Prometheus Concepts

From NovaOrdis Knowledge Base
Revision as of 18:26, 14 October 2020 by Ovidiu (talk | contribs)
Jump to navigation Jump to search

External

Internal

Overview

Prometheus is a time series database plus tools to collect metrics to be stored in the database as aggregated metrics. Prometheus is a solution for system monitoring. It is not designed to catch individual events in time (such as a service outage) but it is designed to gather aggregated metrics about services. Prometheus is not designed to store raw information, such as text. Prometheus provides alerting capabilities with Alertmanager. Prometheus has some built-in visualization capabilities, but Grafana can also be used.

Target

Prometheus will monitor targets. A target can be a server, database, virtual machine, etc. Different targets can have different pull rates. Targets can be discovered dynamically with service discovery.

Pull vs. Push

Prometheus will actively pull (scrap) metrics, by connecting to targets. The retrieval is done by invoking HTTP into endpoints, which are defined in the configuration.

Advantages of pull vs. push:

  • Centralized control - the configuration is done on Prometheus and not on individual targets.
  • Adjustable pull rate - the system has control over pull rate, and avoids being overloaded.

Pull Rate

Sources of Metrics

Instrumented Applications

Prometheus pulls metrics from instrumented applications. An instrumented application exposes Prometheus metrics on a given URL. Such an application will be identified by Prometheus as a target and scrapped at regular periods.

Exporters

An alternative are prebuilt exporters. Examples: node exporter, SQL exporter, HAProxy exporter.

Push Gateway

Push gateways are used in case of applications or short-lived jobs that do not export metrics directly.

Storage

Alertmanager

Prometheus pushes alerts to Alertmanager via custom rules defined in its configuration files.

Service Discovery

Prometheus can discover targets dynamically.

Metrics

A Prometheus metric is a key/value pair plus a set of labels, commonly referred to as metadata. The key is the name of the metric. The value is a float numerical value.

cpu_usage{core="1",ip="127.0.01"} 14.04

Also see:

Monitoring Concepts | Metrics

Metric Types

Counter

A counter counts elements over time. A counter is a monotonically increasing positive value. A counter can only go up or can be reset. It must never decrease.

Gauge

A gauge measures values that can go up and down.

Histogram

Summary