How does polling works for interfaces in down state

[sthon]

I discover my interfaces of my network devices.
Depending on the device I have up to 200-300 interfaces per device.
In some cases most of them are in down or admin down states.
I poll different parameter per interface so my item list grows very fast up 1000 per device.

interfaces which are in state down or admin down must not be polled for interface counters.
anyone know if zabbix do it in that way, or is there a possibillity to implement this ?

thx in advance

[sthon]

No one has an idea ? :-(

[sthon]

still hot,
no answer means no idea, or it is not possible ?

[sthon]

any ideas if something has changed in Zabbix 3.0 or 3.2 ?

[kloczek]

Doesn't matter which one version of the zabbix LLD is working the same way.
To have what you want you cannot do this using only one LLD you need two of them
First LLD would be only with one prototype to collect states of the ports.
Second one with exactly the same LLD item but added filter to filter off interfaces in off or admin off state and this LLD should have prototypes with in/ot bytes metrics.

To be honest what you are trying to archive/implement is a bit pointless.
It is not obvious that in reality you are not saving to much on collecting data.
Why?
First of all if SNMP data are collected over bulk queries so reducing number of interfaces to check does not reduces number of SNMP queries. Zabbix proxy or server on query SNMP OIDs with the same history period is trying to put in single query as much as possible quered OIDs which should be queried at the same time.
Second: if you will be collecting not raw in/out bytes but speed per second in case of inactive interfaces will produce only metrics points with 0 bytes/s values.
All "sped" metrics should be using foat type of data. If you will have look on history table definition which is :

CREATE TABLE `history` (
        `itemid`                 bigint unsigned                           NOT NULL,
        `clock`                  integer         DEFAULT '0'               NOT NULL,
        `value`                  double(16,4)    DEFAULT '0.0000'          NOT NULL,
        `ns`                     integer         DEFAULT '0'               NOT NULL
);

you can see that every history row holding "0.000" value will hold only itemid and clock timestamp which is much less in total size when it contains non-zero value.

[sthon]

> First LLD would be only with one prototype to collect states of the ports.
> Second one with exactly the same LLD item but added filter to filter off
> interfaces in off or admin off state and this LLD should have prototypes
> with in/ot bytes metrics

that´s the Goal. thx for the tip.
In my implementation I can "save" a lot of items with this solution, cause I have a lot of Switches with around 5000 items and if Í only use the "up" Interfaces this items can be reduced to 1000 per host.

thx

[kloczek]

In my implementation I can "save" a lot of items with this solution, cause I have a lot of Switches with around 5000 items and if Í only use the "up" Interfaces this items can be reduced to 1000 per host.

Still you don't get it
If there is no new data for the item or value is default who cares how many iems is defined in the monitoring configuration?
Item definition is only metadata description and takes few tenths of bytes per item.
How many total items you have at the moment in your current zabbix setup?