Re: test_json_parser/002_inline is kind of slow - Mailing list pgsql-hackers

Robert Haas <robertmhaas@gmail.com> writes:
> On Sat, Sep 27, 2025 at 12:43 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> I tried it on my Mac M4 laptop, and saw the 002 test dropping from
>> 3592 ms to 540 ms, about an 85% savings.  So I agree this is worth
>> doing.  It's still unclear why Robert is seeing times so much worse
>> than mine, though.

> I don't have a clear explanation for that slowness either, but the
> patch cuts down the runtime by 89% on my machine.

Over the weekend I updated my macOS buildfarm animals to more recent
OS versions, and noticed this:

sifaka (Mac Mini M4 Pro), on macOS 15 (Sequoia) [1]

[06:50:39] t/001_test_json_parser_incremental.pl .. ok      868 ms ( 0.01 usr  0.00 sys +  0.33 cusr  0.48 csys =  0.82
CPU)
[06:50:44] t/002_inline.pl ........................ ok     5485 ms ( 0.06 usr  0.02 sys +  2.04 cusr  3.20 csys =  5.32
CPU)
[06:50:44] t/003_test_semantic.pl ................. ok       64 ms ( 0.00 usr  0.00 sys +  0.04 cusr  0.02 csys =  0.06
CPU)
[06:50:44] t/004_test_parser_perf.pl .............. ok       41 ms ( 0.00 usr  0.00 sys +  0.03 cusr  0.01 csys =  0.04
CPU)

sifaka (Mac Mini M4 Pro), on macOS 26 (Tahoe) [2]

[02:42:28] t/001_test_json_parser_incremental.pl .. ok      882 ms ( 0.01 usr  0.00 sys +  0.33 cusr  0.48 csys =  0.82
CPU)
[02:42:33] t/002_inline.pl ........................ ok     5573 ms ( 0.06 usr  0.02 sys +  2.06 cusr  3.18 csys =  5.32
CPU)
[02:42:34] t/003_test_semantic.pl ................. ok       66 ms ( 0.00 usr  0.00 sys +  0.04 cusr  0.02 csys =  0.06
CPU)
[02:42:34] t/004_test_parser_perf.pl .............. ok       42 ms ( 0.00 usr  0.00 sys +  0.03 cusr  0.01 csys =  0.04
CPU)

longfin (2018 Mac Mini, Intel-based), on macOS 13 (Ventura) [3]

[07:02:37] t/001_test_json_parser_incremental.pl .. ok     2355 ms ( 0.05 usr  0.01 sys +  1.21 cusr  1.05 csys =  2.32
CPU)
[07:02:52] t/002_inline.pl ........................ ok    14844 ms ( 0.29 usr  0.04 sys +  7.38 cusr  7.02 csys = 14.73
CPU)
[07:02:52] t/003_test_semantic.pl ................. ok      187 ms ( 0.00 usr  0.00 sys +  0.13 cusr  0.06 csys =  0.19
CPU)
[07:02:52] t/004_test_parser_perf.pl .............. ok      144 ms ( 0.01 usr  0.00 sys +  0.11 cusr  0.03 csys =  0.15
CPU)

longfin (2018 Mac Mini, Intel-based), on macOS 15 (Sequoia) [4]

[02:39:51] t/001_test_json_parser_incremental.pl .. ok     4005 ms ( 0.06 usr  0.00 sys +  1.86 cusr  1.92 csys =  3.84
CPU)
[02:40:16] t/002_inline.pl ........................ ok    25139 ms ( 0.33 usr  0.04 sys + 11.47 cusr 12.23 csys = 24.07
CPU)
[02:40:17] t/003_test_semantic.pl ................. ok      244 ms ( 0.00 usr  0.01 sys +  0.15 cusr  0.09 csys =  0.25
CPU)
[02:40:17] t/004_test_parser_perf.pl .............. ok      153 ms ( 0.00 usr  0.00 sys +  0.11 cusr  0.03 csys =  0.14
CPU)

longfin is intentionally an OS release behind sifaka, so these numbers
aren't directly comparable.  However, the previous speed ratio of
about 2.75x is in line with the two machines' single-core performance
difference according to geekbench; the new ratio of 4.5x is not.

I conclude that Apple broke something about the speed of process
forking in the last OS release or two, but only on their Intel
machines.  I doubt that they care a whole lot, given they're about
to stop supporting Intel hardware altogether :-(

Anyway, these numbers prompt me to wonder exactly which macOS
release Robert is running, and on what kind of hardware.

            regards, tom lane

[1]
https://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=sifaka&dt=2025-09-27%2010%3A42%3A05&stg=module-test_json_parser-check
[2]
https://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=sifaka&dt=2025-09-29%2006%3A34%3A17&stg=module-test_json_parser-check
[3]
https://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=longfin&dt=2025-09-27%2010%3A42%3A01&stg=module-test_json_parser-check
[4]
https://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=longfin&dt=2025-09-29%2006%3A20%3A01&stg=module-test_json_parser-check