race condition in OptimisticTransaction::CommitWithParallelValidate

[andoryu-]

Note: Please use Issues only for bug reports. For questions, discussions, feature requests, etc. post to dev group: https://groups.google.com/forum/#!forum/rocksdb or https://www.facebook.com/groups/rocksdb.dev

Expected behavior

OptimisticTransaction::CommitWithParallelValidate() actually prevent race between TransactionUtil::CheckKeysForConflicts() and DBImpl::Write(), as implied by its example code

Actual behavior

OptimisticTransaction::CommitWithParallelValidate() cannot prevect race between TransactionUtil::CheckKeysForConflicts() and DBImpl::Write()

Steps to reproduce the behavior

Read the source code, sepcifically between calling TransactionUtil::CheckKeysForConflicts and calling DBImpl::Write()

The moment thread T1 has just done executing L123, another thread T2 write something into db, then T1 executes L129, the transaction will falsely commit, contrary to what is implied by the occ example code

[ywave620]

I think the entire OptimisticTransaction::CommitWithParallelValidate() is protected by a lock.

[zergvszerg]

I think the entire OptimisticTransaction::CommitWithParallelValidate() is protected by a lock.

@ywave620
However the lock only guarantee mutual exclusion between transactions, not including DB::Write().