Corruption error when calling GetUpdatesSince()

[graetzer]

When using the GetUpdatesSince(..) method we regulary see a corruption error with the message "NO MORE DATA LEFT".

It seems to me like there is a condition in the TransactionLogIteratorImpl::NextImpl() method, that would always lead to this error when a new WAL file is openened (while concurrently scanning the WAL). The code in question is this piece:

is_valid_ = false;
if (current_last_seq_ == versions_->LastSequence()) {
  current_status_ = Status::OK();
} else {
  current_status_ = Status::Corruption("NO MORE DATA LEFT");
}

Seems like the sequence number returned by versions_->LastSequence() would be higher in that case ? Would it be more correct to return a "try again" type error ?

Expected behavior

No corruption error on scanning the WAL when executing basically this code:

std::unique_ptr<rocksdb::TransactionLogIterator> it; 
rocksdb::TransactionLogIterator::ReadOptions ro(false);
rocksdb::Status s = db->GetUpdatesSince(startingFrom, &it, ro);
for (; it->Valid(); it->Next()) {
   // ..
}
if (!it->status().ok())
   std::cout << it->status().ToString();

Actual behavior

We regulary see a corruption error.

Steps to reproduce the behavior

Scan the WAL via GetUpdatesSince and run parallel threads writing into the database instance.

[maysamyabandeh]

Thanks @graetzer. Changing it to ::TryAgain makes sense to me. Are you interested in submitting a PR?

[graetzer]

Sure thing #5474