Zabbix v1.1.x Broken with FreeBSD 5.x and 6.x

[dcrandall]

Hi,

I'm starting a new thread to highlight the major problems with Zabbix v1.1.x running on FreeBSD. My goals are to bring this to a higher profile, but also to get any feedback from other users who are using Zabbix on FreeBSD.

The biggest problems are:

* v1.1.x zabbix_server: When there is a network error or a connection to a client host times out, the clients metrics all say [no data] - which is expected. The problem is that when the client host comes back, zabbix server never updates that the host is available, and so never starts monitoring it again. Furthermore, zabbix_server has the dead client listed as status: monitored and Availability: available the whole time regardless of whether the client is up and running, dead, or timing out for some reason.
The only way to make zabbix_server start monitoring the host again is to massage the database manually by doing this:
mysql -u root -D zabbix -e 'update hosts set errors_from=0'
- This means that if a host client goes into [no data] and then crashes, you will not recieve any alerts from Zabbix.

* v1.1.x zabbix_agentd will build without errors, but will not start. Stack tracing with truss or strace do not produce any output.
- I have had to use v1.1beta9+ agents on FreeBSD.

* v1.1.x zabbix_server builds without errors, but will not start if you are have debug=4 set. I have provided stack traces of this behavior and can again.

As a stab in the dark, I would consider migrating my zabbix_server to Linux if it looked like that would solve some of the server issues.

Let me state that I think Zabbix is an awesome monitoring platform, this post is not meant as a complaint. It's meant to be constructive and help solve these problems.
Also, before you say it, yes I realize it's open source, and yes I am presenting the option of purchasing commercial Zabbix support to my boss - and we may end up doing that.

Thanks for everyones time,

Daniel

[GateKeeper]

http://www.zabbix.com/forum/showthread.php?t=4840

In addition to mentioned very big problem must say that agent does poorly support even those checks that are listed as 'supported' in manual: (ex. net.if.[in|out]). Can't see any reason they can't be supported, can't demand this too, but expect.

From now on I do not use zabbix because if memory leak. Other reasons are not decision-influencing (1.1.3 starts up and works fine both agent and server on my FreeBSD 6.1 box even I compiled it from sources manually, not from ports).

In opposition to thread-starter dcrandall, won't ever migrate to linux due to reasons exclusive to this thread.

[Alexei]

I'm aware of the reported issues and I'm very very interested in a fast resolution of all these problems.

* v1.1.x zabbix_server stop monitoring of unreachable hosts

There is no FreeBSD (or any other OS) specific code in logic of ZABBIX. It means that most of users must be affected. I do not see this. It makes me think that the problem is specific to your setup or, in some strange way, is related to specifics of FreeBSD.

When first network errors hits ZABBIX , it updated hosts.disable_until by 'now+UnavailableDelay', default is 15 seconds. There is no such thing as host status. The disable_until means that ZABBIX won't check the host until the disable_until time.

After UnreachablePeriod, ZABBIX updates the disable_until by 'now+UnreachableDelay', default value is 60 seconds.

May I see your zabbix_server.conf? (delete all sensitive info please).

* v1.1.x zabbix_agentd will build without errors, but will not start.
* v1.1.x zabbix_server builds without errors, but will not start if you are have debug=4 set.

These problem are very hard to debug without direct access to a FreeBSD box. I think that we add a FreeBSD box for our test environment very soon.

* memory leak

This is PostgreSQL-specific. Actually we already spent more than one day tracing is down. It is fixed in 1.1.5 (not released yet).

* more support of freebsd stats

Again, it depends on user demand and availability of FreeBSD server.

Having said this, I'm looking forward to having all reported issues fixed, they all recorded and won't be lost.

[GateKeeper]

Alexei, am I right, thinking that you need just a FreeBSD host with full access granted to you on it?

If so, when I have more time - I'll try to bring it up for you. Of course, if you don't mind a jailed shell.

[Alexei]

Yes, I need a FreeBSD host (I believe jailed environment is fine) with SSH access, a MySQL database, GCC/make and development libs. I would appreciate it very much! Please use my email address a l e x @ z a b b i x . c o m for private communication.

It would be excellent if we could add this host to our automated testing environment. This would improve support of FreeBSD very much.

[Alexei]

All right, we managed to install ZABBIX Agent under FreeBSD 6.1 running under MS Virtual PC and it compiles and runs absolutely fine regardless of debug level.

[dcrandall]

First - It's not the zabbix agent that has problems with debug level, it's the server.

We've had problems getting the agent to run at all on FreeBSD 5 or 6. I know I'm not the only one.

Just now I downloaded the zabbix-1.1.4 tarball, extracted, configured with --enable-agent

tried to start the agent with a valid very minimal config, and it does not start. This is what I get from strace -f....

strace -f /usr/local/bin/zabbix_agentd -p
execve(0xbfbfe810, [0xbfbfed00], [/* 0 vars */]PIOCWSTOP: Input/output error

strace -f /usr/local/etc/rc.d/zabbix_agent start
execve(0xbfbfe810, [0xbfbfecf8], [/* 0 vars */]PIOCWSTOP: Input/output error
-bash-2.05b# Starting zabbix_agentd.

Platform details:
FreeBSD ned 6.1-RELEASE FreeBSD 6.1-RELEASE #0: Sun May 7 04:42:56 UTC 2006 [email protected]:/usr/obj/usr/src/sys/SMP i386

[dcrandall]

I consider the zabbix_server problems to be the bigger concern here anyway.

The problem where server starts putting [no data] for an agents metrics when the agent times out momentarily, then does not pick the agent back up when it's back is huge.

Every time I see a host has [no data] for it's metrics, I check it with zabbix_get on the server side and zabbix_agentd -p on the client side, and both return real values. Only when I execute the command
mysql -u root -D zabbix -e 'update hosts set errors_from=0'
Can I get monitoring back for that host.

I've described the problem over and over again.
I don't really care that I have to use zabbix-1.1beta9 for the zabbix_agentd. It's annoying for sure, but more than that I want the server to behave correctly.

Daniel

[Alexei]

I'm aware of the reported FreeBSD issues. As soon I get FreeBSD MySQL environment ready I will test all of them!

[juswan]

I've tested and confirmed the issues with the agent on Freebsd 6.1. All the beta versions work up until 1.1beta12. So maybe that will give someone a clue about where to look. Very much appreciate any help we can get.

Justin

[nullpt]

Hello,

I have zabbix server and agentd installed in a FreeBSD 6.2 box and it's returning some weird values.

For example, it returns a value of 2 for "Email (SMTP) servers running" but i don't have any e-mail software running and I don't have anything listening on 25/tcp .

Also, I have hundreds of MB's free memory but zabbix reports only 500KB of free memory.

It returns nothing in the free swap space field, total swap size, hoststatus, network stats, Maximum number of processes, etc..

It seems not to work so well under freebsd

Any other freebsd users with this glitches?

Also, when I create a graph it shows the graphic, the min/max/latest values but there is no graphic line drawed. There are no errors in apache errorlog. I can't find out why this is happening :x

Thanks in advance for the help
Cheers

[d_jabsd]

Another Zabbix 1.1.6 problem with FreeBSD 6.2-

The snmp checks are corrupting the udp checksum in v1 and v2c.

15:31:26.144799 IP (tos 0x0, ttl 64, id 33744, offset 0, flags [none], proto: UDP (17), length: 73) xx.xx.xx.xx.60266 > yy.yy.yy.yy.snmp: [bad udp cksum ed43!] { SNMPv1 { GetRequest(30) R=799925457 interfaces.ifTable.ifEntry.ifInOctets.1 } }

15:32:22.195431 IP (tos 0x0, ttl 64, id 33962, offset 0, flags [none], proto: UDP (17), length: 73) xx.xx.xx.xx.64246 > yy.yy.yy.yy.snmp: [bad udp cksum 6132!] { SNMPv2c { GetRequest(30) R=799925458 interfaces.ifTable.ifEntry.ifInOctets.1 } }


a normal snmpget or snmpwalk from the commandline gives [udp sum ok] so its not the card or net-snmp installation.

[ALikD]

Hello,

It returns nothing in the free swap space field, total swap size, hoststatus, network stats, Maximum number of processes, etc..

It seems not to work so well under freebsd

Any other freebsd users with this glitches?

Also, when I create a graph it shows the graphic, the min/max/latest values but there is no graphic line drawed. There are no errors in apache errorlog. I can't find out why this is happening :x

Thanks in advance for the help
Cheers

Same problems. Zabbix_agentd installed on a FreeBSD 7.0 box.
There are Keys from Template_FreeBSD which has ZBX_NOTSUPPORTED Status:
- vm.memory.size[buffers]
- vm.memory.size[cached]
- vfs.file.cksum[/vmlinuz]
- system.cpu.util[,wait,avg1]
- system.swap.size[] (all system.swap.size doesn't work)
- system.boottime
- net.if.in[em0,bytes] (em0 - specific for my server)
- net.if.in[em1,bytes] (em1 - specific for my server)
- net.if.in[lo,bytes]
- net.if.out[em0,bytes]
- net.if.out[em1,bytes]
- net.if.out[lo,bytes]
- proc.num[] (all proc.num checks doesn't work)

Also, I have hundreds of MB's free memory but zabbix reports only 500KB of free memory.

+1. Free memory check (vm.memory.size[free]) returns wrong value.