Re: PATCH: pgbench - aggregation of info written into log - Mailing list pgsql-hackers

From Tomas Vondra
Subject Re: PATCH: pgbench - aggregation of info written into log
Date
Msg-id 505B8318.5050101@fuzzy.cz
Whole thread Raw
In response to Re: PATCH: pgbench - aggregation of info written into log  (Jeff Janes <jeff.janes@gmail.com>)
Responses Re: PATCH: pgbench - aggregation of info written into log  (Alvaro Herrera <alvherre@2ndquadrant.com>)
List pgsql-hackers
On 3.9.2012 01:28, Jeff Janes wrote:
> On Sun, Sep 2, 2012 at 3:46 PM, Tomas Vondra <tv@fuzzy.cz> wrote:
>>
>> Fixed. I've kept use_log_agg only and I've removed the default. Also
>> I've added one more check (that the total duration is a multiple of
>> the aggregation interval).
>
> Hi Tomas,
>
> In the docs, -l is an option, not an application:
> "<application>-l</application>"
>
> Also, the docs still refer to <option>-A</option>, rather than to
> <option>--aggregate-interval</option>, in a few places.

Fixed.

> I think a section in the docs that points out that transactions run by
> specifying mulitple -f will be mingled together into an interval would
> be a good idea, as that could easily be overlooked (as I did earlier).

Fixed, added a paragraph mentioning this.

> The docs do not mention anywhere what the units for the latencies are.
>
> I think the format of the logfile should somehow make it unmistakably
> different from the regular -l logfile, for example by prefixing every
> line with "Aggregate: ".   Or maybe there is some other solution, but
> I think that having two log formats, both of which consist of nothing
> but 6 integers, but yet mean very different things, is a recipe for
> confusion.

Not sure how to do that, I'd say it's a responsibility of the user.

> Is it unavoidable that the interval be an integral number of seconds?
> I found the units in the original code confusing, so maybe there is
> some reason it needs to be like that that I don't understand yet.
> I'll look into it some more.

Not really, but I don't see a real use case for such intervals, so I'm
keeping the integers for now.

Also, I've realized the last interval may not be logged at all - I'll
take a look into this in the next version of the patch.

Tomas

Attachment

pgsql-hackers by date:

Previous
From: Tom Lane
Date:
Subject: Re: Re: proposal and patch : support INSERT INTO...RETURNING with partitioned table using rule
Next
From: "Kevin Grittner"
Date:
Subject: Assigning NULL to a record variable