hashicorp/nomad

Nomad is a workload orchestrator for teams that want to run a mix of containerized, batch, legacy and virtual-machine workloads on the same infrastructure without adding a separate storage or coordination service. It runs as one self-contained binary on Linux, Windows and macOS, combining resource management and scheduling; leader election and state replication provide high availability, and the README states the product has scaled to clusters of 10K+ nodes in production. A commercial Nomad Enterprise edition also exists.

What changes in practice

Instead of requiring every application to be containerized, Nomad uses pluggable task drivers, with documented support for Docker, Podman, executables, Java and QEMU virtual machines README. That means a legacy service, a batch job and a container can share one scheduling layer. It also detects accelerators such as GPUs, FPGAs and TPUs through device plugins, so ML and other accelerator-hungry jobs can be scheduled alongside ordinary services.

Federation is built in, so one deployment can span multiple regions and clouds. Nomad integrates with Terraform, Consul and Vault for provisioning, service discovery and secrets management; these are described as integrations, not prerequisites for every setup, though a cluster using them inherits their operational requirements. Clients can drive it through the CLI or the HTTP API CLI docs, API docs.

Getting started and running it

For a local, non-production cluster the project points to the Getting Started tutorials; the repository's terraform directory optionally contains manifests for a development cluster on a public cloud Getting Started, Terraform manifests. Production use should follow the published reference architecture Production reference.

Constraints to weigh

The source is licensed BUSL-1.1 rather than a standard open-source licence, which matters if you intend to offer Nomad as a service or build a competing product; the badge in the README states the licence and the repository itself carries the authoritative terms Repository. The README is a product overview, not an evaluation: claims about scale and failure handling are the author's, and there is no supplied benchmark of throughput or recovery times. Older Hacker News discussion from the 2020 v1.0 release called it a simpler-to-operate alternative to Kubernetes but asked for a complete application walkthrough; those comments describe that version and are not current documentation.

HotRepo's take

Interest signals here are too thin to read as momentum: one recorded daily-board appearance on 2026-10-30 with +1 star does not show a trend Repository. The practical question is whether a single-binary orchestrator with mixed-workload support fits your stack and licence constraints, not whether the project is trending.