Alerts

Since the control plane has access to real-time PgDog metrics, it can detect outliers and trigger automatic alerts.

How it works

The control plane evaluates the metrics it receives from all PgDog instances on a continuous loop. When they exceed pre-configured thresholds, an incident is created using one of the configured integrations.

Integrations

The following incident management providers are supported:

Provider	Description
incident.io	The control plane will create an incident with the highest available priority.

Configuration

Alerts are disabled by default. To enable them, configure a provider's API key and set thresholds for each supported metric you would like to monitor:

values.yaml

control:
  config:
    alerts:
      evaluation_window_secs: 300
      thresholds:
        clients_waiting: 10
        cpu: 90.0
        memory: 2048
        server_connections: 100
      incident_io:
        api_key: inc_live_xxx

Parameters

Disabled by default

Thresholds have no defaults. If a threshold is not configured, its metric will not be monitored.

The following parameters are configurable via the Helm chart:

Parameter	Description	Unit
evaluation_window_secs	The metrics evaluation window. Metrics are averaged over this period to produce an alert signal.	Seconds, e.g., 300 (5 minutes)
clients_waiting	Average number of clients waiting for a connection from a pool.	Clients, e.g., 10
cpu	Average CPU utilization of each PgDog pod.	Percentage, e.g., 90.0 (90%)
memory	Average memory utilization of each PgDog pod.	MiB, e.g., 2048 (2048 MiB)
server_connections	Average number of open connections from PgDog to Postgres.	Connections, e.g., 100

Evaluation window

Longer evaluation windows reduce the chance of false positives in spiky workloads. Shorter windows will be evaluated quicker and will trigger faster notifications in case of an incident.