Netcup Prometheus alerting for unreachable Alertmanagers

[jchristgit]

We need to configure our Alertmanager to send us alerts on Discord such that we can be informed of anything not being right as part of the monitoring setup on lovelace.

[jb3]

As discussed in the dev-ops channel, I think we can reach a configuration here that utilizes our existing High-Availability AlertManager setup.

We can set up token access for Prometheus on Ansible machines to push alerts through to the Kubernetes HA AlertManager.

Some notes:

• This does not mean we cannot route alerts differently, push to different areas based on different severity or anything like this, we still maintain full granular control over alert routing, even more so since we centralise it.
• We can write a small healthcheck (systemd timer, cronjob, etc.) to check whether the alertmanager server is healthy and responding to requests, if it is not then we can trigger a rudimentary alert from netcup reporting that the AlertManager is down. I don't think this needs to be a separate instance of AlertManager as that feels overcomplex and leads to duplication of routing configuration.

[jchristgit]

Just to clarify this from a discussion on Discord, this is about adding a "dead man's switch" alert that will route to Discord in case the Netcup Prometheus instance can't contact the Alertmanager in Kubernetes properly. To cover this case we want to:

• add alerts in Prometheus in case it cannot talk to Alertmanager properly (there are built-in metrics for this exported by Prometheus)
• add a local alertmanager configured to send alerts to Discord
• configure the local alertmanager & Prometheus instances such that only the newly added alerts from above are routed to the local alertmanager and everything else is still routed to the Kubernetes alertmanager.