From: | "houzj(dot)fnst(at)fujitsu(dot)com" <houzj(dot)fnst(at)fujitsu(dot)com> |
---|---|
To: | "tsunakawa(dot)takay(at)fujitsu(dot)com" <tsunakawa(dot)takay(at)fujitsu(dot)com>, 'Tom Lane' <tgl(at)sss(dot)pgh(dot)pa(dot)us>, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> |
Cc: | Bharath Rupireddy <bharath(dot)rupireddyforpostgres(at)gmail(dot)com>, PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org> |
Subject: | RE: [bug?] Missed parallel safety checks, and wrong parallel safety |
Date: | 2021-04-22 09:08:49 |
Message-ID: | OS0PR01MB5716ED7A2E0199D50DCCBF2E94469@OS0PR01MB5716.jpnprd01.prod.outlook.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
> Thank you, fmgr_info() looks like the best place to do the parallel safety check.
> Having a quick look at its callers, I didn't find any concerning place (of course,
> we can't be relieved until the regression test succeeds.) Also, with fmgr_info(),
> we don't have to find other places to add the check to deal with functions calls
> in execExpr.c and execExprInterp.c. This is beautiful.
>
> But the current fmgr_info() does not check the parallel safety of builtin
> functions. It does not have information to do that. There are two options.
> Which do you think is better? I think 2.
>
> 1) fmgr_info() reads pg_proc like for non-builtin functions This ruins the effort
> for the fast path for builtin functions. I can't imagine how large the adverse
> impact on performance would be, but I'm worried.
For approach 1): I think it could result in infinite recursion.
For example:
If we first access one built-in function A which have not been cached,
it need access the pg_proc, When accessing the pg_proc, it internally still need some built-in function B to scan.
At this time, if B is not cached , it still need to fetch function B's parallel flag by accessing the pg_proc.proparallel.
Then it could result in infinite recursion.
So, I think we can consider the approach 2)
Best regards,
houzj
From | Date | Subject | |
---|---|---|---|
Next Message | Justin Pryzby | 2021-04-22 09:16:04 | Re: truncating timestamps on arbitrary intervals |
Previous Message | Patrik Novotny | 2021-04-22 08:57:37 | Re: RFE: Make statistics robust for unplanned events |