Todd Lipcon has posted comments on this change.

Change subject: KUDU-495 (part 2): ensure all catalog writes for an operation 
are batched
......................................................................


Patch Set 2:

(1 comment)

(just a response to a comment here - haven't looked at the new revision yet to 
see if the comment is moot)

http://gerrit.cloudera.org:8080/#/c/2695/2/src/kudu/master/catalog_manager.cc
File src/kudu/master/catalog_manager.cc:

Line 36: //   in iteration order (i.e. the same order as 
TableInfo::GetAllTablets()).
> Ah yes, I forgot that it's using std::map and not std::unordered_map. But t
But if you add a new tablet (or remove a tablet) from an unordered_map, the 
underlying implementation is free to reorder everything, right? i.e it's 
probably a hashtable, and adding an element invalidates any prior iteration 
orders.

So, I think it is better to explicitly state an order rather than say 
"arbitrary but consistent", because it's pretty hard to get anything like the 
latter if you're just relying on container iteration order of an unordered 
container.

(one option that's reasonable and unambiguous is pointer comparison order)


-- 
To view, visit http://gerrit.cloudera.org:8080/2695
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I5cbccf5ce22c005d7aa25bbdefe7502873a8ed7d
Gerrit-PatchSet: 2
Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-Owner: Adar Dembo <[email protected]>
Gerrit-Reviewer: Adar Dembo <[email protected]>
Gerrit-Reviewer: David Ribeiro Alves <[email protected]>
Gerrit-Reviewer: Kudu Jenkins
Gerrit-Reviewer: Mike Percy <[email protected]>
Gerrit-Reviewer: Todd Lipcon <[email protected]>
Gerrit-HasComments: Yes

Reply via email to