Alertmanager

Why it's ranked here

Alertmanager is the default choice here, not one alternative among several: it is the alert handler the rest of the Prometheus ecosystem is built to talk to, released under the Apache license, with a high-availability design that states its own tradeoff plainly rather than hiding it, choosing a duplicate notification over a missed page during a network split. The codebase backs that up with real engineering care: an internal lifecycle type with an explicit teardown stack, built to run both as a standalone binary and embedded inside tests.

What's good

The routing tree does real work: nodes can attach arbitrary labels that flow into notification templates, and matching rules inherit down the tree so a team-wide default only needs to change in one place. Templates run on the standard Go template engine with built-in helpers for JSON encoding, timezone conversion, and duration formatting, so receivers get well-formed messages without a separate templating layer bolted on. Clustering uses a gossip protocol with no leader election, so any surviving instance keeps serving alert state and history if the others go down.

Tradeoffs

The high-availability design is explicit about what it does not guarantee: peers replicate state over a best-effort, eventually-consistent gossip layer with no quorum or leader, so a network split during an incident can produce the same page twice rather than risk losing it, and that tradeoff cannot be switched off. The routing configuration format still carries an older matching syntax alongside the newer one meant to replace it, so real configuration files often mix both styles. Outbound email notifications default to requiring an encrypted connection, worth checking before pointing the tool at an internal mail relay that does not support one.

How to use it well

Point it at whatever already emits alerts over its HTTP API, most naturally a Prometheus server, and spend the setup time on the routing tree rather than the receiver list: get grouping and inheritance right first, then layer in rules that quiet the noise a major outage produces. Running more than one instance is worth doing early, since the cluster flags are meant to be set from the start rather than bolted on later, and every instance should get traffic directly rather than sitting behind a load balancer. It is a router and notifier only; it does not evaluate metrics or generate alerts itself.

Observed

License

Apache License 2.0, per license headers in source files.

Language

Implemented in Go; module path github.com/prometheus/alertmanager.

Interfaces

Ships a companion CLI, amtool, alongside the alertmanager binary.

Interfaces

Interfaces: HTTP API (v2, OpenAPI/go-swagger generated), YAML configuration file, web UI for silences, and a CLI.

Packaging

Distributed as precompiled binaries and Docker images, in addition to building from source.

Config reload

Configuration can be reloaded without a restart, via SIGHUP or an HTTP POST to /-/reload.

Clustering

High-availability clustering uses a gossip protocol (Hashicorp memberlist) with no leader election or quorum requirement.

Templating

Notification templates use Go's text/template system with a documented set of built-in functions.

SMTP

SMTP notifications require an encrypted connection to the mail server by default.