The Road to 2.18 - Customfields

[Sean McAfee]

Gervase Markham <gerv at mozilla.org> wrote:
>Christopher Hicks wrote:
>> Bravo.  Real data stumps more fear and loathing.

>But can you accurately scale up data gathered from an installation of 
>only 1760 bugs?

Fair enough.  I copied the production database I used to gather my
previously-described data over to a test machine, then ran a program which
replicated each existing bug one hundred times.  I did not replicate notes
or attachments, since they're irrelevant for this purpose.  The production
database had grown to about 2100 bugs; between that and a few dry runs
of the replication program, I eventually got this as a starting point:

bash-2.05a$ mysql -e 'select count(*) from bugs'
+----------+
| count(*) |
+----------+
|   214524 |
+----------+

Here are the new results.  Against short_desc only:

bash-2.05a$ time ./count-bugs.pl short_desc allwords Astro
35350 bugs found.

real    0m35.397s
user    0m4.890s
sys     0m0.380s

Against one custom fields table, CF_INTEGER:

bash-2.05a$ time ./count-bugs.pl short_desc allwords Astro tlc_incident_num '>=' 2700
11817 bugs found.

real    0m10.935s
user    0m2.800s
sys     0m0.240s

Against a second custom fields table, CF_SHORTSTRING:

bash-2.05a$ time ./count-bugs.pl short_desc allwords Astro tlc_incident_num
'>=' 2700 tlc_platform_serial_num allwords 'n/a'
2121 bugs found.

real    0m6.262s
user    0m1.850s
sys     0m0.170s

Against a third custom fields table, CF_LONGSTRING:

bash-2.05a$ time ./count-bugs.pl short_desc allwords Astro tlc_incident_num '>=' 2700 tlc_platform_serial_num allwords 'n/a' tlc_reproducibility_details allwords happens
1010 bugs found.

real    0m4.902s
user    0m1.760s
sys     0m0.140s

Far from revealing a performance penalty, it appears that the more (scalar)
custom fields one queries against, the better the results.


--Sean