Observability XXXII · Logging Pipeline ArchitectureLoggingPipeline

Agents vs Sidecars

Intermediate⏱ ~18 minbash

What you'll learn

Explain agents vs sidecars in production terms
Configure and operate agents vs sidecars in a production observability stack
Recognise and diagnose the most common failure modes
Apply the discipline to a real environment

Prerequisites

The Logging Pipeline

Verified against Prometheus 2.55.x · Alertmanager 0.28.x · node_exporter 1.8.x · blackbox_exporter 0.26.x · Grafana 11.x · Loki 3.x · Tempo current · OpenTelemetry Collector 0.110.x · Grafana Alloy current · Docker Engine 28.x · Ubuntu 24.04 LTS · Debian 12 (Bookworm) · RHEL / Rocky / AlmaLinux 9.x · 2026-08-13

Not yet marked complete on this device.

Agent runs as a host service. Sidecar runs alongside the application.

A sidecar per container is high-overhead at scale; an agent per host is most common.

What it is

A precise definition of agents vs sidecars, scoped to production operations.

Why a sysadmin cares

Production framing.

How it works

The mental model.

example_setting: value

How to configure it

Real configuration examples with annotated options.

promtool check config /etc/prometheus/prometheus.yml

How to validate it

Commands the operator runs to confirm the configuration is live and correct.

How it can fail

The high-frequency failure modes: silent misconfiguration, crash on load, performance regression, permissions failure, schema / version drift.

How to troubleshoot it

The diagnostic order.

Security implications

Agents vs Sidecars has security implications wherever the relevant component exposes an HTTP endpoint, an authentication layer, or a credential.

Performance implications

Performance implications come from cardinality, scrape / push interval, rule size, retention, and query cost.

Production guidance

Validate before applying.
Test changes in a non-production environment.

Verification

You should now be able to answer:

What is agents vs sidecars in production terms?
Why does a sysadmin care about it?
How does it fail and how do you diagnose the failure?