关于Prometheus中label_replace函数及对应告警规则的功能咨询

Hey there! Let me break down this Prometheus alert rule for you clearly, no jargon overload promise 😊

First, let's look at when this alert will trigger — it fires if either of these two conditions is met:

• In the last 5 minutes, the number of changes to the hikaricp_connections_timeout_total metric is greater than 2
• The hikaricp_connections metric doesn't exist at all (absent)

Now let's unpack each part one by one:

1. The trigger conditions explained #

• changes(hikaricp_connections_timeout_total[5m]) > 2

  • hikaricp_connections_timeout_total is a counter metric that tracks the total number of connection timeouts in your HikariCP connection pool (commonly used in Java applications).
  • The changes() function counts how many times this counter's value has increased (since each timeout increments the counter). So this condition means "there have been more than 2 connection timeouts in the last 5 minutes" — a clear sign your app is struggling to establish or maintain database connections.

• absent(hikaricp_connections) == 1

  • hikaricp_connections tracks the current number of active connections in the pool.
  • The absent() function returns 1 only if the metric has no data at all. This condition triggers when Prometheus isn't receiving any connection pool metrics from the target instance — which could mean the app is down, the monitoring agent is broken, or the connection pool itself has crashed.

2. What the label_replace part does #

The label_replace(..., "ip", "$1", "instance", "([^:]+)(:[0-9]+)?") part is just adding a practical label to the alert:

• It takes the existing instance label (which usually looks like 192.168.0.10:8080 or just a raw IP)
• Uses a regex to extract just the IP address (ignoring the port if one exists)
• Creates a new label called ip with that extracted value

This makes your alerts cleaner — instead of seeing a full instance string with a port, you just get the IP, which makes it way faster to identify which server or app is having issues.

To sum it up: This alert watches for two critical HikariCP connection pool problems — frequent timeouts (which hurt app performance) and complete loss of metrics (which signals a potential app/pool outage) — and adds an IP label to simplify troubleshooting.

备注：内容来源于stack exchange，提问作者Андрей