Zabbix agent on server is unreachable for 5 minutes

[ilia]

i had many much more of them, then i set the params:

StartPollers=1000
StartPreprocessors=10
StartPollersUnreachable=1000
StartTrappers=100
StartPingers=1000
CacheSize=32M

now i have much less but sill got a approx 20-30 radom unreachable events a day.
any idea what i should do ?

[ingus.vilnis]

Setting max values for all data gathering processes is a bad idea, particularly if you can live with as little as 32MB cache, meaning you have a very small environment.

Start by lowering the number of started processes so they are about 30-40% busy. Same goes for internal processes for which you have not posted any graphs and settings. Actually I'd recommend to go with default values on all settings and then gradually increasing the values till performance is optimal.

[ilia]

i thought of it and set to

Pollers=100
StartIPMIPollers=10
StartPreprocessors=100
StartPollersUnreachable=100
StartTrappers=100
StarStarttPingers=100
StartDiscoverers=100
CacheSize=64M
HistoryCacheSize=32M
HistoryIndexCacheSize=16M
TrendCacheSize=16M
ValueCacheSize=32M
Timeout=10


the system have 4 cores and 8 GB ram (approx 1.5GB ram is used)

and it was stable for few hours
and then few Unreachable events occured

[Mayday]

Please kindly check if the clock is synchronized for agent servers. thanks.

[ilia]

I did, all computers have a synced ntp clock.
i have upgraded from 4.0 to 4.2 few hours ago let hope it will be enough

[ingus.vilnis]

Upgrade alone will be no help if you have such high values for started processes set in the configuration file.

In Template App Zabbix Server there are five graphs that very well illustrate the health of Zabbix server. based on the reading all settings in the config file can be adjusted. You can show them all here to have more insight along with all currently configured values in the config file.

[ilia]

since upgrade it seems good,
maybe the upgrade did some sort of flush or clean some values

here are the charts: first with the load upon start, and second excluding startup load

[ingus.vilnis]

When you receive the "false" alerts, are there really gaps in the collected values from those hosts? Compare one such time when you receive the alert with graphs from that host.

[ilia]

check the chart on my 3rd post, when i have a spike at 12:20 i received a lot of alerts,
ut before i had received them all the time, usually at large chunks

never i had a gap in the data, but sometimes i had a relative spike (most of the time under the 80% mark)

[ingus.vilnis]

12:20 Pollers busy + UnreachablePollers busy = Network problem where Zabbix can't connect to devices it monitors.

Not sure what spikes you mention but go to Monitoring -> Latest data or Graphs for any device which had this agent unreachable alert and carefully look if you have all the values with exact timestamps collected at the problem period.

[ilia]

just a note, i had a successful ping and a working ssh connection while zabbix gave the error
i never had internal issues when this occurs, cause in most time it occurs on hosts that are idle,

i cannot find something special

[ingus.vilnis]

Special would be missing data in for example CPU utilization graph for a host that had the false alert.

Look in Zabbix server log file for that period and analyze if there are any entries about the hosts with alerts.

[ilia]

i tried, no gaps or error, next error ill dive deeper into it, cause now i enabled some external monitoring