On Sat, 02 Jun 2007 00:14:28 +0200, Ron Johnson <ron.l.johnson@cox.net>
wrote:
> On 06/01/07 16:38, PFC wrote:
>>
>>>>> Will the synchronized seq scan patch be able to do this by
>>>>> issuing all the CREATE INDEX commands at the same time from several
>>>>> different database connections ?
>>>> No, but it could someday.
>> Actually I tested, it does it right now, albeit unconsciously (pg
>> doesn't do anything to synchronize the scans, but if you launch the
>> concurrent connections at the same time and issue all your "create
>> index" at the same time, only 1 table scan is needed). Maybe if the
>> tables were bigger, it would lose sync between the 3 concurrent scans
>> and would end up going slower. That's why I spoke about the
>> "synchronized scan" patch.
>
> How much of this, though, is from the OS's disk cache? Or are Seq Scans
> O_DIRECT and bypass the OS cache?
Well, the file was larger than disk cache, and I checked in vmstat's
number of actual bytes read from disks...
Three threads read the table once, One thread reads the table 3 times.
So it works right now, except it doesn't have (yet) the infrastructure to
keep the scans synchronized, and psql can't open several connections (yet).