Summary

  • On 22 September 2026, the MPLS working group sent draft-ietf-mpls-on-path-telemetry-flag-05 to the IESG. “Publication Requested” is a process state, not IESG approval, RFC publication, an IANA allocation or deployment evidence.
  • The proposal uses a single P flag to trigger postcards at MPLS Network Actions-aware nodes whose data collection is enabled. The bit does not specify a universal data set and cannot show by itself that every hop exported.
  • A defensible operational claim therefore needs a trigger-to-export receipt joining the mark, expected path, node capability and configuration, template version, postcard identity, transport loss, collector correlation and any missing-hop result.

The attractive part of postcard-based telemetry marking is easy to see on the wire. A head-end places an MPLS Network Actions sub-stack on a selected packet. One Format D bit, called P, asks nodes along the label-switched path to generate postcards. The packet does not have to carry every measurement that each node may produce.

That economy is also the system boundary. The P flag is an instruction, not a sealed record of what happened after the instruction was seen.

Publication Requested marks a handoff

Datatracker records revision 05 on 21 September. A day later, the MPLS working-group state changed to “Submitted to IESG for Publication” and the IESG state changed to “Publication Requested”, with Jim Guichard named as responsible area director and action holder.

Those labels matter because they identify who holds the next decision. They do not say that the IESG has approved the text. No RFC has been published from this draft, and the live IANA table for Network Action Flags Without Ancillary Data contained no registration at the evidence freeze. The shepherd write-up asks for Proposed Standard, reports broad working-group support and no known implementations. It also identifies two IPR disclosures. None of that is a deployment claim.

There is one existing number that can create confusion. RFC 9994 already assigns MNA Sub-Stack Opcode 1. The PBT-M draft uses that action and asks for a new flag within it. An existing opcode and a requested flag allocation are different registry facts.

One bit activates many local decisions

Revision 05 places a Format D label stack entry after the action entry. The P action is hop-by-hop, has no ancillary-data entry, and sets U=0 so a node that does not understand the action skips it. An aware node generates one postcard for a marked packet only when data collection is enabled, and it exports the data types selected by local control-plane configuration.

The head-end therefore decides that a packet should be observed, but it does not carry a complete measurement contract. One node may be unsupported. Another may understand PBT-M but have collection disabled. Two enabled nodes may use different templates or versions. A node may suppress export after reaching its generation cap. The packet can keep forwarding in every case.

The draft makes that last distinction explicit. It says only a small subset of packets should be marked, never every packet, and gives a default of no more than one in 1,000 packets per flow. Each node limits postcard generation to an average of 1,000 per second with a burst of 2,000. Packets above the cap continue to be forwarded without a postcard, while a counter records the excess.

An unbroken user packet is thus compatible with an incomplete postcard set. Forwarding success and observation completeness are separate results.

A postcard is a local claim

Each postcard describes what one exporting node observed under its own configuration. The collector still has to decide which postcards belong to the same user packet and which ordered traversal they represent.

The draft warns that postcards may arrive out of order or be lost. A single time-to-live value is not always enough when a path pushes or pops labels. It discusses a vector of TTL values across the label stack, a node identifier, a flow identifier and a timestamp as correlation material. If two packets can share the available key while both are in flight, their postcards may be ambiguous. Ambiguous correlations must be discarded, not completed by assumption.

That makes absence a classified condition rather than a blank to fill. A missing postcard can indicate an unaware node, disabled collection, a different template, export suppression, transport loss, collector loss, a path change that was not captured, or a correlation failure. The P flag alone cannot distinguish them.

Keep a trigger-to-export receipt

The smallest useful record should join the initiating decision to the collector result without claiming more than the evidence supports.

Receipt field Evidence to preserve
Trigger P flag, ingress point, marking policy, sampler version and sampling rate
Scope Trust domain, expected label-switched path and observation window
Node readiness Node identity, PBT-M capability, enabled state and relevant software/configuration version
Semantics Local telemetry template, template version and selected data types
Postcard identity Packet or flow correlation key, sequence material and export timestamp
Delivery Export destination, transport, sent count, loss, suppression and rate-limit counters
Reconstruction Correlation rule, received node order, discarded ambiguities and time boundary
Completeness Expected nodes, received nodes, missing hops and explicit partial-path status
Safety Per-node caps, exceptions, exceeded counters and trust-boundary handling

This is not an extra packet format. It is an operational receipt assembled from the control plane, exporter and collector. A team can keep the in-packet mechanism small while preserving a reproducible explanation of what the resulting path statement means.

Integrity has a boundary too

The draft treats the P flag as mutable and unauthenticated. At a trust-domain boundary, an incoming flag is to be cleared or the packet dropped. Inside the trusted domain, a compromised or misconfigured node could set or clear the bit. Rate limiting bounds the postcard load; it does not authenticate the observation or stop low-rate tampering.

The specification also defines no YANG model. That is not a defect claim. It means capability discovery, enablement, template identity, export health and collection policy will need explicit operational treatment rather than being inferred from the data-plane bit.

PBT-M can offer a fixed 12-byte label-stack instruction while moving richer data out of band. That is a real design advantage. It should be described accurately: the packet carries a cheap trigger; the network and collector must carry the proof.

Sources

  1. IETF Datatracker: Postcard-Based Telemetry using MPLS Network Actions
  2. Datatracker document history
  3. Working-group shepherd write-up
  4. Revision 05 plain text
  5. IANA MPLS Network Actions registry
  6. RFC 9994: MPLS Network Actions Framework