|From:||Fabien COELHO <coelho(at)cri(dot)ensmp(dot)fr>|
|To:||Michael Paquier <michael(at)paquier(dot)xyz>|
|Cc:||Kyotaro Horiguchi <horikyota(dot)ntt(at)gmail(dot)com>, rulyox(at)gmail(dot)com, PostgreSQL Developers <pgsql-hackers(at)lists(dot)postgresql(dot)org>|
|Subject:||Re: Error on pgbench logs|
|Views:||Raw Message | Whole Thread | Download mbox | Resend email|
>> + /* flush remaining stats */
>> + if (!logged && latency == 0.0)
>> + logAgg(logfile, agg);
> You are right, this is missing the final stats. Why the choice of
> latency here for the check?
For me zero latency really says that there is no actual transaction to
count, so it is a good trigger for the closing call. I did not wish to add
a new "flush" parameter, or a specific function. I agree that it looks
> That's just to make the difference between the case where doLog() is
> called while processing the benchmark for the end of the transaction and
> the case where doLog() is called once a thread ends, no?
> Wouldn't it be better to do a final push of the states once a thread
> reaches CSTATE_FINISHED or CSTATE_ABORTED instead?
The call was already in place at the end of threadRun and had just become
ineffective. I did not wish to revisit its place and change the overall
structure, it is just a bug fix. I agree that it could be done differently
with the added logAgg function which could be called directly. Attached
another version which does that.
|Next Message||Tom Lane||2021-06-11 14:37:21||Re: recovery test failures on hoverfly|
|Previous Message||Alvaro Herrera||2021-06-11 14:08:20||Re: pg_regress.c also sensitive to various PG* environment variables|