Skip to content

v0.27.0: Network services and HTTPProxy traffic controls

Choose a tag to compare

@scotwells scotwells released this 16 Sep 22:15
3166239

Adds NetworkService for grouping VPC workloads behind a proxy, and gives HTTPProxy weighted multi-backend rules, load balancing algorithms, and passive health checks.

What's new

  • NetworkService. A new resource groups network interfaces by label and reports per-location member and health counts. (#439, #441)
  • networkService backends for HTTPProxy. A proxy can target a network service port, served at the edge through the tenant VRF with an offline page when no member is healthy. (#443, #444)
  • Hub federation for networks and VPC pods. Networks are mirrored to the federation hub, and VPC pod endpoints reach only the edges that serve them. (#440, #445)
  • Weighted backends per rule. An HTTPProxy rule can carry up to 16 backends, each with an optional weight. (#453)
  • Load balancing algorithms. spec.loadBalancer selects round-robin, random, least-request, or consistent-hash routing. (#455)
  • Passive health checks. Consecutive 5xx responses eject an origin for a while, defaulting to 5 errors, 30s, and 50% of the pool. (#468)
  • Declared hypervisor attachment. A network interface claim can ask for a hypervisor device the guest must be told about. (#450)
  • Networks portal plugin. Consumers can browse a project's Networks in the cloud portal. (#463, #466, #467)
  • Large responses no longer stall on VPC networks. New Networks default to an MTU of 1440 instead of 1460. (#452)
  • Host header override for connector backends. The override now reaches origins behind a tunnel. (#456)
  • HTTPProxy creation fixed. A validation rule no longer rejects rules without backends. (#461)

Note

New CRDs: NetworkService and EdgeReachability. HTTPProxy changes only add fields or relax validation, so existing resources keep working without any conversion. Existing Networks keep their MTU; running Instances pick up a new value only when recreated.