HashiCorp Nomad

HashiCorp Nomad is a workload orchestrator: a cluster scheduler that places and keeps applications running across a fleet of machines. It runs containers, binaries, Java apps, and other workloads through pluggable task drivers, using one declarative job format instead of a container-only control plane.

itCloud native tools and technologies

Don't Panic — HashiCorp Nomad

Nomad is HashiCorp's workload orchestrator: you declare what should run, and it keeps placing that work on machines until reality matches the declaration. It is the "make sure this is running somewhere sensible" layer, not the "own the entire cloud-native universe" layer.

Before Nomad (and friends), operators pinned apps to named hosts and hoped the hosts stayed polite. Container platforms fixed a lot of that for Linux containers, then shrugged at Windows services, Java archives, batch pipelines, and plain binaries. Nomad's bet is one job format for that mixed pile, executed through task drivers (plugins that actually start Docker, exec, Java, QEMU, and so on).

Three ideas carry most of the weight.

First, servers and clients. Servers store cluster state and run the scheduler. Clients are the machines that execute work. A production region usually runs three or five servers so Raft can keep a quorum; clients can be numerous. A datacenter is a placement label inside a region, not a second control plane.

Second, the job hierarchy: job → group → task. Tasks in a group are co-located on one client. When Nomad assigns a group to a client, that assignment is an allocation. Job type (service, batch, system, sysbatch) picks which scheduler personality you get.

Third, placement is two acts: feasibility (can this node run it?) then ranking (which feasible node is best?). Ranking leans on bin packing: fill machines densely without exhausting any resource dimension. That is great for utilization and slightly rude to your blast radius, which is why spreads and rack metadata exist.

The surprise for Kubernetes-shaped brains: Nomad refuses to become your service mesh, secret engine, and ingress platform. It composes with Consul, Vault, and Terraform on purpose. Native Nomad services can cover basic discovery; they do not retire Consul when you already standardized on the catalog or need a mesh.

Pending allocations with idle CPU are usually inventory lies (missing driver, bad constraint), not a plea to buy hardware. Read allocation events before you autoscale.

Next: the Intro for the architecture walk; Slides for the decision map; Cheatsheet when you need the tables; Practice and Exercise when you want a local nomad agent -dev loop; Field Notes when you want the scars without the brochure.

Don't Panic — HashiCorp Nomad

Relevant careers