How to structure triggers with clusters & dynamic inventory

[sfr]

Hi

Let's say I have a MySQL cluster of around 50 amazon instances that dynamically scale up and down depending on load.

I want to have a trigger in zabbix that checks on a specific return value from a MySQL query on the cluster. Could be getting a

COUNT()

from a table, or checking the status of cluster synchronization (which is done via something like

show global status where Variable_name='wsrep_evs_repl_latency'

)

So I add this trigger to my MySQL Cluster Server Zabbix template.

But what I don't want is for all of the 50 servers to generate 50 duplicate pager calls and email alerts whenever this trigger fires – 1 is enough.

What I've done so far is simply creating the trigger directly on a random MySQL node, but then the trigger obviously stops working when auto scaling removes that node/server.

So where do I put my trigger?

Since Zabbix is 100% host-based, I guess I could put it on the Zabbix Server itself, but of course I also autoscale my zabbix environment due to load, and then I have the same problem.

Any pointers on this?

[sfr]

bumpety bump

[shuum]

Have you come up with solutions?

[sfr]

I just tried searching for this again in 2023, and only found my own post.

Zabbix has zero support for clustered systems, still.

[seren]

I believe you can do what you are/were trying to do using aggregations (https://www.zabbix.com/documentation...ated/aggregate).

So each MySQL host would have an item such as `mysql.cluster.latency`. Then you create a pseudo host (ex. "MySQL-Cluster") with a single trigger like (assuming the servers are in a host group "MySQL-Servers" and you have tags per environment).

Some made up examples:

This takes the max value of the reported cluster latency (which should be the same across all nodes):

max(last_foreach(/*/mysql.cluster.latency?[group="MySQL-Servers" and tag="env:production"])) > {$MYSQL.CLUSTER.MAXLATENCY}

This checks that all running servers are in the cluster:

min(last_foreach(/*/mysql.cluster.membercount?[group="MySQL-Servers" and tag="env:production"]​)) < sum(last_foreach(/*/agent.ping?[group="MySQL-Servers" and tag="env:production]))