Call for discussion on ZABBIX 1.6

[helfrez]

More Robust SNMP support

The main thing lacking for me is more robust snmp support. I like the way cacti handles snmp, but it is very limited in scope of capability.

- SNMP base setting tied to template rather than items (port, community, version)

- Basic templates for graphing data from standard MIBs

- Easier handling of scaling graphical data values.

I don't know how hard it would be, but possibly even handing off some graphing functions to rrd. RRD does a awesome job of creating simple and scalable graphs. (long term)

[Marie]

Refresh rate should be configurable on a per screen basis : when a screen contains several items that are already refreshed, it doesn't need to refresh.

I hope this makes sense

[x0tsing]

Better win32 logging and reports

We desperately need to be able to capture Event ID's from Windows event log. It's theoretically possible to get it with the source, severity, and a text match of the description, but that's too much work to be scalable.

It would be nice to mask certain triggers with other triggers for a period of time. For example, every time I see a log entry that Windows Update has run, I know the server will be reboot shortly and don't need to be notified of it.

Customizable reporting. I would like to receive all alerts in realtime, but e-mail management a list of currently active triggers at the end of the day. I would also like to receive an e-mail with less severe triggers at the end of the day (or week).

"All" should be listed in the "host" drop down wherever a Group dropdown exists. If I am able to choose a group, I would like to be able to see the data from all members of that group without switching between each one.

[dunbart]

Logging

Alexei,

Perhaps it is already implemented in ZABBIX 1.4. If so, you can ignore this post. We are using 1.1.6 and it would be very helpfull that the zabbix_server.log could tell us the IP address of the host that has an "unsupported item" configured. We are monitoring about 1500 hosts, and using the bulkloader it is difficult to find that specific SNMP-host that has to be adjusted.

[akschu]

Alexei,

A couple of thoughts on improving zabbix.

Short version:

1. Find a way to group up actions or events so that the admin doesn't get 100 separate pages to the pager when 1 key interface fails.

2. More macro support, specifically {LOCATION} and other profile elements so that the location of a host can be placed inside of an email action.

3. Less database load. Find a way to group queries.

4. Actions that take 6-8 queries can probably gain lots of performance with database procedures.


Long version:

1. If I'm reading the code right we look for a trigger on every insert of the data causing several queries for every new recorded element. This takes a lot of time so perhaps if we backed this off to a process that runs every minute then the query load would be reduced but then we could also group of the triggers into a single action. In other words the look for tripped triggers would only happen every minute, and the action would handle every trigger for that minute.

2. Several of the items in the host profile are not available as macros for actions. It would be nice to be able to use some of this information in a notification email.

3. This is like the first suggestion. It seems the do everything every time approach causes zabbix to have insane query load with lots of elements.

4. I don't know how far along mysql proc support is, and I know most people use mysql, however my own testing (and many new benchmarks) show that psql 8.2.x on linux 2.6.20.x outperforms mysql 4.x and 5.x even in raw query volume. Because I'mat using a database system that has very good procedure support, many of the dabase operations that require several queries can be grouped into a single procedure which would really improve the performance. That said i realize that database independence is important and that it would make a lot of database independent code, however it may also simplify things as zabbix would call DB functions and leave the db specific code inside the DB. I'm not meaning to draw a mysql vs psql comparison, but my personal opinion is that using database systems that have good procedure support and using procedures to cut down the query load is the best way to add more scalability to the product as the database will be MUCH faster at managing it's data.

Thanks,
schu

[Alexei]

Alexei,

A couple of thoughts on improving zabbix.

Short version:

1. Find a way to group up actions or events so that the admin doesn't get 100 separate pages to the pager when 1 key interface fails.

2. More macro support, specifically {LOCATION} and other profile elements so that the location of a host can be placed inside of an email action.

3. Less database load. Find a way to group queries.

4. Actions that take 6-8 queries can probably gain lots of performance with database procedures.

1,3,4 are already under development. I hope to increase ZABBIX performance 5x-10x times using a database cache module.

2. These macros are already supported in 1.4.x. {PROFILE.LOCATION}, ...

[MartinZapa]

Cisco boxes templates

Dear Alexei,
zabbix is very usefull tool for monitoring our network. A weak point for me is tricky configurations of Cisco boxes. I dont want zabbix equate to other products but for example cacti have a nice templates with easy interfaces octets,bcst,errors and more configuration. So if you think about this it will be great.
Second concept from me is about map. If you have link backup or for example ATM ima interface with two or more E1 lines and first created of them going down in map is connection line all along green because lines are crisscross.
Best regards,
Martin

[Zapnix]

Hello,

(1) if there's a way to stabilize the DB interface (use ODBC?) for some more databases (e.g. Oracle) would be nice, and
(2) while setup (PHP frontend), letting the user choose the DB connection type earlier would make a better requirement check possible (had a PHP with only Oracle support, but a Zabbix using sqlite3, finding out that php-sqlite3 is needed took me some time) - with a unified connection mechanism this would completly be obsolete.

[treyh]

My Thoughts

Suggestions:

The scability of the UI would get very ugly if you had more than 100 hosts with 25 items being monitored.

Add support for negative numbers


Compliments:

Great forum and feedback from the community and developers!

[mcarbonneaux]

first it's very greet fine product... thank!

1. pub/prive ssl key authentification and encryption between server/agent
   use simply openssl to do that.
   i think is important feature to add to zabbix, to be abel to work on hostile environement, without hint like vpn.
2. screen template like graph template.
   simply host variabilisation it's cool feature to add.
3. mass update in "host"
   to be abel to modify host port , or group, template of group of host
4. database creation integrate in web install
   why in web install interface they not create directly the database schema?
5. in "add host" snmp discovery action add the posibility to specify how to set hostname
   • like from reverse dns of the ip
   • from returned discovery value (regexp or full value?!)
   • from formula (based on ip!?)
   • from external script called
   • ...
   
   auto set dns name when reverse dns name is ok
   and to define the default port of the host in the add host action.
6. embed php and/or perl interpreter, to used in item get in addition of zabbix agent/internal, snmp, external script...
   to use also in differente location with complex formula...
   to used also in discovery/trigger action...
   
7. tendancy based trigget
   like if filesystem "/" has > 3% per hours and <30% free
   
8. no need to update screen who use graph when modify graph
   
   bug??!!!
9. posibility to include external php script or web page in screen box
   for exemple with php include.
   or iframe
   
10. more posibility to work on the project (cvs/svn acces)
    not only with patch submit...
    i'm interessed to work with you on this project...
    
11. more communication with other developper
    
    if open acces to the développements...
    
12. clear strategie of update (like linux kernel and many big other open source project)
    
    for exemple : stable realease with only correction (defining one person to maintain this branch and accept patch) (use of bug tracker like bugzilla)
    and dev release open to modification (also open to other developper)
13. php class to acces to the data model...
    to developpe external intergration
    
14. plugable ui authentification
    to add the possibility to integrate the ui in enterprise sso or ldap
15. plugable poller as so/dll
    to add custom poller: like ldap poller, different sgbd poller...
    
16. mini snmp browser to permit to select element in target device
    in add item to smplify item creation
17. snmptable support (in mini snmp browser also) and/or walk get
    to get multiple data in one defined item.
    it's very important feature beceause, when i have loadbalancer style of device... if i need to monitor all Virtual IP and it back end server status, actualy is very hard to manage... if i'have like 100 vip per device to transforme in 3 or more item per vip... it's large work to add in zabbix...
    it's important to be abel to add item and associate trigger...
    
18. item template trigger or generic trigger
    item template trigger: when i add item, automaticly add trigger
    generic trigger: ability to add one trigger generic for many item simply on linking item to this trigger (ex: the alarm is fired with the name of the item in the trigger label). all item must me of the same type.
19. more import/export possibility
    - import/export screen
    - batch import/export (not from web, to automate from crontab device intergration)
    
    batch import can made with simplified version of exp_imp.php who can be used from shell.
    

many more but steel enougth for the moment...
Best regards,
Mathieu

[yvesd]

I would like to say simply, support for AIX...

The agent is not working (collecting values of 0), and the zabbix_sender is not working too...

I had to recreate my parameters with external script in perl and injecting vias socket the XML request to server...

The rest is fine...

[azilber]

Leaner network resource consumption.

I'm still running zabbix 1.1.4, so this may or may not be an issue with later versions.

But, we have only 4 external servers that monitored for standard stuff.

Looking at network utilization graphs, we are using 24% of our T1 (outgoing).
While only 5% incoming for the zabbix port.

If requests could be serialized toogether and maybe the protocol optimised that would be significant in my book.

[mcarbonneaux]

[*] snmptable support (in mini snmp browser also) and/or walk get
to get multiple data in one defined item.
it's very important feature beceause, when i have loadbalancer style of device... if i need to monitor all Virtual IP and it back end server status, actualy is very hard to manage... if i'have like 100 vip per device to transforme in 3 or more item per vip... it's large work to add in zabbix...
it's important to be abel to add item and associate trigger...

in cacti they use xml file to define index polling like this...

<interface>
        <name>Alteon Group Actives sessions</name>
        <description>get Alteon Actives sessions for Group</description>
        <oid_index>.1.3.6.1.4.1.1872.2.5.4.2.3.1.1</oid_index>
        <index_order>ifName:ifIndex</index_order>
        <index_order_type>numeric</index_order_type>
        <index_title_format>|chosen_order_field|</index_title_format>

        <fields>
                <ifIndex>
                        <name>Index</name>
                        <method>walk</method>
                        <source>value</source>
                        <direction>input</direction>
                        <oid>.1.3.6.1.4.1.1872.2.5.4.2.3.1.1</oid>
                </ifIndex>

                <ifName>
                        <name>Group Name</name>
                        <method>walk</method>
                        <source>value</source>
                        <direction>input</direction>
                        <oid>.1.3.6.1.4.1.1872.2.5.4.1.1.3.4.1.10</oid>
                </ifName>

                <currSess>
                        <name>Actives Sessions</name>
                        <method>get</method>
                        <source>value</source>
                        <direction>output</direction>
                        <oid>.1.3.6.1.4.1.1872.2.5.4.2.3.1.2</oid>
                </currSess>
        </fields>
</interface>

i think can be in database in zabbix...

to poll the complete list of the Altheon VIP like that without to configure manualy...

http://www.cacti.net/downloads/docs/...query_xml.html

and the same for external script...
http://www.cacti.net/downloads/docs/...query_xml.html

in that way when on this equipement i add new VIP they are atomaticly add in monitoring...

best regards,
Mathieu

[mcarbonneaux]

[*] embed php and/or perl interpreter, to used in item get in addition of zabbix agent/internal, snmp, external script...
to use also in differente location with complex formula...
to used also in discovery/trigger action...

on cacti:
http://www.cacti.net/downloads/docs/...pt_server.html

some link how to embede php un c programm...

http://circle.ch/blog/p786.html

http://www.the-asw.com/post/2005/05/...rogramme-c-php (french...sorry!)

http://www.sklar.com/page/article/myphp

http://books.slashdot.org/article.pl?sid=04/06/22/233250

the Sara Golemon book!

http://safari.oreilly.com/067232704X

and excelent tutorial from sara: http://www.cluesheet.com/lectures/phptek-embedding.pdf



the idea are the compilation of php with --enable-embed sapi
in sapi/embed of php source tree... generate libphp that can be used un c programme...that make possible to call php code from zabbix server for exemple to poll item...

sample of code:

#include <php_embed.h>
#include <stdio.h>
int main (int argc, char *argv[])
{
PHP_EMBED_START_BLOCK(argc, argv)
zval retval;
zend_eval_string(“return 1+1;”, &retval,
“Code returning a value” TSRMLS_CC);
convert_to_string(&retval);
printf(“The script code returned: %s\n”, Z_STRVAL(retval));
zval_dtor(&retval);
PHP_EMBED_END_BLOCK()
}

the same with perl...

http://circle.ch/blog/p788.html

http://search.cpan.org/dist/perl/pod/perlembed.pod

[azilber]

Also, connection optimization..

Zabbix tends to open 1 connection per item. Which is crazy. Here's the connection list, monitoring 4 servers:

I'm still running zabbix 1.1.4, so this may or may not be an issue with later versions.

But, we have only 4 external servers that monitored for standard stuff.

Looking at network utilization graphs, we are using 24% of our T1 (outgoing).
While only 5% incoming for the zabbix port.

If requests could be serialized toogether and maybe the protocol optimised that would be significant in my book.