BUG #10404: auto_explain for EXPLAIN ANLYZE.

Started by Tomonari Katsumataalmost 12 years ago4 messagesbugs
Jump to latest
#1Tomonari Katsumata
katsumata.tomonari@po.ntts.co.jp

The following bug has been logged on the website:

Bug reference: 10404
Logged by: Tomonari Katsumata
Email address: katsumata.tomonari@po.ntts.co.jp
PostgreSQL version: 9.3.4
Operating system: CentOS 6.4 x86_64
Description:

Hi,

I'm testing auto_explain.
And I noticed that auto_explain has a strange behavior.

The settings for auto_explain are like below.

-------------------------------
testdb=# select name, setting from pg_settings where name like
'%auto_explain%';
name | setting
------------------------------------+---------
auto_explain.log_analyze | on
auto_explain.log_buffers | on
auto_explain.log_format | text
auto_explain.log_min_duration | 0
auto_explain.log_nested_statements | off
auto_explain.log_timing | on
auto_explain.log_verbose | on
(7 rows)
-------------------------------

And then, I tested five cases.

-------------------------------
a.EXPLAIN ONLY
b.EXPLAIN ANALYZE
c.EXPLAIN ANALYZE BUFFERS
d.EXPLAIN ANALYZE BUFFERS-OFF
e.EXPLAIN ANALYZE TIMING
f.EXPLAIN ANALYZE TIMING-OFF
-------------------------------

The results of a to e are good, but the result of f is not good.
I thought "actual time" would be omitted from the result of EXPLAIN,
but it is there.
Is this an intentional thing? or bug?

The test results are below.

===============
a.EXPLAIN ONLY
===============
testdb=# explain select * from hoge where i = 3;
QUERY PLAN
------------------------------------------------------
Seq Scan on hoge (cost=0.00..40.00 rows=12 width=4)
Filter: (i = 3)
(2 rows)

****************************************************************
LOG: duration: 0.000 ms plan:
Query Text: explain select * from hoge where i = 3;
Seq Scan on public.hoge (cost=0.00..40.00 rows=12 width=4)
Output: i
Filter: (hoge.i = 3)
****************************************************************

=================
b.EXPLAIN ANALYZE
=================
testdb=# explain analyze select * from hoge where i = 3;
QUERY PLAN

--------------------------------------------------------------------------------------------------
Seq Scan on hoge (cost=0.00..40.00 rows=12 width=4) (actual
time=23.484..23.484 rows=0 loops=1)
Filter: (i = 3)
Rows Removed by Filter: 1
Total runtime: 25.275 ms
(4 rows)

****************************************************************
LOG: duration: 23.488 ms plan:
Query Text: explain analyze select * from hoge where i = 3;
Seq Scan on public.hoge (cost=0.00..40.00 rows=12 width=4) (actual
time=23.484..23.484 rows=0 loops=1)
Output: i
Filter: (hoge.i = 3)
Rows Removed by Filter: 1
Buffers: shared hit=1
****************************************************************

=========================
c.EXPLAIN ANALYZE BUFFERS
=========================
testdb=# explain (analyze, buffers) select * from hoge where i = 3;
QUERY PLAN

------------------------------------------------------------------------------------------------
Seq Scan on hoge (cost=0.00..40.00 rows=12 width=4) (actual
time=0.152..0.152 rows=0 loops=1)
Filter: (i = 3)
Rows Removed by Filter: 1
Buffers: shared hit=1
Total runtime: 0.326 ms
(5 rows)

****************************************************************
LOG: duration: 0.155 ms plan:
Query Text: explain (analyze, buffers) select * from hoge where i = 3;
Seq Scan on public.hoge (cost=0.00..40.00 rows=12 width=4) (actual
time=0.152..0.152 rows=0 loops=1)
Output: i
Filter: (hoge.i = 3)
Rows Removed by Filter: 1
Buffers: shared hit=1
****************************************************************

=============================
d.EXPLAIN ANALYZE BUFFERS-OFF
=============================
testdb=# explain (analyze, buffers off) select * from hoge where i = 3;
QUERY PLAN

------------------------------------------------------------------------------------------------
Seq Scan on hoge (cost=0.00..40.00 rows=12 width=4) (actual
time=0.024..0.024 rows=0 loops=1)
Filter: (i = 3)
Rows Removed by Filter: 1
Total runtime: 1.407 ms
(4 rows)

****************************************************************
LOG: duration: 0.026 ms plan:
Query Text: explain (analyze, buffers off) select * from hoge where i = 3;
Seq Scan on public.hoge (cost=0.00..40.00 rows=12 width=4) (actual
time=0.024..0.024 rows=0 loops=1)
Output: i
Filter: (hoge.i = 3)
Rows Removed by Filter: 1
Buffers: shared hit=1
****************************************************************

========================
e.EXPLAIN ANALYZE TIMING
========================
testdb=# explain (analyze, timing) select * from hoge where i = 3;
QUERY PLAN

------------------------------------------------------------------------------------------------
Seq Scan on hoge (cost=0.00..40.00 rows=12 width=4) (actual
time=0.024..0.024 rows=0 loops=1)
Filter: (i = 3)
Rows Removed by Filter: 1
Total runtime: 0.207 ms
(4 rows)

****************************************************************
LOG: duration: 0.026 ms plan:
Query Text: explain (analyze, timing) select * from hoge where i = 3;
Seq Scan on public.hoge (cost=0.00..40.00 rows=12 width=4) (actual
time=0.024..0.024 rows=0 loops=1)
Output: i
Filter: (hoge.i = 3)
Rows Removed by Filter: 1
Buffers: shared hit=1
****************************************************************

============================
f.EXPLAIN ANALYZE TIMING-OFF
============================
testdb=# explain (analyze, timing off) select * from hoge where i = 3;
QUERY PLAN

------------------------------------------------------------------------------------------------
Seq Scan on hoge (cost=0.00..40.00 rows=12 width=4) (actual
time=0.023..0.023 rows=0 loops=1)
Filter: (i = 3)
Rows Removed by Filter: 1
Total runtime: 0.194 ms
(4 rows)

****************************************************************
LOG: duration: 0.026 ms plan:
Query Text: explain (analyze, timing off) select * from hoge where i = 3;
Seq Scan on public.hoge (cost=0.00..40.00 rows=12 width=4) (actual
time=0.023..0.023 rows=0 loops=1)
Output: i
Filter: (hoge.i = 3)
Rows Removed by Filter: 1
Buffers: shared hit=1
****************************************************************

regards,
----------------------
Tomonari Katsumata

--
Sent via pgsql-bugs mailing list (pgsql-bugs@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-bugs

#2Tom Lane
tgl@sss.pgh.pa.us
In reply to: Tomonari Katsumata (#1)
Re: BUG #10404: auto_explain for EXPLAIN ANLYZE.

katsumata.tomonari@po.ntts.co.jp writes:

[ auto_explain changes the user-visible behavior of EXPLAIN ]
Is this an intentional thing? or bug?

Yeah, that's a bug. The code in explain.c is looking at whether it got
timing results, not whether it asked for them ...

regards, tom lane

--
Sent via pgsql-bugs mailing list (pgsql-bugs@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-bugs

#3Tomonari Katsumata
katsumata.tomonari@po.ntts.co.jp
In reply to: Tom Lane (#2)
Re: BUG #10404: auto_explain for EXPLAIN ANLYZE.

Hi Tom,

Thank you for checking it.

(2014/05/20 20:51), Tom Lane wrote:

katsumata.tomonari@po.ntts.co.jp writes:

[ auto_explain changes the user-visible behavior of EXPLAIN ]
Is this an intentional thing? or bug?

Yeah, that's a bug. The code in explain.c is looking at whether it got
timing results, not whether it asked for them ...

I made a patch for this bug.
Please see the attached patch.

According to my understanding,
explain.c flows like below.

ExplainOnePlan
- call ExecutorStart(explain_ExecutorStart)
- call ExecutorRun(explain_ExecutorRun)
- call ExecutorFinish(explain_ExecutorFinish)
- call ExplainPrintPlan *A*
- call ExecutorEnd(explain_ExecutorEnd) *B*

The query result is printed at *A*, and then
auto_explain log message is printed at *B*.

So I add a check in explain_ExecutorFinish if the query has
"EXPLAIN ANALYZE TIMING OFF" or not to supress timing output
at *A*.
And I add a check in explain_ExecutorEnd if the setting of
auto_explain need timing information or not.

To distinguish "EXPLAIN ANALYZE TIMING OFF" or not, I use
queryDesc->instrument_options has INSTRUMENT_ROWS flag or not.
And also to distinguish if auto_explain needs timing info or not,
I use queryDesc->instrument_options has INSTRUMENT_TIMING flag or not.

This works fine, but I'm not sure I should to use INSTRUMENT_ROWS flag
for this. Because it seems that the flags is used but not referenced
from anywhere in current code.
Will this flag be obsoleted?

regards,
--------------------
Tomonari Katsumata

Attachments:

fix-auto_explain-explain-analyze-timing-off.patchtext/x-patch; name=fix-auto_explain-explain-analyze-timing-off.patchDownload+6-0
#4Tomonari Katsumata
katsumata.tomonari@po.ntts.co.jp
In reply to: Tomonari Katsumata (#3)
Re: BUG #10404: auto_explain for EXPLAIN ANLYZE.

Hi Tom,

You fixed this problem already.
Sorry for confusing you...
please forget my last mail.

Thank you for alot!

(2014/05/21 11:37), Tomonari Katsumata wrote:

Hi Tom,

Thank you for checking it.

(2014/05/20 20:51), Tom Lane wrote:

katsumata.tomonari@po.ntts.co.jp writes:

[ auto_explain changes the user-visible behavior of EXPLAIN ]
Is this an intentional thing? or bug?

Yeah, that's a bug. The code in explain.c is looking at whether it got
timing results, not whether it asked for them ...

I made a patch for this bug.
Please see the attached patch.

According to my understanding,
explain.c flows like below.

ExplainOnePlan
- call ExecutorStart(explain_ExecutorStart)
- call ExecutorRun(explain_ExecutorRun)
- call ExecutorFinish(explain_ExecutorFinish)
- call ExplainPrintPlan *A*
- call ExecutorEnd(explain_ExecutorEnd) *B*

The query result is printed at *A*, and then
auto_explain log message is printed at *B*.

So I add a check in explain_ExecutorFinish if the query has
"EXPLAIN ANALYZE TIMING OFF" or not to supress timing output
at *A*.
And I add a check in explain_ExecutorEnd if the setting of
auto_explain need timing information or not.

To distinguish "EXPLAIN ANALYZE TIMING OFF" or not, I use
queryDesc->instrument_options has INSTRUMENT_ROWS flag or not.
And also to distinguish if auto_explain needs timing info or not,
I use queryDesc->instrument_options has INSTRUMENT_TIMING flag or not.

This works fine, but I'm not sure I should to use INSTRUMENT_ROWS flag
for this. Because it seems that the flags is used but not referenced
from anywhere in current code.
Will this flag be obsoleted?

regards,
--------------------
Tomonari Katsumata

--
Sent via pgsql-bugs mailing list (pgsql-bugs@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-bugs