2012/6/4 Tom Lane <tgl@sss.pgh.pa.us>:
> Kohei KaiGai <kaigai@kaigai.gr.jp> writes:
>> Here is two problems around RLSBYPASS. The first is we have
>> no idea to handle invalidation of prepared-statement when current
>> user is switched, right now.
>
> How is that specifically the fault of RLSBYPASS? *Any* of the schemes
> you're proposing for inlined RLS checks will have problems with userID
> switching.
>
Really? I don't find out a scenario that cause a problem with user-id
switching in case when RLS policy is *unconditionally* appended then
evaluated on executor stage. I'd like to see the scenario.
> My guess is we'd have to treat the effective userID as part of the
> plancache lookup key to make it safe to inline anything related to RLS.
>
It might be a solution, if we append individual RLS policy at the
planner stage, depending on user-id.
Thanks,
--
KaiGai Kohei <kaigai@kaigai.gr.jp>