Ad Widget

Collapse

Help with lots of proxies

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • cronuscronus
    Junior Member
    • Jan 2010
    • 29

    #1

    Help with lots of proxies

    Hello. We have a unique setup. Our system has a lot of remote locations (~400) and each location has a small load, only about 1.25 values per second.

    The ZBX manual states you can have "unlimited proxies" and right now we have all ~400 proxies running with 255 trappers on our Zabbix server.

    We are also running 18 DB Syncers load balanced to an Oracle RAC.

    Oracle is handling the load just fine and the server has 4 processors and an average idle time of 70%.

    Our queue tends to back up though with our proxies. Using strace I can see that the trapper processes are working the hardest. So it looks like we have an adequate number of DB Syncers.

    I was wondering the best way to balance the load coming from our proxies. Can we start more than 255 trappers? Can we assign the trappers to listen to specific proxies? Should we have half of our proxies in passive mode so we can start 255 trappers and 255 proxy pollers?

    We currently send data from the proxies every 30 seconds.... has anybody noticed any good results with different timing configurations?

    Any information is helpful..

    Thanks!
  • xs-
    Senior Member
    Zabbix Certified Specialist
    • Dec 2007
    • 393

    #2
    This is an interesting question.

    We have a similar setup (not as much proxies tho), with lots of remote locations.

    A possible approach could be to setup a proxy per region, not per site. In my situation, i split up all sites into regions and picked the biggest site (or most stable one) to host the proxy and use that to monitor all sites surrounding that.
    This way you would make better use of each proxy and have a smaller monitoring infrastructure footprint to worry about.

    Back to the real question:
    Officially, the proxies require a 'Zabbix Trapper' thread on the main node. I have done some tests with this, and i'm not sure if this is really the case. During my tests i set the pollers to 0 (as the main node does not actively monitor anything) and the proxies stopped working. I think this was during 1.6, perhaps stuff has changed. You could perform some tests to check if a combination of pollers and trappers gives you a better result.

    Please post your results, i'm curious as to what your findings are. (we are scaling up proxies and i expect to run into similar problems).

    Comment

    Working...