Loading should be transactional?

[GoogleCodeExporter]

Currently loading is not transactional. This causes the following issue:

Use more than one partition, and then load a dataset that has a duplicate key. 
You will end up getting an exception saying duplicate key found. However, if 
you query the dataset afterwards, you will get partial results. The reason is 
that the load in those partitions that didn't get duplicate keys ended up 
succeeding and the disk components are marked as valid.

Remember that if you insert multiple records using a single insert statement, 
and an error occur, we don't roll back those inserts that made it successfully.
Is the loading issue comparable to this case? if yes, then it is not an issue. 
Otherwise, we must fix it.

Original issue reported on code.google.com by salsuba...@gmail.com on 5 Jul 2013 at 7:10

[GoogleCodeExporter]

I don't think loading needs to be transactional, exactly, but it has to be 
clear that the result of trying to load isn't complete.  I.e., it seems like 
the problem here was that JArod thought he'd loaded properly, but hadn't, and 
didn't get feedback saying "Nice try - bad data - partly loaded, partly not was 
the outcome".  It could then be left up to users to decide if they want to drop 
and recreate stuff, delete all and reload stuff, or try to pick up the pieces.  
Does that make sense?  (If not, I may have missed some of the points being 
made; apologies if that's the case...)

Original comment by dtab...@gmail.com on 5 Jul 2013 at 9:52

[GoogleCodeExporter]

@Sattam
Could you close this issue and file an another issue saying "Erroneous loading 
due to the bad data doesn't return proper error message to user"?
According to the Mike's comment, this issue is not vaild any more. 

Young-Seok

Original comment by kiss...@gmail.com on 15 Nov 2013 at 5:28

[GoogleCodeExporter]

Original comment by kiss...@gmail.com on 15 Nov 2013 at 7:41

• Added labels: Type-Enhancement
• Removed labels: Type-Defect