Обсуждение: PG slowdown

Поиск
Список
Период
Сортировка

PG slowdown

От
"Abu Mushayeed"
Дата:
Hello,

My hardware

CPU: Intel 3 GHZ X 4
Memory: 9 GB
LINUX

My conf file are as follows:

shared_buffers=10000
wal_buffers=128
vacuum_mem = 256000
sort_mem=32768
max_fsm_pages = 3000000
max_fsm_relations = 32768
checkpoint_segments = 2048
checkpoint_timeout = 1800
checkpoint_warning = 300
effective_cache_size =  262144
random_page_cost = 2
default_statistics_target = 250

Some of my queries that used to run within a minute, before I changed the
"default_statistics_target" from 10 to 250, now runs for 20-30 minutes. Some
queries did see improvement in their run time.

I have two questions:

1. Why did the other queries slowdown? I tried to tune them, as their is a
join between two tables and it reads all the rows, I forced it to do a seq
scan rather than index scan but it quadrupled it run time. Why?
2. How can I disable the logging of updates? What are the pitfalls? When an
update or insert happening there are lot of logging going on and the system
writes messages like this "2006-07-11 10:08:54 [2330] LOG:  checkpoints are
occurring too frequently (182 seconds apart)
HINT:  Consider increasing the configuration parameter
"checkpoint_segments"."

Thank you for the help.

Abu



Re: PG slowdown

От
adey
Дата:
If you turn WAL logging off (via fsynch parameter in .conf), it is my understanding that you will no longer log database updates, and will therefore not be able to recover any uncommitted transactions in the event of failure, (which seems contradictory to database practice).
 
Gut feel to me says your chckpoint_segments and timeout are too high.

 
On 7/12/06, Abu Mushayeed <amushayeed@hotmail.com> wrote:
Hello,

My hardware

CPU: Intel 3 GHZ X 4
Memory: 9 GB
LINUX

My conf file are as follows:

shared_buffers=10000
wal_buffers=128
vacuum_mem = 256000
sort_mem=32768
max_fsm_pages = 3000000
max_fsm_relations = 32768
checkpoint_segments = 2048
checkpoint_timeout = 1800
checkpoint_warning = 300
effective_cache_size =  262144
random_page_cost = 2
default_statistics_target = 250

Some of my queries that used to run within a minute, before I changed the
"default_statistics_target" from 10 to 250, now runs for 20-30 minutes. Some
queries did see improvement in their run time.

I have two questions:

1. Why did the other queries slowdown? I tried to tune them, as their is a
join between two tables and it reads all the rows, I forced it to do a seq
scan rather than index scan but it quadrupled it run time. Why?
2. How can I disable the logging of updates? What are the pitfalls? When an
update or insert happening there are lot of logging going on and the system
writes messages like this "2006-07-11 10:08:54 [2330] LOG:  checkpoints are
occurring too frequently (182 seconds apart)
HINT:  Consider increasing the configuration parameter
"checkpoint_segments"."

Thank you for the help.

Abu



---------------------------(end of broadcast)---------------------------
TIP 4: Have you searched our list archives?

               http://archives.postgresql.org

Re: PG slowdown

От
"Joshua D. Drake"
Дата:
adey wrote:
> If you turn WAL logging off (via fsynch parameter in .conf), it is my
> understanding that you will no longer log database updates, and will
> therefore not be able to recover any uncommitted transactions in the
> event of failure, (which seems contradictory to database practice).
>

You can not turn of WAL logging. You can only delay how the data is
written to disk.

Sincerely,

Joshua D. Drake



--

    === The PostgreSQL Company: Command Prompt, Inc. ===
Sales/Support: +1.503.667.4564 || 24x7/Emergency: +1.800.492.2240
    Providing the most comprehensive  PostgreSQL solutions since 1997
              http://www.commandprompt.com/