ZABBIX Forums - Blogs

Out for a bit.

cstackpole — Tue, 17 Jun 2008 15:54:50 GMT

I know I told a bunch of people that I would help out with some of their issues and I know that I have got several projects that I was wanting to do or already working on for the Zabbix community, but I am afraid those plans are on hold for now.

The short of it is a friend of mine was in an accident and past away. I am just now getting online and letting people know why I have been out of touch this past weekend.

I really do want to help the Zabbix community and I do want to get more involved...but this isn't the time. I am sorry if I have left someone I was helping in an awkward position; as soon as I get some things worked out I will be back.

Take care everyone.

Zabbix HA

nelsonab — Tue, 10 Jun 2008 18:31:35 GMT

High Availability Zabbix is something a lot of people have wanted. Recently my boss came to me and wanted to see if we can do it. I believe it can be hacked together to work.

I'm hoping I can hack together some scripts to work with Linux-HA and STONITH. Here's the setup.

I will have two Zabbix servers, let's call them Zabbix-A and Zabbix-B. The database will also sit on another machine, likely in a Master-Master setup.

Zabbix-A will always be the first primary. When A is running B will monitor. If B detects that A is down it will then power off A and startup it's own Zabbix server. After it has come up it will then restart A. Zabbix-A will then check to see if Zabbix is running on B, if so then it will start to monitor Zabbix on B. If it fails on B, A will power down B and the above process will be repeated.

How will I monitor the health of Zabbix? This is a tricky one. I'm thinking the standby node will have to send values to the active node and then check to see if they make it to the database through a direct query. I'm also thinking I will also perform some Zabbix_get operations and perhaps some SNMP trap information too.

Either way this will be an interesting journey.

Nagios, zabbix and stderr

Shmuma — Sun, 01 Jun 2008 03:27:17 GMT

Hello, all

zabbix is much more powerfull than nagios. But during migration, we sometimes find some missed features which implemented in nagios. One of them -- ability to attach text message to item result. It's quite usefull sometimes.

For example:
You want to monitor linux software raid status. It returns some numerical value (0/1 or amount of failed disks or something similiar). But on agent's side, this script have more information about of system's status. For example, it knows array and disk names that failed. In zabbix you have no way to pass this additional information to server, but it would be quite good to have a way to use it (in SMS or mail message).

Situation is even worse, when your check must handle possibly unlimited amount of server's events. One of such issues is kernel messages. There is almost no noise in dmesg output. You always must react in some way when kernel talks to you. But kernel have almost unlimited number of different messages. Of course, you can write item which will tell you number of new lines in dmesg output, but this will force you to go to server to see what happened.

This issue also makes migration from nagios to zabbix quite hard if you have large amount of user items.

In past few days I tried to solve this problem. I extended agent->server protocol to transfer not only key and value but text written by UserParameter command to stderr. The amount of text is limited to 256 characters. On server side, this text string (if any) stored in item's table and displayed in 'Last values' and 'Configure->Items' interfaces. This value can also be used as text macro {ITEM.STDERR} in action mesages.

The patch is attached to this post, and based against our patched zabbix version (which in turn based on 1.4.4), so it may not apply cleanly. But it is quite enougth to get the idea what happend inside. If you really interested in this feature, drop me a note, I'll port this patch to mainstream version.

wbr, Shmuma.

Attached Files

stderr-in-items.patch (32.5 KB, 31 views)

Not able to find php-gd package for rhel 4

Gajendra — Tue, 27 May 2008 04:06:12 GMT

Hi,
I am not able to find php-gd package for redhat enterprise linux 4.
Can any one tell me where can i find the rpm or can u please mail me @ gajendra2k2@gmail.com

Agent Forum post

nelsonab — Tue, 22 Apr 2008 07:54:20 GMT

This blog is only here as a placeholder to the forum post for the Ruby agent.

http://www.zabbix.com/forum/showthread.php?p=32923

Ruby Agent

nelsonab — Mon, 21 Apr 2008 23:48:12 GMT

Work on the packet monitoring Ruby agent is coming along nicely. I've managed to work out the dynamic update issues I was having and now I'm testing for stability. I still need to add a few other filter types. Currently it only supports IP address and Port which means it a filter for a host on port 25 will capture both TCP and UDP traffic. Why you would run both on the same port is a mystery to me. I need to add a specific filter for TCP, UDP, general IP and perhaps MAC address. Yes, IP6 would also be a good idea, but I think I'll leave that for version 2.

I hope to be releasing this to the general community soon.

It works!!!

nelsonab — Thu, 17 Apr 2008 07:12:19 GMT

The basic packet capture agent works!! So far it can handle only one filter being added dynamically, but I hope to work on that tomorrow. However for now I'm one step closer to bringing packet flow data into zabbix.

Ruby, Zabbix and Packet Capture

nelsonab — Wed, 16 Apr 2008 18:44:59 GMT

I started work on a new contract and the first thing they had me do was learn Ruby. Ok, not bad. However as I learned it more and began to see how I could use it for some Zabbix stuff. So I wrote a simple demo of a Zabbix Agent and am now working on integrating it with a slightly more complex Ruby app, a packet capture agent for Zabbix.

The purpose of the agent is to monitor network traffic at the packet level. Say you want to know how much data is going to one server on a certain port with this agent you can do this. So far I've written a small app in Ruby which sets up some Pcap filters and when packets come in matching the Pcap filters a set of counters are updated. Every so often the counter information is displayed to the screen.

Right now I'm working on integrating it into a Zabbix agent. Here's how I envision it will work, we'll see if this is how it will really work in the end however. ;) When the agent starts it will have no filters enabled. Filters are added when the Zabbix Server contacts the agent. The agent will then check it's local filter list, if it does not exist one will be created. If it does exist the pertinent information will be returned.

Pretty easy right? HA! My biggest question is can I dynamically add filters to a running Pcap session? The next question is do I want to process the data into pkts/sec and bytes/sec and return it to the server or do I want to send it back to the server "as is"? Well I guess it's kinda a moot point as I can just set that in the key argument.

Either way I hope to keep updating as time progresses. If anyone is interested in testing, or may wish to see a feature please let me know.

Agent data buffers

Shmuma — Sat, 05 Apr 2008 20:43:59 GMT

Hi!

Some time ago I implemented buffers of data collected by zabbix_agent via active checks. A bit later, similiar functionality appeared in development version of zabbix (1.5 at the moment). It's pleasant to me that my guesses about functionality missed in zabbix agreed with zabbix developers' views.

But these two implementations differ in one very important topic -- where to store historical data.

My implementation uses disk buffers to store active checks' data that we failed to send to server. Beside the data, agent store list of currently configured active checks. If it failed to get list of checks from server, it takes them from buffer.

Version 1.5 of zabbix performs these in a different manner. It uses memory buffers which haven't preserved between agent restarts.

I have strong belief that such buffers must allow agent to work independently of server for some (maybe quite long) time. Of course we won't have triggers and events there -- only history. But this history must be carefully collected and safely stored for later analisys. What can cause agent<->server communication failures? A lot of things:

Minor network errors. In large networks, some packets can be lost and this is ok.
Major network errors. Physical cable failure, central router failure, etc.
Power failure.
Overheating.
Zabbix server hardware or software problems.

Buffers in memory do their job in first kind of problems (most frequent, of course). But they mostly useless when some of Bad Things from the rest of list happens, because the servers with zabbix_agent can be rebooted and right after reboot we lost all history data collected from the start of failure. This is even worse -- agent won't start data collection after the reboot. Server is up and running, but if agent can not reach zabbix_server, it doesn't collect data -- it trying to obtain the list of checks. The more servers you have, the higher the probability that some of them rebooted and history have lost.

Such bad situation are rare, of course, but they are possible. And value of information about what happened with servers during these failures is much higher than some piece of disk space (100Mb disk buffer allows an average server to collect history during a week :)) and some little overhead of writing data to disk.

Of course, all above is actual for monitoring in large systems (hundreeds of servers in separate locations, 24/7, 99.999%, bla-bla-bla :)) and almost useless when you have two servers on top of your desk. But if zabbix is planning to grow, we should consider both use cases. It can be easily implemented both ways with configuration file option where to store buffer (in memory or disk) and how large it must be. But my Jira ticket with patch simply closed with comment 'The functionality is implemented differently in 1.5.x.'. Cool.

zabbix

divyavij — Wed, 12 Mar 2008 14:28:31 GMT

i also wanted to knw how is zabbix an advantage over nagios and cacti. wa is the advantage n disdvatage

how to monitor switches in zabbix

divyavij — Wed, 12 Mar 2008 14:26:50 GMT

hi friends, can u please tellme hw to we monitor switches using zabbix. we cant istall th agents rgt ? so wat is the exact procedure can u pls tell me:)

About users, rights and left join

Shmuma — Sun, 02 Mar 2008 14:58:57 GMT

Yesterday, we discovered and started fixing one very unpleasant bug in zabbix rights management subsystem. If we grant some group of users right to read some hosts group, and other group of users have some rights on the same group, first group sometimes can gain no permission at all.

Quick search on zabbix forum have shoun that this drawback is here during long time and even have partial fixes.

During whole day we hacked around this and make things a bit cleaner, but still have no 100% solution. I hope that a bit more work on this can help.

So, the details follows. It's known that zabbix rights management is quite flexible, and this is great. We have users, which are joined to user groups. From other side we have hosts which can belong to one or more hosts groups. These two kind of groups can be joined by rights (deny, read or read-write). This can be shown by this nice ER diagram:

The central part of this rights management system is a routine get_accessible_hosts_by_user. This function gets user_id and compare permission function (on this later) and returns list of hosts accessible by the user. The heart of this routine is the following query:

Code:

select distinct h.hostid,h.host, min(r.permission) as permission,ug.userid 
from hosts h left join hosts_groups hg on hg.hostid=h.hostid 
left join groups g on g.groupid=hg.groupid 
left join rights r on r.id=g.groupid and r.type=1
left join users_groups ug on ug.usrgrpid=r.groupid and ug.userid=2
group by h.hostid,h.host,ug.userid
order by h.host, permission desc, userid desc

I simplified this query a little -- in original code there are clustered monitoring stuff, which isn't neccesarry at the moment.

So, this query does quite simple thing: it takes all our hosts and left joins other tables to get permissions associated with user, passed in arguments.

Simple configuration works well:

User with ID=2 (guest) have granted right to read group, consisted by one host (fry.homelinux.ru).

But if we make our configuration a bit more complex by granting RW-access to this group to Admin user, the picture changes dramatically:

Here we have two lines with fry.homenet.ru hosts. Routine get_accessible_hosts_by_user iterates over result of query and stores host's permissions in a hash using host name as a key (checking for new hash key existence before inserting). So second line of data is blindly ignored. As a result, user guest have no permission to this host. Bad.

It's trivial that this row mustn't be in a result -- it related to different user. And it appears there because of left join condition ug.userid=2. We can filter such occurences, moving this condition out of join to global where clause, but we'll lost hosts we hadn't specified permissions (there are default policy, related to nodes in cluster config).

To my mind, this can be resolved only by splitting this query into two.

The above query also have another issue with min(r.permission) as permission. If we have one host, belonging to two different groups adn some user have separate rights on these groups, it gain only minimal rights. It seems to me that this is incorrect, because we have no way to extend user's rights on some hosts if we prohibited that hosts via some other group. It needs more experminets, but I think that min shoud be replaced with max.

Intro

Shmuma — Fri, 29 Feb 2008 22:06:31 GMT

Hello all,

I'm starting this blog to keep thougths, investigations and plans about our company migration from nagios to zabbix. I don't think that telling company name here is a good idea (NDA and other nice things, yes). But amount of servers in our infrastructure is quite large (5000 at the moment, 10k planned to the end of this year), so I think that this experience can be quite useful to others.

I have no separate plans for these posts, so I will write here from time to time. And my English is bad, sorry.

New Zabbix templates!

qix — Fri, 29 Feb 2008 15:52:35 GMT

I've been uploading some Cisco and Dell templates to the wiki lately.
See the template gallery thread or the wiki.

These include:

Cisco Catalyst C3750-48TS switch stack
Cisco PIX 515E with 6 Fastethernet Interfaces
Cisco VPN3000 concentrator series with 3 NIC's
Dell Powerconnect 5224 switch
Dell Powerconnect 5324 switch

Please expect the following still to come:

Cisco Catalyst 2924M-XL switch
Cisco Catalyst 2980 Switch
Cisco Catalyst 3524-XL Switch
Cisco Catalyst 3548-XL Switch
Cisco Catalyst 3550-48 Switch
Cisco Catalyst 4003 Switch
And more...

I'm not a programmer, so I cannot contribute directly to the code, but perhaps this way we can contribute back to the community :p

Back to the office

Alexei — Wed, 27 Feb 2008 14:50:05 GMT

I am glad that I visited the Fosdem 2008 event even despite the fact I had to wake up at 4:30 a.m. in order to be in Riga airport in time for the flight to Brussels.

It was worth it: I had a chance to meet members of the Open Source community and to listen to the lighting talks, 15 minutes long presentations on different topics, software-related mostly. Kees Jan Koster has made a very good job: his presentation (the reason why I was there) about Zapcat (monitoring of Java by Zapcat and ZABBIX) was excellent and he earned a bunch of applauses.

The whole event was a demonstration of the strong spirit of Open Source.

Unfortunately I missed PHP6 talk, which I was very interested in. However I had an excellent opportunity to discuss some technical topics at the dinner we had later.

I�ve made some photos at the meeting � find them attached.

Attached Images

Attached Files