[ovs-dev] [PATCH] ovsdb: Force columns that contain weak references to be immutable.

[Elluru, Krishna Mohan]

Hi Ben,
	If the below idea of allowing NULL values for Indexed columns makes sense., shall we add an attribute (allow_null_index = false kind) as an optional parameter for such type of columns and ensure compatibility with existing/old schema? Does it sound okay?

Thanks
Krishna Mohan

-----Original Message-----
From: Ben Pfaff [mailto:blp at ovn.org] 
Sent: Monday, April 11, 2016 11:00 AM
To: Ansari, Shad <shad.ansari at hpe.com>
Cc: Elluru, Krishna Mohan <elluru.kri.mohan at hpe.com>; dev at openvswitch.org
Subject: Re: [ovs-dev] [PATCH] ovsdb: Force columns that contain weak references to be immutable.

On Mon, Apr 04, 2016 at 10:38:40PM +0000, Ansari, Shad wrote:
> > 
> > HI Ben,
> > 	How does column which is weak reference  and also configured as an 
> > Index? If it is mutable, and the corresponding UUID going 
> > away(deletion of record) leaves column value as NULL right? And what 
> > would be the behavior if the table is a Root table also?
> > I am listing down, my assumptions:
> > 
> > 	1. In table X, If the column is index and a weak reference to table 
> > y, on table y row deletion, the index field would become NULL and 
> > since it is an index, constraint violation would occur and to avoid, 
> > it is expected to remove the row first in table X and table Y row 
> > deletion can happen in same txn or consequent transactions. Correct?
> > 	2. If the table X is also a root table, and one of its column is an 
> > index and weak reference to table Y, on row deletion in table Y, 
> > constraint violation would occur, hence similar to above table X row 
> > cleanup is expected to happen prior to table Y row deletion.
> > 
> > 	If above 2 statements is true, index with weak reference is 
> > becoming equivalent to strong reference. Am I correct?
> > 
> > Thanks
> > Krishna Mohan
> 
> Would it make sense to allow NULL values for columns that are references and indexes. Something like what MySQL allows for unique keys.
> 
> http://dev.mysql.com/doc/refman/5.0/en/create-index.html

That's a reasonable idea.  It could theoretically break some currently working use cases, if someone has an OVSDB schema that relies on an index to ensure the uniqueness of an empty column.  I don't know whether there are schemas out there that rely on this property.

I don't know how this idea would extend to multi-column indexes.