2012/6/27 Robert Haas <robertmhaas@gmail.com>:
> On Wed, Jun 27, 2012 at 7:21 AM, Florian Pflug <fgp@phlo.org> wrote:
>> On Jun27, 2012, at 07:18 , Kohei KaiGai wrote:
>>> The problem is the way to implement it.
>>> If we would have permission checks on planner stage, it cannot handle
>>> a case when user-id would be switched prior to executor stage, thus
>>> it needs something remedy to handle the scenario correctly.
>>> Instead of a unique plan per query, it might be a solution to generate
>>> multiple plans depending on user-id, and choose a proper one in
>>> executor stage.
>>>
>>> Which type of implementation is what everybody is asking for?
>>
>> I think you need to
>>
>> a) Determine the user-id at planning time, and insert the matching
>> RLS clause
>>
>> b1) Either re-plan the query if the user-id changes between planning
>> and execution time, which means making the user-id a part of the
>> plan-cache key.
>>
>> b2) Or decree that for RLS purposes, it's the user-id at planning time,
>> not execution time, that counts.
>
> Or b3, flag plans that depend on the user ID inside the plan-cache,
> and just flush all of those (but only those) when the user ID changes.
> In the common case where RLS is not used, that might ease the sting.
>
Probably, PlannedStmt->invalItems allows to handle invalidation of
plan-cache without big code changes. I'll try to put a flag of user-id
to track the query plan with RLS assumed, or InvalidOid if no RLS
was applied in this plan.
I'll investigate the implementation for more details.
Do we have any other scenario that run a query plan under different
user privilege rather than planner stage?
Thanks,
--
KaiGai Kohei <kaigai@kaigai.gr.jp>