Context
Third-party clients building network-inventory or monitoring views can today retrieve full static NIC configuration via info.networkInterfaces (IPs, gateway, DHCP, IPv6, status) and hardware metadata via info.devices.network (vendor, model, speed, virtual-flag). What's missing is the dynamic side — live throughput counters that would let clients render bandwidth graphs or per-interface activity indicators.
Existing pattern that fits perfectly
The schema already exposes live system stats via Metrics, with both query and subscription surfaces:
type Metrics {
cpu: CpuUtilization # live CPU load (percentTotal + per-core)
memory: MemoryUtilization # live RAM (used/free/available/swap/percent)
temperature: TemperatureMetrics
}
type Subscription {
systemMetricsCpu: CpuUtilization!
systemMetricsMemory: MemoryUtilization!
systemMetricsTemperature: TemperatureMetrics
}Network is the obvious sibling that should fit right alongside — same shape, same access pattern, same subscription cadence. The data is already trivially available on the host (/proc/net/dev, /sys/class/net/<iface>/statistics/*).
Proposal
type Metrics {
# … existing fields
network: NetworkUtilization
}
type NetworkUtilization implements Node {
id: PrefixedID!
interfaces: [NetworkInterfaceStats!]!
}
type NetworkInterfaceStats implements Node {
id: PrefixedID!
iface: String!
rxBytes: BigInt!
txBytes: BigInt!
rxPackets: BigInt!
txPackets: BigInt!
rxErrors: BigInt!
txErrors: BigInt!
rxDropped: BigInt!
txDropped: BigInt!
carrier: Boolean
mtu: Int
}
type Subscription {
# … existing
systemMetricsNetwork: NetworkUtilization!
}Why this matters
CPU, memory, and temperature already stream live to clients; network is the conspicuous gap. Adding it lets dashboards/mobile clients show real-time per-interface throughput without falling back to SSH-based scraping of /proc/net/dev, and slots into the existing Metrics consumer code path on the client side with near-zero new abstractions.
Context
Third-party clients building network-inventory or monitoring views can today retrieve full static NIC configuration via
info.networkInterfaces(IPs, gateway, DHCP, IPv6, status) and hardware metadata viainfo.devices.network(vendor, model, speed, virtual-flag). What's missing is the dynamic side — live throughput counters that would let clients render bandwidth graphs or per-interface activity indicators.Existing pattern that fits perfectly
The schema already exposes live system stats via
Metrics, with both query and subscription surfaces:Network is the obvious sibling that should fit right alongside — same shape, same access pattern, same subscription cadence. The data is already trivially available on the host (
/proc/net/dev,/sys/class/net/<iface>/statistics/*).Proposal
Why this matters
CPU, memory, and temperature already stream live to clients; network is the conspicuous gap. Adding it lets dashboards/mobile clients show real-time per-interface throughput without falling back to SSH-based scraping of
/proc/net/dev, and slots into the existing Metrics consumer code path on the client side with near-zero new abstractions.