BREAKING: Rename remaining network attributes from net.* to network.* and align definitions with ECS

[trask]

Fixes #3371
Fixes #3438

Changes

Introduce:

• network.transport defined as OSI Transport Layer or Inter-process Communication method.
• network.type defined as OSI Network Layer or non-OSI equivalent.

(note: ECS defines both network.transport and network.type only in terms of OSI, but we expand the definition here to be more general)

Remove net.transport and net.sock.family.

Rename:

• net.protocol.* --> network.protocol.*
• net.host.connection.* --> network.connection.*
• net.host.carrier.* --> network.carrier.*

Summary of benefits

More precise definition of network.transport and network.type in terms of OSI model where applicable.

These attributes are used in ECS security use cases, so the cost of breaking these on the ECS side is higher, while these remaining net.* attributes are not heavily used in OpenTelemetry instrumentation (and we can bundle this into the transition plan for the larger breaking changes).

We can bring in the remaining ECS network.* attributes without causing ECS more breakage.

[tigrannajaryan]

I wonder if this impacts our ebpf implementation.

@open-telemetry/ebpf-maintainers @open-telemetry/ebpf-approvers can you comment? Is this a breaking change from your perspective?

[jmw51798]

This is not a breaking change for opentelemetry-ebpf because it does not currently emit any metrics defined in the opentelemetry-specification (although that should probably change). Instead, it uses historical metric names documented here: https://github.com/open-telemetry/opentelemetry-ebpf/blob/main/docs/metrics/metrics.yaml.

[jsuereth]

After a long discussion, +1. I feel like this really clarifies some very confusing x-layer attributes we had previously.

[Oberon00]

@jsuereth

x-layer attributes

Can you clarify what you mean by this?

[trask]

@jsuereth

x-layer attributes

Can you clarify what you mean by this?

I believe what he liked was that the attributes are now defined in terms of OSI layers.

[trask]

don't merge yet, I need to double check the merge conflicts...

[trask]

good to go!