CircleCI Server v3.x Operations Overview

The following guide contains information useful for CircleCI server Operators, or those responsible for ensuring CircleCI server 3.x is running properly, via maintenance and monitoring.

It is assumed that you have already read the Server 3.x Overview.

CircleCI server schedules CI jobs using the Nomad scheduler. The Nomad control plane runs inside of Kubernetes, while the Nomad clients are provisioned outside the cluster. The Nomad clients need access to the Nomad control plane, output processor, and VM service.

CircleCI server can run Docker jobs on the Nomad clients, but it can also run jobs in a dedicated VM. These VM jobs are controlled by the Nomad clients, therefore the Nomad clients must be able to access the VM machines on port 22 for SSH and port 2376 for remote Docker jobs.

Job artifacts and output are sent directly from jobs in Nomad to object storage (S3, GCS, or other supported options). Audit logs and other items from the application are also stored in object storage so both the Kubernetes cluster and the Nomad clients will need access to object storage.

Build Environment

CircleCI server 3.x uses Nomad as the primary job scheduler. Refer to our Introduction to Nomad Cluster Operation to learn more about the job scheduler and how to perform basic client and cluster operations.

By default, CircleCI Nomad clients automatically provision compute resources according to the executors configured for each job in a project’s .circleci/config.yml file.

Nomad Clients

Nomad Clients run without storing state, enabling you to increase or decrease the number of containers as needed.

To ensure enough Nomad clients are running to handle all builds, track the queued builds and increase the number of Nomad Client machines as needed to balance the load. For more on tracking metrics see the Metrics and Monitoring section.

If a job’s resource class requires more resources than the Nomad client’s instance type has available, it will remain in a pending state. Choosing a smaller instance type for Nomad clients is a way to reduce cost, but will limit the Docker resource classes CircleCI can use. Review the available resource classes to decide what is best for you. The default instance type will run up to xlarge resource classes.

See the Nomad Documentation for options on optimizing the resource usage of Nomad clients.

The maximum machine size for a Nomad client is 128GB RAM/64 CPUs. Contact your CircleCI account representative to request use of larger machines for Nomad Clients.

For more information on Nomad port requirements, see the Hardening Your Cluster section.

GitHub

CircleCI uses GitHub or GitHub Enterprise as an identity provider. GitHub Enterprise can, in turn, use SAML or SCIM to manage users from an external identity provider.

CircleCI does not support changing the URL or back-end GitHub instance after it has been set up.

The following table describes the ports used on machines running GitHub to communicate with the services and Nomad Client instances.

Source Ports Use

Services

22

Git access

Services

80 or 443

API access

Nomad Client

22

Git access

Nomad Client

80 or 443

API access



Help make this document better

This guide, as well as the rest of our docs, are open-source and available on GitHub. We welcome your contributions.