| From: | Craig Ringer <craig(at)2ndquadrant(dot)com> |
|---|---|
| To: | Ashutosh Bapat <ashutosh(dot)bapat(at)enterprisedb(dot)com> |
| Cc: | Robert Haas <robertmhaas(at)gmail(dot)com>, Kevin Grittner <kgrittn(at)ymail(dot)com>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org> |
| Subject: | Re: Getting sorted data from foreign server for merge join |
| Date: | 2015-11-16 10:52:52 |
| Message-ID: | CAMsr+YGgASgk1wU6v92A=qw30Yb84yJk2kkT2aAt0GGS9jAmeQ@mail.gmail.com |
| Views: | Whole Thread | Raw Message | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
On 16 November 2015 at 17:47, Ashutosh Bapat <
ashutosh(dot)bapat(at)enterprisedb(dot)com> wrote:
>
>
> collations arising from a foreign table's var are considered to be safer
> (FDW_COLLATE_SAFE) to push down to the foreign server , since they are
> either local default collation or are assumed to be same on foreign and
> local server as per user declaration. The onus of making that sure is on
> the user who declares particular collation for a foreign table var. As said
> upthread, different glibc implementations can cause collation ordering
> mismatch, this patch will be susceptible to the same problem as
> master/standby problem.
>
>
Yeah. It's less bad than the related problems we already have:
* Upgrading glibc can cause your indexes to no longer match your local
collations
* Different glibc versions on master and replica(s) can have the same effect
I don't see a problem with adding another way this same issue can be
expressed, since there's no sane way to fix it _properly_ without either
versioned collations in glibc or bringing collation onboard into Pg.
--
Craig Ringer http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Craig Ringer | 2015-11-16 11:01:53 | Re: Proposal: "Causal reads" mode for load balancing reads without stale data |
| Previous Message | Simon Riggs | 2015-11-16 10:44:29 | Re: Proposal: "Causal reads" mode for load balancing reads without stale data |