Thread: self join for history analyzis

Hello the list,

For some time now, I'm struggling with a problem of self join of a table:

I have a usage recording table: CREATE TABLE readings(tm timestamp,
bytesin int, bytesout int);

The readouts are made "occasionally" - the timespan between the readouts
are not very precise, but there is a lot of those readouts.

I need to make a self join of that table to analyze the bandwidth usage
e.g.: when presenting data, for every readout I need to compute the
difference between it and the one immediately preceding it, and divide
that by respective measurement interval.

Initially I've put an additional column with a serial into the readouts
table, and did a join on (p.serial = n.serial+1); but that had an
occasional glitch, when serial actually skipped a value. So I'm trying
to work out a more resiliant/general solution. So far to no avail.

Is there an "sql-idiom" (receipt?) to do such join? the better if
without the spurious seiral column.

Thenx


-R

On Jan 26, 2013, at 13:32, Rafał Pietrak <rafal@zorro.isa-geek.com> wrote:

> I have a usage recording table: CREATE TABLE readings(tm timestamp, bytesin int, bytesout int);
>
> The readouts are made "occasionally" - the timespan between the readouts are not very precise, but there is a lot of
thosereadouts. 
>
> when presenting data, for every readout I need to compute the difference between it and the one immediately preceding
it,and divide that by respective measurement interval. 

You don't need a self-join, you need a window function.
See: http://www.postgresql.org/docs/9.1/static/tutorial-window.html

For example:
SELECT timestamp, lag(timestamp) OVER (ORDER BY timestamp) FROM readings;

W dniu 01/26/2013 02:49 PM, Alban Hertroys pisze:
> On Jan 26, 2013, at 13:32, Rafał Pietrak <rafal@zorro.isa-geek.com> wrote:
>
>> I have a usage recording table: CREATE TABLE readings(tm timestamp, bytesin int, bytesout int);
>>
>> The readouts are made "occasionally" - the timespan between the readouts are not very precise, but there is a lot of
thosereadouts. 
>>
>> when presenting data, for every readout I need to compute the difference between it and the one immediately
precedingit, and divide that by respective measurement interval. 
> You don't need a self-join, you need a window function.
> See: http://www.postgresql.org/docs/9.1/static/tutorial-window.html
>
> For example:
> SELECT timestamp, lag(timestamp) OVER (ORDER BY timestamp) FROM readings;
>

Yes. That's what I needed (another example, of how fragmented knowledge
of a subject makes one (myself) use quite inapropriate keywords when
searching).

Thenx,


-R