Federated Analytics — Nodes¶

Overview¶

This page is for node operators. It explains how to enable or disable Federated Analytics (FA) on a node and what that means in practice.

Federated Analytics lets researchers compute statistics across data held by multiple nodes without the raw data ever leaving any node. When a researcher sends an analytics request, each node computes the statistics locally and sends back only the aggregated summaries.

• For which datasets support FA and how to make a custom dataset FA-compatible, see Federated Analytics — Datasets.
• For how to run analytics as a researcher, see Federated Analytics — Researcher.

Enabling or Disabling FA¶

Federated Analytics is enabled by default on every node. The setting lives in the [security] section of the node configuration file:

[security]
allow_federated_analytics = True

To disable FA on a node, set the value to False:

[security]
allow_federated_analytics = False

Once a dataset is registered on a node with the appropriate tags, it is automatically eligible for analytics requests — no further configuration is needed beyond this flag.

Secure aggregation¶

Analytics requests can run under secure aggregation, where each node masks its contribution so the researcher only ever recovers the combined result. This relies on the same node-side secure-aggregation mechanism as federated training — no analytics-specific setup is required beyond enabling FA.

Enforcing secure aggregation¶

A node can require secure aggregation for every action that uses its datasets — including analytics. Set force_secure_aggregation = True in the [security] section of the node configuration:

[security]
secure_aggregation = True
force_secure_aggregation = True

When forced, the node rejects any analytics request that does not carry a secure-aggregation context. The researcher must then run the experiment with secagg=True.

Common Errors & Troubleshooting¶