Overview --- This PR ensures the rest of our `dpd` configuration is covered by a RPW to help recover state in the event of `dendrite` crashing, the switch zone restarting / being replaced, or the sled restarting. This is accomplished via a background task in Nexus that periodically ensures `dpd` is up to date with the tables in Nexus. The tradeoffs of this design is that we don't track versioning and reconcile the entire state every time, but since the actual number of ports will never be that high (relative to something like NAT entries) the tradeoff of less efficiency for much greater simplicity seems to make sense today, and it requires much less rework in Nexus and Dendrite should we choose to replace this strategy down the road. Tasks --- - [x] Ensure that Service Zones configured during rss, cold boot, and nexus have their NAT entries added to the NAT RPW table (extracted into #4857) - [x] Create background task that periodically reconciles switch port configuration for dendrite instances - [x] Move switch zone uplink SMF property updates to RPW - [x] Move routing updates (via mg) to RPW - [x] Static Routing - [x] BGP - [x] Move bootstore updates to RPW - [x] Move loopback address management to RPW - [x] Move Nexus-side switch zone service on-demand lookups as outlined in #5092 Verifications Performed --- - [x] Basic instance deployment - [x] Loopback Address Creation - [x] BGP configuration (a4x2) - [ ] BGP configuration modification (a4x2) - [x] Static routing - [x] Static routing configuration modification Related --- Closes #4715 Closes #4650 Depends on https://github.com/oxidecomputer/dendrite/pull/838 --------- Co-authored-by: Levon Tarver <levon@oxidecomputer.com>

Service NAT entries missing after dendrite restart #4650

Description

Metadata

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

Issue actions