Zabbix milestone achieved - 1,000 hosts monitored

[walterheck]

what's inaccurate about graphing speed per second from questions ?
mysql qps, yes, that's not really useful for long running mysql installations.

Sorry, for some reason I didn't see this reply before. My bad

Questions is increased for a whole lot more then just queries. SET, use, show, desc statements and even syntax errors increase it as well.
Syntax errors shouldn't be too much of a problem on a production machine, but SET and things like desc can be executed quite a lot depending on the application (a lot of frameworks do nasty stuff for every execute query to get metadata and such).
Try looking at the amount of questions that get asked when you say 'use zabbix;' (on a machine that has no other connections using mysql obviously):

mysql> show global status like 'questions';
+---------------+-------+
| Variable_name | Value |
+---------------+-------+
| Questions     | 53    |
+---------------+-------+
1 row in set (0.00 sec)

mysql> use zabbix;
Reading table information for completion of table and column names
You can turn off this feature to get a quicker startup with -A

Database changed
mysql> show global status like 'questions';
+---------------+-------+
| Variable_name | Value |
+---------------+-------+
| Questions     | 146   |
+---------------+-------+
1 row in set (0.00 sec)

mysql>

The better way would be to count the specific com_% statusses and the questions that get asked to the query cache.

[richlv]

ah, right. yeah, questions would also include pretty much everything sent to the mysql - although i haven't seen a huge impact from non-query questions usually, i can imagine that some applications do have a lot of them.

on the other hand, as far as i recall, com ones were very detailed, so one might miss some important actions... while not an issue with zabbix database usually (selects/inserts/deletes/updates, and that's pretty much it during normal operation), maybe that would not work that great with some other application.

not sure whether there is a decent solution - maybe by gathering both 4 most often used types of questions + total questions as well ?

although this is departing from the original topic subject and might have to go into a separate thread

[walterheck]

ah, right. yeah, questions would also include pretty much everything sent to the mysql - although i haven't seen a huge impact from non-query questions usually, i can imagine that some applications do have a lot of them.

Load-wise the non-query questions are indeed usually very lightweight. Which potentially skews your idea of how many queries your database is actually handling.

on the other hand, as far as i recall, com ones were very detailed, so one might miss some important actions... while not an issue with zabbix database usually (selects/inserts/deletes/updates, and that's pretty much it during normal operation), maybe that would not work that great with some other application.

not sure whether there is a decent solution - maybe by gathering both 4 most often used types of questions + total questions as well ?

although this is departing from the original topic subject and might have to go into a separate thread

It doesn't _really_ matter which figure you look at, as long as you know what it means. What is now queries per second is really "commands fired at the database per second". If you agree on that convention and that is what you always look at when looking at servers, that is totally fine.

imho it should be a combination of the query cache variables and some of the com_ variables added together.

[cronuscronus]

Our deployment

We're using Zabbix to monitor transmitters. We run a nationwide network of them here.

Currently we have ~2200 hosts monitoring 27,000 items and we're pushing about 1150 new values per second.

We're using an Oracle RAC for the database processing and have the actual data collection local. We have 500 proxies and the damn thing works!

[dalle]

Today I enabled host number 1,000

1 Zabbix App server (RHEL - HP DL360 G6, DQC 2.40Ghz , 8GB RAM)
1 Zabbix DB server (RHEL, MySql - HP DL380 G6, DQC 2.53Ghz, 24GB RAM, SAN storage)
13 Zabbix proxies (RHEL, MySql - HP DL360 G6, DQC 2.40Ghz, 8GB RAM)
1 Zabbix Administrator

Hosts span OS's of Win, AIX, HP-UX, Linux and Solaris
Zabbix version 1.8.2
All items are passive

Made possible through talented teammates and Zabbix support.

on our environment we have:

Number of hosts 1572
Number of items 62303
Number of triggers 11062
Number of users (online) 79
Required server performance, new values per second 171.48

[tzn]

One question there: how do you get MySQL CPU Time value?

[xsbr]

Our updated status:
3771 hosts
314698 items
163266 triggers
2707 vps (values per second)

http://twitpic.com/4zaui9

OS: CentOS 5.6 64bits

Database: MySQL Percona 5.5
* InnoDB with a file per table
* Housekeeper disabled
* Table daily partitioning on history tables
* Table monthly partitioning on trend tables
* Created an index by clock column on some tables

1 Database Server:
* Dell R610 Poweredge
* 2x L5520 CPUs
* 24GB RAM
* 6x HDs 300GB SAS 10K (RAID 10)

1 Zabbix Server:
* Dell R610 Poweredge
* 2x L5520 CPUs
* 24GB RAM
* 6x HDs 72GB SAS 15K (RAID 10)

10 Zabbix Proxies:
* VM Xen 4 Cores
* 2GB RAM
* SQLite

1 Zabbix FrontEnd:
* VM Xen 8 Cores
* 4GB RAM

[bcarpio]

Proxies vs DM

Its interesting to see people primarly using proxies vs distributed monitoring, is there a reason for this? What about links going down between sites?

[xcalibur]

I must ask what do you use as an availability trigger for each and every of those hosts. How do you know if they're down or not?
Myself cant seem to get a proper trigger that wont false positive and vise verso :\

[jix]

!

Good job
is it possible for you to tell me
what solution chosen for mysql backup?