Summary

  • An individual IS-IS draft describes a link-information TLV moved from LSP fragment F1 to F2. A receiver that sees new F1 beside old F2 can briefly conclude that the link disappeared.
  • Its proposed Originating Generation Identifier (OGID) labels related updates. When the originator asks for coordinated handling with Fragment Count above one, the receiver floods and stores each fragment immediately but delays topology processing until the set is complete or a bounded timer expires.

The example is almost comically administrative. A parent TLV grows, no longer fits in fragment F1, and is repacked into F2. The originator sends updated versions of both. The distant router receives new F1 first. Its database now contains an F1 without the link description and the old F2 without it too. A link that is still up can look absent long enough to provoke route calculation or tunnel movement. The 2 November draft presents this as a possible transient failure mode, not a documented outage.

The proposed fix gives every logical update a 16-bit OGID and places it in each LSP fragment the originator changes. For updates the originator wants processed together, a Fragment Count greater than one tells the receiver how many unique participating fragments to collect. The receiver keeps accepting the LSPs into its link-state database and flooding them to neighbours. What waits is higher-level use of that partial state: Traffic Engineering Database updates and shortest-path computation. When the count is met, the held fragments are released for processing together. Fragment Count one allows immediate handling; it can even accompany an update that touched multiple fragments if coordinated treatment is unnecessary.

This is a bounded hold, not a promise to wait forever. The transaction timer starts with the first held fragment and later arrivals cannot extend it. If a piece never comes, the receiver releases what it has and processes it under ordinary IS-IS rules. The draft’s LRGID back-reference can help a late fragment resolve the last multi-fragment transaction it joined. Several rapid transactions can outrun that single back-reference, leaving an older group unresolved until its timer fires. An operator who buys “atomic updates” as a feature slogan should ask what happens at that timeout.

Mixed fleets have their own limit. RFC 8918 lets a node ignore an unknown TLV while still accepting a valid LSP. Such a node continues the old behaviour, including the possible temporary false removal. The IETF tracker lists OGID as an individual Internet-Draft, and the IANA IS-IS registry has no final OGID codepoint. A named multi-vendor trace showing fragment arrival order, timer expiry, computed paths and actual forwarding would determine whether the proposed receipt prevents this false link-loss decision in practice.