Configuration file

When deploying with NixOS, NixOps or nix-darwin you should use the module documentation instead, or skip ahead to read about less frequently used options to use with extraOptions.

The configuration format for the agent is TOML markup. Its location can be specified by invoking hercules-ci-agent --config agent.toml. Alternatively, the NixOS and nix-darwin module can generate this file for you. See also the setup guide.

A basic agent.toml looks as follows:

# See https://docs.hercules-ci.com/hercules-ci-agent/agent-config/
baseDirectory = "/var/lib/hercules-ci-agent"
# concurrentTasks = "auto"

If you are deploying an agent, the guide is a better starting point.

apiBaseUrl

Optional. Defaults to https://hercules-ci.com.

HTTP API agent will connect to.

baseDirectory

Required.

Directory with all the agent state: secrets, work, etc.

binaryCachesPath

Optional. Defaults to staticSecretsDirectory/binary-caches.json if that file exists. With the NixOS module this defaults to /var/lib/hercules-ci-agent/secrets/binary-caches.json.

Path to a JSON file containing binary cache secret keys.

The contents of the file are described in binary-caches.json.

clusterJoinTokenPath

Optional. Defaults to staticSecretsDirectory/cluster-join-token.key.

Path to a secret token retrieved when creating a new agent via hercules-ci.com/dashboard.

This token is used for authentication with apiBaseUrl.

concurrentTasks

Optional. Defaults to "auto". Must be an integer or "auto".

Number of tasks to perform simultaneously.

A task is a single derivation build, an evaluation or an effect run. At minimum, you need 2 concurrent tasks for <literal>x86_64-linux</literal> in your cluster, to allow for import from derivation.

When the value is "auto", the number becomes the number of CPU cores, but with a lower bound of 2 in order to avoid deadlock during import from derivation.

The optimal value depends on the resource consumption characteristics of your workload, including memory usage and in-task parallelism. This is typically determined empirically.

When scaling, it is generally better to have a double-size machine than two machines, because each split of resources causes inefficiencies; particularly with regards to build latency because of extra downloads.

labels

Optional. A key-value map of user data.

This data will be available to organization members in the dashboard and API.

The values can be of any TOML type that corresponds to a JSON type, but arrays can not contain tables/objects due to limitations of the TOML library. Values involving arrays of non-primitive types may not be representable currently.