Обсуждение: High %SYS CPU usage
Hello there,
I have PostgreSQL 8.3.18 server running on Centos 6.2 (2.6.32-220.7.1) with this specs:
2x CPU AMD Opteron 6282
128GB RAM
Raid 10 (12HD 15k rpm 1GB cache) with data
Raid 10 (4HD 15k rpm 1GB cache) with xlog
Raid 1 (15k rpm 1GB cache shared with xlog) with system
On this server I have only one database with 312GB of data. The database had run fine during 4 months, but from two months ago, during high work load periods, the server is collapsed by "%sys" type load.
For example "dstat -ar --socket --tcp" during %sys load problem:
Reboot the server mitigates the problem during few days, but always reappear.
Server not is swapping, don't have excessive I/O, don't have %IRQ load.
I don't have any ideas...
Thank you very much for your help.
My sysctl and postgres.conf:
sysclt -a:
My Postgres.conf:
max_connections = 500 # (change requires restart)
unix_socket_directory = '/var/run/postgres' # (change requires restart)
shared_buffers = 18GB # min 128kB or max_connections*16kB
work_mem = 30MB # min 64kB
maintenance_work_mem = 1GB # min 1MB
max_fsm_pages = 8553600 # min max_fsm_relations*16, 6 bytes each
max_fsm_relations = 409000 # min 100, ~70 bytes each
fsync = on # turns forced synchronization on or off
synchronous_commit = off # immediate fsync at commit
wal_buffers = 8MB # min 32kB
checkpoint_segments = 64 # in logfile segments, min 1, 16MB each
checkpoint_completion_target = 0.9 # checkpoint target duration, 0.0 - 1.0
archive_mode = on # allows archiving to be done
archive_command = 'exit 0'
effective_cache_size = 100GB
constraint_exclusion = on
default_text_search_config = 'pg_catalog.spanish'
max_locks_per_transaction = 100
Cesar, > On this server I have only one database with 312GB of data. The database > had run fine during 4 months, but from two months ago, during high work > load periods, the server is collapsed by "%sys" type load. Hmmm. Have you updated Linux any time recently? I'm wondering if this is a PostgreSQL problem at all. It sounds like an OS issue. Can you give us the results of mpstat -P ALL 3 ? The dstat output doesn't tell me much. -- Josh Berkus PostgreSQL Experts Inc. http://pgexperts.com
Hi Josh,
--
César Martín Pérez
cmartinp@gmail.com
Today is not the worse day for this issue, because Fridays DB have little workload... But I have this output:
I don't have updated OS server from the install.
I also begin to think that the problem is OS misconfiguration (Kernel parameters??) or Hardware problem...
Thanks for your help.
2012/10/30 Josh Berkus <josh@agliodbs.com>
Cesar,Hmmm. Have you updated Linux any time recently? I'm wondering if this
> On this server I have only one database with 312GB of data. The database
> had run fine during 4 months, but from two months ago, during high work
> load periods, the server is collapsed by "%sys" type load.
is a PostgreSQL problem at all. It sounds like an OS issue.
Can you give us the results of mpstat -P ALL 3 ? The dstat output
doesn't tell me much.
--
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com
--
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance
César Martín Pérez
cmartinp@gmail.com
Finally I resolv the problem setting /sys/kernel/mm/redhat_transparent_hugepage/enabled to "never"
2012/11/2 Cesar Martin <cmartinp@gmail.com>
Hi Josh,Today is not the worse day for this issue, because Fridays DB have little workload... But I have this output:I don't have updated OS server from the install.I also begin to think that the problem is OS misconfiguration (Kernel parameters??) or Hardware problem...Thanks for your help.2012/10/30 Josh Berkus <josh@agliodbs.com>Cesar,Hmmm. Have you updated Linux any time recently? I'm wondering if this
> On this server I have only one database with 312GB of data. The database
> had run fine during 4 months, but from two months ago, during high work
> load periods, the server is collapsed by "%sys" type load.
is a PostgreSQL problem at all. It sounds like an OS issue.
Can you give us the results of mpstat -P ALL 3 ? The dstat output
doesn't tell me much.
--
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com
--
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance
César Martín Pérez
cmartinp@gmail.com