From: | Kyotaro HORIGUCHI <horiguchi(dot)kyotaro(at)lab(dot)ntt(dot)co(dot)jp> |
---|---|
To: | tgl(at)sss(dot)pgh(dot)pa(dot)us |
Cc: | noah(at)leadboat(dot)com, peter_e(at)gmx(dot)net, robertmhaas(at)gmail(dot)com, pgsql-hackers(at)postgresql(dot)org |
Subject: | Re: UNION ALL on partitioned tables won't use indices. |
Date: | 2014-01-14 09:04:47 |
Message-ID: | 20140114.180447.145186052.horiguchi.kyotaro@lab.ntt.co.jp |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
This is cont'd from previous CF3.
You'll see the overview and the discussion since in the thread
begins from there. The issue ramains as of current 9.4dev head.
http://www.postgresql.org/message-id/20131024.193953.233464126.horiguchi.kyotaro@lab.ntt.co.jp
The issue in brief is that UNION ALL on inheritance tables does
not make use of indices in contrast to that on bare tables. This
is because of appendrels with the depth more than 2 levels
brought by inheritance tables.
Some of UNION ALL operation - especially using ORDERBY and LIMIT
- will be accelerated by this patch. Details in the message
above.
I proposed 3 types of solution but the one of them - tweaking
Equivalence Class (Type 1)- was dismissed because of
inappropriate manipulation on Equivalence Class. The rest do
essentially the same thing - flattening appendrels - at different
timings.
In type 2, the process is implemented as a generic appendrel
flattening function and applied after expand_inherited_tables()
in subquery_planner.
In type 3, the equivelant process is combined in
expand_inherited_rtentry().
Type 2 loops once for whole appendrel list (in most cases) so it
might be more effective than type 3 which searches the parent
appendrel for each appended ones. Type 3 is more approprieate
design assuming that appendrel tree deeper than 2 levels will be
generated only by expand_inherited_tables().
The attached two patches are rebased to current 9.4dev HEAD and
make check at the topmost directory and src/test/isolation are
passed without error. One bug was found and fixed on the way. It
was an assertion failure caused by probably unexpected type
conversion introduced by collapse_appendrels() which leads
implicit whole-row cast from some valid reltype to invalid
reltype (0) into adjust_appendrel_attrs_mutator().
Attached files are the two versions of patches mentioned above.
- unionall_inh_idx_typ2_v4_20140114.patch
- unionall_inh_idx_typ3_v4_20140114.patch
regards,
--
Kyotaro Horiguchi
NTT Open Source Software Center
Attachment | Content-Type | Size |
---|---|---|
unionall_inh_idx_typ3_v4_20140114.patch | text/x-patch | 6.5 KB |
unionall_inh_idx_typ2_v4_20140114.patch | text/x-patch | 9.0 KB |
From | Date | Subject | |
---|---|---|---|
Next Message | Kyotaro HORIGUCHI | 2014-01-14 09:08:10 | Re: Get more from indices. |
Previous Message | Alexander Korotkov | 2014-01-14 08:59:20 | Re: GIN improvements part 1: additional information |