No single point of failure configuration

[little_martian]

Hi,
I installed 6.0.1 version with 2 zabbix servers configured in HA mode, 2 MySQL DB servers with replication, and a proxy server (for the moment).

I am trying some configurations and have some problems. My scope is to be able to store metrics in several failure scenarios: 1 db server down, 1 zabbix server down, proxy server down.
The zabbix servers config relevant to this:

ZBX1 server (version 6.0.1)
ListenPort=11051
HANodeName=Primary node, NodeAddress=zbx1:11051

ZBX2 server (version 6.0.1)
ListenPort=11051
HANodeName=Backup node, NodeAddress=zbx2:11051

PRX server (version 6.0.1)
ProxyMode=0
Server=zbx1:11051;zbx2:11051
ListenPort=11051

Agent (version 3.2.0)
Server=prx,zbx1,zbx2
ServerActive=prx:11051,zbx1:11051,zbx2:11051

In this configuration, my scope was for the agent to send metrics to the proxy first and if the proxy is down, send metrics to zabbix servers (the one that is listening in the HA configuration)

The proxy log contains these kind of messages (anonymized):
"cannot send list of active checks to "x.y.z.t": host [monitored_host] not found"

I checked in the logs with debug level and found these two queries:

select h.hostid,h.status,h.tls_accept,h.tls_issuer,h.tls_ subject,h.tls_psk_identity,a.host_metadata,a.liste n_ip,a.listen_dns,a.listen_port,a.flags from hosts h left join autoreg_host a on a.proxy_hostid is null and a.host=h.host where h.host='monitored_host' and h.status in (0,1) and h.flags<>2 and h.proxy_hostid is null

insert into proxy_autoreg_host (clock,host,listen_ip,listen_dns,listen_port,tls_a ccepted,host_metadata,flags) values (1647006413,'monitored_host','x.y.z.t','monitored_ host',11050,1,'',0)

So I checked in the proxy database and I found the host is present in the first query periodically and the dissapears. It seems the monitored host is registered with the proxy and the de-registered (probably due to the Server/ServerActive parameter in the agent configuration).

My question is: how do I achieve no single point of failure between monitored host and zabbix database? Is the layout described previously sufficient ? How should I configure the agent, proxy and server to achieve my goal ?

Thank you.

[cyber]

There is no such way as you want, "first send to proxy and if proxy fails, then to server". In Zabbix your host is configured with assignment to a proxy. So no other proxy or server will accept data for that host. They have no knowledge of that host.
So there is no point to configure that agent to point to proxy and both servers.

Your 3.2 agent does not understand the concept of HA on server side.

ServerActive=zabbix.cluster.node1;zabbix.cluster.n ode2:20051;zabbix.cluster.node3
3.2 won't understand this, probably fails to start (note the ; instead of ,) . So you cannot even send to the "one that is listening in the HA configuration"...

What you want, may come with version 6.2. Roadmap says "Proxies to provide automatic load balancing and HA" https://support.zabbix.com/browse/ZBXNEXT-5911 top voted feature.. Until that comes, you need to build that cluster and HA with the help of other software... Corosync/pacemaker for example.

[little_martian]

Thank you for the answer. I knew that version 3.2 would be no help in this architecture. I was hoping that with agent version 6, that supports ServerActive configuration with HA nodes, I could configure the Proxy as a HA node on the agent and would behave as I want.
Just to be clear, this has to do active checks only, but, because when the server receives data from the proxy it automatically marks the host as monitored by that proxy, no passive checks work either ?

[cyber]

Your host it tied to a proxy in its config (Monitored by proxy...). All items, active or passive, will use that proxy.

[little_martian]

Thank you, I upvoted ZBXNEXT-5911 also, it seems like what I want to achieve.