Skip to content

Adoption & Ecosystem

Who uses it

The project's ADOPTERS.md lists one organisation publicly: G-Research, the company that started Armada. No other organisations are listed there, so only this one is cited below.

OrganisationUse caseSource
G-ResearchRuns Armada in production to process millions of jobs daily across tens of thousands of nodesADOPTERS.md

If you adopt Armada and want to be listed, the ADOPTERS file invites a pull request (ADOPTERS.md:5). Beyond G-Research, there is no public, citable adopter list, so the GitHub signals below are the available evidence of usage.

Adoption signals

Measured from the GitHub REST API on 2026-06-26:

  • Stars: 602
  • Forks: 166
  • Open issues: 109
  • Contributors: roughly 102 (contributors API last page)
  • Repository created: 2019-06-19; last push: 2026-06-26
  • Latest release: v0.21.6 (2026-06-26)

Armada is a CNCF Sandbox project, accepted on 2022-07-25 (CNCF project page). Releases are frequent, which is itself a signal of active maintenance.

Ecosystem

  • Apache Pulsar is a required dependency; it is the event log that routes all messages between subsystems (docs/system_overview.md:62).
  • PostgreSQL backs the scheduler and Lookout; Redis is part of the local development stack (README:81).
  • Prometheus integration exposes analytics on system behaviour and resource allocation (README:30).
  • An Airflow operator integrates Armada into Airflow workflows (docs/armada_airflow_operator.md).
  • Client libraries exist for Python, Java, and .NET (docs/client_libraries.md).
  • The Armada Operator is the recommended install path for Kubernetes (README:47-50).

Alternatives

Armada is a multi-cluster, out-of-cluster meta-scheduler. Most alternatives schedule within a single cluster, which is the main distinction.

AlternativeDiffers by
Volcano (CNCF Incubating)In-cluster batch scheduler for one Kubernetes cluster; Armada queues and schedules across many clusters from outside them.
Kueue (Kubernetes SIG)In-cluster job queueing; single-cluster focus rather than Armada's cross-cluster queues.
Apache YuniKornBatch and data scheduler for Kubernetes, still oriented to in-cluster scheduling.
Karmada / Open Cluster ManagementMulti-cluster, but for general workload federation rather than high-throughput batch queueing.

Pick Armada when batch scale exceeds a single cluster and you need queues of millions of jobs without overloading etcd (README:21). Pick an in-cluster scheduler when one cluster covers your scale and you want a simpler deployment.