Re: Simple (hopefully) throughput question?

Поиск
Список
Период
Сортировка
От Samuel Gendler
Тема Re: Simple (hopefully) throughput question?
Дата
Msg-id AANLkTine9evHx1tTGvZXvA0wTMmarYgbtD=t0AiAP7gS@mail.gmail.com
обсуждение исходный текст
Ответ на Simple (hopefully) throughput question?  (Nick Matheson <Nick.D.Matheson@noaa.gov>)
Список pgsql-performance


On Fri, Nov 5, 2010 at 12:23 PM, Samuel Gendler <sgendler@ideasculptor.com> wrote:
On Thu, Nov 4, 2010 at 8:07 AM, Vitalii Tymchyshyn <tivv00@gmail.com> wrote:
04.11.10 16:31, Nick Matheson написав(ла):

Heikki-

Try COPY, ie. "COPY bulk_performance.counts TO STDOUT BINARY".

Thanks for the suggestion. A preliminary test shows an improvement closer to our expected 35 MB/s.

Are you familiar with any Java libraries for decoding the COPY format? The spec is clear and we could clearly write our own, but figured I would ask. ;)
JDBC driver has some COPY support, but I don't remember details. You'd better ask in JDBC list.



The JDBC driver support works fine.  You can pass a Reader or InputStream (if I recall correctly, the InputStream path is more efficient.  Or maybe the Reader path was buggy.  Regardless, I wound up using an InputStream in the driver which I then wrap in a Reader in order to get it line-by-line.

You can write a COPY statement to send standard CSV format - take a look at the postgres docs for the COPY statement to see the full syntax.  I then have a subclass of BufferedReader which parses each line of CSV and does something interesting with it.  I've had it working very reliably for many months now, processing about 500 million rows per day (I'm actually COPYing out, rather than in, but the concept is the same, rgardless - my outputstream is wrapper in a writer, which reformats data on the fly).



I should mention that I found basically no documentation of the copy api in the jdbc driver in 8.4.  I have no idea if that has changed with 9.x. I had to figure it out by reading the source code.  Fortunately, it is very simple:

return ((PGConnection) con).getCopyAPI().copyIn(sql, this.fis);


Where this.fis is an InputStream. There's an alternative copyIn implementation that takes a Reader instead.  I'm sure the copyOut methods are the same.


Note: my earlier email was confusing.  copyIn, copies into the db and receives an InputStream that will deliver data when it is read.  copyOut copies data from the db and receives an OutputStream which will receive the data.  I inverted those in my earlier email.


You can look at the source code to the CopyAPI to learn more about the mechanism.

В списке pgsql-performance по дате отправления:

Предыдущее
От: Greg Smith
Дата:
Сообщение: Re: Major Linux performance regression; shouldn't we be worried about RHEL6?
Следующее
От: Samuel Gendler
Дата:
Сообщение: Re: Simple (hopefully) throughput question?