Summary

  • Revision 07 defines compact IPFIX fields for MPLS and SRv6 Path Segment Identifiers, but a PSID is meaningful only inside the scope and mapping maintained by the responsible operator.
  • A candidate-path switch can change the PSID while local fast reroute can change forwarding without changing it; neither a stable nor a changed value closes an incident by itself.

The collector showed the same PSID before and after the link failure. The incident timeline therefore labelled the route “unchanged.” That conclusion was tidy, machine-readable and wrong.

The logical Segment Routing path had remained the same, so its Path Segment Identifier remained the same. A local protection mechanism had nevertheless moved packets around the failed resource. The exporter had decoded the packet correctly. The collector had stored the field correctly. The error appeared later, when an identifier for a logical path was promoted into proof of the physical forwarding history.

That is the useful leadership question raised by revision 07 of Export of Segment Routing Path Segment Identifier Information in IPFIX. The draft proposes two Information Elements: a three-octet psidMplsLabelStackSection for the MPLS label entry containing the PSID, and a 128-bit srhPsidIPv6 for SRv6. Both can make flow telemetry more compact than repeatedly exporting a full segment list. Neither makes a PSID self-describing.

The field is not the mapping

A PSID can identify one segment list. It can also aggregate several lists under a candidate path, or several candidate paths under one SR Policy. The entity that establishes the association—perhaps a controller, ingress or egress—chooses the intended granularity and must keep values distinguishable within its scope. Reuse outside that scope is undefined for correlation.

The numeric value is therefore only half a key. A defensible join also needs the exporter and Observation Domain, SR technology, direction, mapping authority, declared scope and the mapping version or interval effective when the Flow Record was created. The draft explicitly leaves the procedures for obtaining and maintaining that mapping outside scope. That is not a flaw in the field definition. It is a warning against asking the field to govern more than it carries.

A collector may query a controller or consume BGP-LS state. Either route gives it a second evidence stream. If that stream is late, overwritten or drawn from another domain, a correctly decoded PSID can be joined to the wrong policy. Retaining only the latest mapping makes historical records especially dangerous: yesterday’s value may be reinterpreted using today’s meaning.

A change and a non-change say different things

Revision 07 is unusually clear about change semantics. When traffic switches to a backup candidate path, packets carry the PSID assigned to that new path and subsequent Flow Records can show the new value. When the logical path stays intact but fast reroute or another local protection action changes forwarding, the segment list and PSID can stay unchanged.

So a changed PSID can support the claim that the declared path context changed. It does not prove that every packet used the new path, that the mapping reached the collector on time, that convergence completed, or that the service recovered. An unchanged PSID supports an even narrower claim: the exported logical identifier did not change. It cannot exclude a local detour, different links, queue movement, degradation or loss.

The same discipline applies to asymmetry. The forward record names only the current unidirectional context. It does not identify a reverse path or prove that one exists. And a PSID is explicitly not a quality metric. Delay, loss and service behaviour require independent observations joined to the same bounded interval.

SRv6 adds a decoding gate

For SRv6, the Metering Process must understand the G-Flag. Only when that flag is set and the last Segment Routing Header entry actually carries an SRv6 PSID should the new field be used. If the field appears while the G-Flag is unset, revision 07 says it should be ignored.

That rule matters because 128 bits do not authenticate their own type. Without the flag and application context, the last entry can be an opaque object with another meaning. Existing segment-list fields may accompany the PSID, but reduced mode can omit the first SID from the header, while compressed SIDs require additional decoding. More bytes can improve context without turning one packet view into a universal path oracle.

The receipt chain leadership needs

The minimum useful operational receipt is not PSID=42. It is a versioned chain:

  1. the Exporting Process, Observation Domain, Metering Process and Template that produced the record;
  2. the packet direction, SR technology, PSID field and observation time;
  3. the responsible mapping source, scope, effective interval and policy/candidate/list meaning;
  4. the evidence that the mapping reached the analytics system before interpretation;
  5. independent forwarding, local-protection and performance observations; and
  6. the application or service outcome whose risk authorises a decision.

This chain also limits disclosure. PSID exports can reveal path topology and traffic distribution beyond a normal five-tuple. The right to collect the field is not automatically the right to expose the mapping to every analyst, vendor or tenant.

Heng Lu’s Running-Code Primacy supplies the ordering. A registered field describes an observation; it does not create forwarding reality. Minimum Initial Specification supports a small portable encoding and deterministic decoding rule. Mapping retention, incident thresholds and remediation stay with the operator that bears the consequence. The draft becomes real through protected export, maintained associations and repeated operational use—not through Working Group status.

Document status and uncertainty

Revision 07 is an active OPSAWG Working Group Internet-Draft. Its header says Standards Track, while the frozen Datatracker record exposes no intended-status value. The two Element IDs remain TBD1 and TBD2. The record establishes no IANA assignment, implementation, deployment, interoperability result, incident or measured outcome.

That uncertainty strengthens rather than weakens the decision rule: never let a PSID value alone close a path incident or trigger remediation. First prove which scoped mapping applied at the record’s time. Then prove what forwarding did and what the service experienced.

Sources