Weird icmpping behavior from zabbix-proxies (v1.6)

[Sawn]

Hi,

We have a setup constitued of one zabbix-server, five proxies and two MySQL servers (master/slave in replication).
All (server and proxies) are Zabbix (from sources) v1.6 under Debian Etch x86.

Everything was working fine since last september.

Since a few days, we have noticed a weird behavior from one of our proxy.

This proxy is monitoring ~150 Hyper-V VM, there are about thirty hosts being declared as "down" after an icmpping check.

Problem is : those hosts are not down.

Here is some tests we performed from the concerned proxy :

- ping -c1000 from the proxy to one of the "false positive" IP address : no packet lost.
- fping from the proxy to the same host : all ok
- traceroute from the proxy to the host : ok
- moving monitored hosts from this proxy to another : problem repeat itself on the other proxy

Here is what we saw from zabbix_proxy.log (debug=4 with a tail -f and grep -A3 on the hostname) :

                        "host":"hostname.domainname",
                        "key":"icmpping",
                        "clock":1247129052,
                        "value":"0"},
--
                        "host":"hostname.domainname",
                        "key":"icmpping",
                        "clock":1247129082,
                        "value":"0"},
--
                        "host":"hostname.domainname",
                        "key":"tcp,3389",
                        "clock":1247129103,
                        "value":"1"},
--
                        "host":"hostname.domainname",
                        "key":"icmpping",
                        "clock":1247129112,
                        "value":"0"},
--
                        "host":"hostname.domainname",
                        "key":"icmpping",
                        "clock":1247129142,
                        "value":"0"},
--
                        "host":"hostname.domainname",
                        "key":"tcp,3389",
                        "clock":1247129163,
                        "value":"1"},

And here is a fping from the same proxy :

# fping -c5 hostname.domainname
hostname.domainname : [0], 84 bytes, 0.96 ms (0.96 avg, 0% loss)
hostname.domainname : [1], 84 bytes, 0.75 ms (0.85 avg, 0% loss)
hostname.domainname : [2], 84 bytes, 0.93 ms (0.88 avg, 0% loss)
hostname.domainname : [3], 84 bytes, 0.59 ms (0.80 avg, 0% loss)
hostname.domainname : [4], 84 bytes, 0.76 ms (0.79 avg, 0% loss)

hostname.domainname : xmt/rcv/%loss = 5/5/0%, min/avg/max = 0.59/0.79/0.96

Any idea ?

Thanks

[froztbyte]

Not quite the same with me, but also some strange behaviour with monitoring via proxies. I've got hosts that I monitor in networks that I have very high latency to (>1500ms latency) and quite a few hosts start to indicate packetloss to varying degrees in the ZABBIX interface on my master server, but when I log into my proxy and run checks from there I don't get the same problem. This leads me to believe that ZABBIX is somehow making decisions about the data it's getting in via the proxies, perhaps substituting a value of '0' if it doesn't get all the data in during the sync period. This doesn't happen to all hosts, though. Anyone maybe seen this happening and perhaps have a solution?

[Sawn]

More informations :

When I set the problematic hosts to be monitored directly by the Zabbix Master (server on which zabbix_server runs), the icmpping status is correct.

But if i set them back on the proxy, after a few minutes, they are seen as down again.

[Sawn]

We found the problem, it had nothing to do with Zabbix.
I've a script which synchronise an SQL Server database with Zabbix (add/delete hosts & templates) but some ip addresses in the SQL Server database contain space after the IP.
Those spaces where replicated into Zabbix causing the hosts to be seen as down