Network awareness and traffic policies

Argo separates observing the network, scheduling its own work, and governing system traffic.

NetworkManager

argod observes active connectivity, connection type, interface, and metered state over D-Bus. It recovers subscriptions after transient failures and reconciles policy when interfaces disconnect or change.

When NetworkManager is unavailable, ordinary downloads continue where the host network still works. The daemon reports awareness as unavailable instead of pretending it has current interface state.

Priority is not policy

Download priority (low, normal, or high) orders Argo’s queued transfers. Traffic policy controls how active Argo connections compete with the default system class. Selecting high priority does not automatically enable kernel shaping.

Policy	Current behavior
off	No active Argo shaping state
focus	Argo receives a 20% guaranteed class; unused capacity remains borrowable
balanced	Argo and default traffic receive equal guaranteed classes
throughput	Argo receives an 80% guaranteed class
latency	A bounded controller adjusts Argo’s ceiling from smoothed latency telemetry

argo policy balanced
argo status

Policy application also requires an active transfer, a connected interface, a nonzero qos.link_rate, and the argo-qosd helper. If those conditions are absent, the requested policy can remain selected while shaping is reported inactive or unavailable.

Receive-path shaping

Argo identifies argod sockets through a cgroup selector, transfers that identity through conntrack, and redirects matching ingress packets to an Argo-owned IFB. HTB classes and fq_codel then shape the received payload path. This is distinct from merely limiting outbound TCP acknowledgements.

The helper owns narrowly named nftables and tc objects and refuses arbitrary shell commands or invalid interface/rate inputs. It does not intentionally flush administrator rulesets or remove unrelated qdiscs.

Adaptive latency

The latency policy establishes a baseline from multiple samples or an explicit manual_baseline. It changes the current Argo ceiling gradually within configured min_rate and max_rate bounds. Missing or stale telemetry produces a safe bounded state rather than an unbounded numerical decision.

Use argo status to inspect measured latency, baseline, controller state, current rate, and helper errors. Results depend on a representative configured probe_target; an empty target means active latency sampling is unavailable.