On 18/07/18 01:43, Fabien COELHO wrote:
The more reasonable alternative could be to always last 2 seconds under
-T 2, even if the execution can be shorten because there is nothing to do
at all, i.e. remove the environment-based condition but keep the sleep.
That sounds reasonable. It's a bit silly to wait when there's nothing to do,
but it's also weird if the test exits before the specified time is up. Seems
less surprising to always sleep.
I did that in the attached version: no more environment variable hack, and
no execution shortcut even if there is nothing to do.
I also had to reproduce the progress logic to keep on printing report of
(no) progress in this tailing phase.
On second thoughts, there's one problem with this approach of always
waiting until -T is up. What if all the threads died because of errors?
For example:
pgbench postgres -T 10 -P 1 -c 2
starting vacuum...end.
client 0 aborted in command 5 (SQL) of script 0; ERROR: relation
"pgbench_accounts" does not exist
LINE 1: UPDATE pgbench_accounts SET abalance = abalance + -4963 WHER...
^
client 1 aborted in command 5 (SQL) of script 0; ERROR: relation
"pgbench_accounts" does not exist
LINE 1: UPDATE pgbench_accounts SET abalance = abalance + -2985 WHER...
^
progress: 1.0 s, 0.0 tps, lat 0.000 ms stddev 0.000
progress: 2.0 s, 0.0 tps, lat 0.000 ms stddev 0.000
progress: 3.0 s, 0.0 tps, lat 0.000 ms stddev 0.000
progress: 4.0 s, 0.0 tps, lat 0.000 ms stddev 0.000
progress: 5.0 s, 0.0 tps, lat 0.000 ms stddev 0.000
progress: 6.0 s, 0.0 tps, lat 0.000 ms stddev 0.000
progress: 7.0 s, 0.0 tps, lat 0.000 ms stddev 0.000
progress: 8.0 s, 0.0 tps, lat 0.000 ms stddev 0.000
progress: 9.0 s, 0.0 tps, lat 0.000 ms stddev 0.000
progress: 10.0 s, 0.0 tps, lat 0.000 ms stddev 0.000
transaction type: <builtin: TPC-B (sort of)>
scaling factor: 1
query mode: simple
number of clients: 2
number of threads: 1
duration: 10 s
number of transactions actually processed: 0
I don't think you want to wait in that situation. I think we should wait
at the end only if there some threads still alive, with nothing to do
only because of --rate.
- Heikki
