reproducible for you (poller_interval)

[bfek-18]

Hallo,

after alterations of the cacti-settings,increased itself powerfully the CPU-Load and poller-runtime.
I found the settings value which produced this behavior.
It is the „poller_intervall“ field.

Here are my Testresults:

System: cacti-0.8.7b with fc8
Method:spine Processes:2 Threads:4 Hosts:715 HostsPerProcess:358 DataSources:14438 RRDsProcessed:8308


with poller_interval = 300:
echo "insert into settings (name,value) value ('poller_interval',300);"| mysql -u cacti -p -D cacti

poller.php runtime : 50-60 seconds
CPU Load : 15 % (System 5%, User 10%)


without poller_interval:
echo "delete from settings where name='poller_interval';"| mysql -u cacti -p -D cacti

poller.php runtime : 90 – 110 seconds
CPU Load : 45 % (System 15%, User 30%)


Is that a problem of my cacti installation ?
Can the problem be reproduced with you?
Simply delete the poller_interval from the Database table settings and test it.


Thank you for your help.
Uwe

[bfek-18]

Hi,

the source of the problem I could find.

The high timeout value of the spine process are due to two mysql queries and query6 query7 caused.
You can find both of these queries in the source code of spine (poller.c).

query6="UPDATE poller_item SET rrd_next_step=rrd_next_step-300 WHERE host_id=<HOST-ID>"
query7="UPDATE poller_item SET rrd_next_step=rrd_step-300 WHERE rrd_next_step < 0 and host_id=<HOST-ID>"

Why these two queries are exactly I am not entirely clear.
They are, each host up to 5 seconds duration.

The delete of poller_interval prevented the call from query6/query7. Was not my intention.
And the burden (duration / CPU) were much smaller.
This worked well until version cact-spine-0.8.7e.

From version cacti-spine-0.8.7g were the times bad again.

Together with TheWitness I had tried to find the cause. http://forums.cacti.net/viewtopic.php?t=38045

For the version cacti-spine-0.8.7g I only have the following dirty patch.

bye
UK

[gandalf]

Please

Code: Select all

show index from poller_item;

to find out, which indexes are defined to your poller_item table
R.

[bfek-18]

Please

Code: Select all

show index from poller_item;

to find out, which indexes are defined to your poller_item table
R.

Hi,

here are the output.

mysql> show index from poller_item;
+-------------+------------+---------------+--------------+---------------+-----------+-------------+----------+--------+------+------------+---------+
| Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment |
+-------------+------------+---------------+--------------+---------------+-----------+-------------+----------+--------+------+------------+---------+
| poller_item | 0 | PRIMARY | 1 | local_data_id | A | NULL | NULL | NULL | | BTREE | |
| poller_item | 0 | PRIMARY | 2 | rrd_name | A | 21798 | NULL | NULL | | BTREE | |
| poller_item | 1 | local_data_id | 1 | local_data_id | A | NULL | NULL | NULL | | BTREE | |
| poller_item | 1 | rrd_next_step | 1 | rrd_next_step | A | NULL | NULL | NULL | | BTREE | |
| poller_item | 1 | action | 1 | action | A | NULL | NULL | NULL | | BTREE | |
+-------------+------------+---------------+--------------+---------------+-----------+-------------+----------+--------+------+------------+---------+

The long runtime is to be observed only under poller.php / spine.
If I run two sql commands by hand, it does not take that long.

thx

[TheWitness]

How big is this system? How about a quick top screenshot and a

Code: Select all

du -hs rra

[bfek-18]

How big is this system? How about a quick top screenshot and a

Code: Select all

du -hs rra

Code: Select all

2.5G

[TheWitness]

Well you provided half the information i requested anyway. Make sure your rrdfiles are on separate physical disk from the database.

TheWitness

[bfek-18]

Well you provided half the information i requested anyway. Make sure your rrdfiles are on separate physical disk from the database.

TheWitness

They are on different hard drives.

Code: Select all

mysql/OS Centos5.5 Raid1 2x149GB/15K  ext3

cacti/rra          Raid1 2x149GB/15k  ext4 (defaults,noatime,nodiratime,data=writeback,commit=120)

UK

[TheWitness]

Hmm. Do the following:

Code: Select all

#!/bin/bash
tables=`mysql -uroot -p<password> -e "show tables in cacti" | grep -v Table`
for table in $tables;do
  mysql -uroot -p<password> -e "analyze table $table"
done

Do that once a week....

TheWitness