Peter Eisentraut <peter_e(at)gmx(dot)net> wrote:
> On ons, 2011-04-06 at 11:01 -0700, Jeff Davis wrote:
>> One of the use cases of SSI is to enforce constraints, which are
>> certainly of interest to DBAs.
>
> Well, but users can freely change the isolation level, so it would
> not really be an effective constraint mechanism.
In our shop we plan to have a trigger check to ensure the isolation
level is serializable for all production tables. In the long run,
it sure would be convenient to have a GUC setting to prohibit
explicitly setting the isolation level to something less strict than
the default, but it didn't seem wise to try to do that in the
initial release with the feature. People are probably going to want
a little wiggle room until the technique has some miles on it.
So, sure -- in our shop a user can change the isolation level, but
if they try to modify production data they will get an error with
such a setting, and if they view data from a read only transaction
in a state where a transient anomaly is visible to them, I guess
that's on them.
-Kevin