Horton

Re: pgbench logging broken by time logic changes

Fabien COELHO <coelho@cri.ensmp.fr>

View thread View on postgr.es
From: Fabien COELHO <coelho@cri.ensmp.fr>
To: Hannu Krosing <hannuk@google.com>
Cc: Kyotaro Horiguchi <horikyota.ntt@gmail.com>, nagata@sraoss.co.jp, gregsmithpgsql@gmail.com, pgsql-hackers <pgsql-hackers@postgresql.org>, david.christensen@crunchydata.com
Date: 2021-07-08T17:25:38Z
Lists: pgsql-hackers 
Hello Hannu,

>> I'm not sure we have transaction lasts for very short time that
>> nanoseconds matters.
>
> Nanoseconds may not matter yet, but they could be handy when for
> example we want to determine the order of parallel query executions.
>
> We are less than an order of magnitude away from being able to do 1M
> inserts/updates/deletes per second, so microseconds already are not
> always 100% reliable.

ISTM that 1M tps would be with really a lot of parallel clients, thus the 
latency of each would be quite measurable, so that µs would still make 
sense for measuring their performance? If an actual network is involved, 
the network latency is already 100-200 µs even before executing any code.

-- 
Fabien.

Commits

  1. Fix pgbench timestamp bugs.

  2. Fix pattern matching logic for logs in TAP tests of pgbench

  3. pgbench: Improve time logic.

  4. pgbench: Synchronize client threads.

  5. pgbench: refactor handling of stats tracking