributed transaction case, we have

bool CheckConflictWithPending(const TransactionConflictData& transaction_data) override {
  return transaction_data.conflict_info->subtransactions.empty();
}

which I think can be re-used for the single shard waiters as well.

Issue Type

kind/bug

Warning: Please confirm that this issue does not contain any sensitive information

• I confirm this issue does not contain any sensitive information.

[basavaraj29]

Ignoring intents from aborted subtransactions was introduced in 4fb1676 (v2.13.3.0). Could look at backporting this enhancement upto 2.14.