On Aug 17, 2011, at 11:46 AM, Jean-Pierre Fiset wrote:

> I think that the operations of replication and backing up are quite 
> different. Although some are using the replication features for backing up, I 
> tend to think of replication as an operation taking place between two nodes 
> that do not necessarily trust one another.

That's one possible use case for replication, but hardly the only one.  Anyway, 
if you don't trust the replication then I certainly hope the replication 
doesn't use credentials that map to _admin powers on your database.  If the 
replication doesn't have _admin powers it cannot bypass validation.

> If what you are proposing is a special privilege given to the admin party, 
> then I do not have much of an issue with this, since administrators already 
> have intimate access to the server. However, the concept of creating a new 
> "replicator" role, which would supersede the validation functions is another 
> thing.

Yes, I probably should have picked one approach and stuck with it.  Either way, 
my intent was that a replicator could bypass validation only if an admin had 
given it credentials that mapped to a powerful role (possibly _admin), *and* if 
the admin had explicitly asked for the replicator to bypass validation.

> In applications that must ensure that some document types have a given 
> structure, opening the door to a user (and here I assume a user that attempts 
> a replication from a different node, not a local administrator performing a 
> back up) to work around the validation function is probably a bad idea.

That's not going to happen, unless you granted the user this really powerful 
role.  Don't do that.

> If the validation function could not be counted on, it would really affect 
> the way an application must be written.

Understood, I'm certainly not asking for the replicator to bypass validations 
in general.  Cheers,

Adam

Reply via email to