From: | Ashutosh Bapat <ashutosh(dot)bapat(at)enterprisedb(dot)com> |
---|---|
To: | Robert Haas <robertmhaas(at)gmail(dot)com> |
Cc: | amul sul <sulamul(at)gmail(dot)com>, Dilip Kumar <dilipbalaut(at)gmail(dot)com>, Yugo Nagata <nagata(at)sraoss(dot)co(dot)jp>, David Steele <david(at)pgmasters(dot)net>, Greg Stark <stark(at)mit(dot)edu>, PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: [POC] hash partitioning |
Date: | 2017-05-18 03:58:22 |
Message-ID: | CAFjFpRdYOY4O5S0rB4dxhcF5iQoFKeXf4Pb+vf_-fgt2EeQRAg@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
On Wed, May 17, 2017 at 11:51 PM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> On Wed, May 17, 2017 at 1:41 AM, Ashutosh Bapat
> <ashutosh(dot)bapat(at)enterprisedb(dot)com> wrote:
>>> Fixed in the attached version; used "hash partition remainder must be
>>> greater than or equal to 0" instead.
>>
>> I would suggest "non-zero positive", since that's what we are using in
>> the documentation.
>
> Well, that's not very good terminology, because zero is not a positive
> number. Existing error messages seem to use phrasing such as "THING
> must be a positive integer" when zero is not allowed or "THING must be
> a non-negative integer" when zero is allowed. For examples, do git
> grep errmsg.*positive or git grep errmsg.*negative.
Ok. We need to change all the usages in the documentation and in the
comments to non-negative. The point is to use same phrases
consistently.
>
>> In partition_bounds_equal(), please add comments explaining why is it safe to
>> check just the indexes? May be we should add code under assertion to make sure
>> that the datums are equal as well. The comment could be something
>> like, "If two partitioned tables have different greatest moduli, their
>> partition schemes don't match. If they have same greatest moduli, and
>> all remainders have different indexes, they all have same modulus
>> specified and the partitions are ordered by remainders, thus indexes
>> array will be an identity i.e. index[i] = i. If the partition
>> corresponding to a given remainder exists, it will have same index
>> entry for both partitioned tables or if it's missing it will be -1.
>> Thus if indexes array matches, corresponding datums array matches. If
>> there are multiple remainders corresponding to a given partition,
>> their partitions are ordered by the lowest of the remainders, thus if
>> indexes array matches, both of the tables have same indexes arrays, in
>> both the tables remainders corresponding to multiple partitions all
>> have same indexes and thus same modulus. Thus again if the indexes are
>> same, datums are same.".
>
> That seems quite long.
I have shared a patch containing a denser explanation with my last set
of comments.
--
Best Wishes,
Ashutosh Bapat
EnterpriseDB Corporation
The Postgres Database Company
From | Date | Subject | |
---|---|---|---|
Next Message | Thomas Munro | 2017-05-18 04:43:04 | Re: Race conditions with WAL sender PID lookups |
Previous Message | Amit Kapila | 2017-05-18 03:43:55 | Re: UPDATE of partition key |