> On 23 March 2011 16:36, Jeff Janes <
jeff.janes@gmail.com> wrote:
>>
>> On Wed, Mar 23, 2011 at 6:19 AM, Jochen Erwied
>> <
jochen@pgsql-performance.erwied.eu> wrote:
>> > Wednesday, March 23, 2011, 1:51:31 PM you wrote:
>> >
>> > [rearranged for quoting]
>> >
>> >> background writer stats
>> >> checkpoints_timed | checkpoints_req | buffers_checkpoint |
>> >> buffers_clean |
>> >> maxwritten_clean | buffers_backend | buffers_alloc
>> >>
>> >> -------------------+-----------------+--------------------+---------------+------------------+-----------------+---------------
>> >> 3 | 0 | 99754 |
>> >> 0
>> >> | 0 | 115307 | 246173
>> >> (1 row)
>> >
>> > buffers_clean = 0 ?!
>> >
>> >> But I don't understand how postgres is unable to fetch a free buffer.
>> >> Does any body have an idea?
>> >
>> > Somehow looks like the bgwriter is completely disabled. How are the
>> > relevant settings in your postgresql.conf?
>>
>> I suspect the work load is entirely bulk inserts, and is using a
>> Buffer Access Strategy. By design, bulk inserts generally write out
>> their own buffers.
>>
>> Cheers,
>>
>> Jeff
>
> Yes. that's true. We are converting databases from one schema into another
> with a lot of computing in between.
> But most of the written data is accessed soon for other conversions.
> OK. That sounds very simple and thus trustable ;).