1.8.X vs Solaris vs SNMP

[ilikejam]

Has anyone managed to get Zabbix server running on Solaris to talk to SNMP devices? I'm running Zabbix 1.8.5 (also tried 1.8.4) compiled against various versions of Net-SNMP on Solaris 10 (sparc), but I can't get it to work. No matter what I do, the SNMP community string never gets sent/set with the SNMP GET.

I've switched on debug on the target system's snmpd...

Here's the snmpd log from a command line snmpget (which works OK and returns a sensible answer):

Received 43 bytes from 10.27.4.153
0000: 30 29 02 01  01 04 06 70  75 62 6C 69  63 A0 1C 02    0).....public�..
0016: 04 75 71 04  8E 02 01 00  02 01 00 30  0E 30 0C 06    .uq........0.0..
0032: 08 2B 06 01  02 01 01 01  00 05 00                    .+.........

Received SNMP packet(s) from 10.27.4.153
  GET message
    -- SNMPv2-MIB::sysDescr.0

Sending 89 bytes to 10.27.4.153
0000: 30 57 02 01  01 04 06 70  75 62 6C 69  63 A2 4A 02    0W.....public.J.
0016: 04 75 71 04  8E 02 01 00  02 01 00 30  3C 30 3A 06    .uq........0<0:.
0032: 08 2B 06 01  02 01 01 01  00 04 2E 53  75 6E 4F 53    .+.........SunOS
0048: 20 68 6E 75  6E 78 6A 65  74 30 31 73  20 35 2E 31     hnunxjet01s 5.1
0064: 30 20 47 65  6E 65 72 69  63 5F 31 34  31 34 31 34    0 Generic_141414
0080: 2D 30 39 20  73 75 6E 34  75                          -09 sun4u

And here's what the zabbix snmp probe looks like in the logs:

Received 37 bytes from 10.27.4.153
0000: 30 23 02 01  01 04 00 A0  1C 02 04 5A  A3 33 EF 02    0#.....�...Z.3�.
0016: 01 00 02 01  00 30 0E 30  0C 06 08 2B  06 01 02 01    .....0.0...+....
0032: 01 05 00 05  00                                       .....

Received SNMP packet(s) from 10.27.4.153
  GET message
    -- SNMPv2-MIB::sysName.0

Note the lack of 'public' in the request. The SNMP community is definitely set on the Item, both on the web interface and in the Zabbix DB.

Anyone else seeing this?

Cheers.

Dave

[untergeek]

Yes, I had this exact same problem. This was one of the biggest reasons we migrated from using Solaris as the Zabbix server to RedHat EL5.

The second reason is that Zabbix is optimized for high-speed single-thread models. The deep thread (but slower individual thread speed) model of Sun hardware was not fast enough to keep up with the load we were seeing.

Best of luck. We couldn't make Solaris work for us. If you have an NVPS of less than 200, you'll probably be okay.

[ilikejam]

Gah! Don't tell me these things!

I'm sneaking Zabbix in so I don't have a budget for x86 kit - all I've got is a rescued 2-core V240 for the server. It's not doing too badly so far: 1934 items monitored, 60 values per second. There are items in the queue for >30seconds most of the time, but nothing's getting dropped.

I can't really see why SNMP would behave differently between Solaris/Sparc and Linux/x86, though.

[untergeek]

Bait the hook. That's all it took for us. Once they started seeing the benefit they were absolutely dependent on Zabbix, like it was cocaine or something. They bent over backwards to get me x86 hardware for Zabbix.

I would log a bug for with Zabbix over Solaris SNMP. Now that you've pinpointed the fact that no credentials are being sent they may be able to fix it. I never went that far in my debugging.

What DB is on your backend? Is it local or remote? I/O could be in your favor, if things are local instead of going out remotely.

[ilikejam]

SNMP is part of the hook-baiting

The DB is on the local host, with the table storage on its own ZFS mirror on the 3rd and 4th internal disks. Haven't really had a chance to look at optimising MySQL yet - just using the pretty much out-the-box MyISAM config from the GlassFish WebStack download (which Oracle have, of course, stopped providing now - all the more reason to move to Linux).

I'll log a bug and see where it goes.

[ilikejam]

Raised a bug:

https://support.zabbix.com/browse/ZBX-3737

[ilikejam]

Probably a subject that needs a thread of its own, but I've got your attention for a few moments, so...

Since you've managed to introduce Zabbix to your environment, have you got any tips for things that Zabbix can do that the Management types really go for?

We've got an incumbent BMC Patrol monitoring system that's Installed And Paid For, but which all of the admins - Windows and Unix alike - hate (with occasionally frightening fervour, I have to add). I'd like to do some kind of Zabbix presentation for the Wintel admins and the PHBs, so I'm looking for stuff that would decisively shoot down any opposition and bring everyone round to our way of monitoring.

The Zabbix dashboard on recent versions is pretty convincing on its own (compared with BMC at least) for the PHBs, but I have a feeling I'll have to dig pretty deep to convince the people who control the contracts that we should move to a better monitoring system.

Any thoughts?

[untergeek]

Screens & Graphs
Escalations
History
Extensibility

One of the things that continues to wow the management is our ability to add monitoring for things which were previously not monitor-able. Finding a few of those and beginning to monitor them, and if they're the kind that can be trended, graph them. These kinds of things are the ones which continue to impress at my company.

Also, auto-discovery is a huge deal for us. If it can be implemented in your setup, the ability to remove human error from your monitoring is a huge plus.