Files
tailscale/kube/egressservices/egressservices.go
T
BeckyPauley 6ee7bcb458 cmd/{k8s-operator,containerboot,kube}: support IPv6 in egress ProxyGroup (#19898)
* cmd/{k8s-operator,containerboot,k8s-proxy},kube: support IPv6 in egress ProxyGroup

  Add support for dual-stack and IPv6 clusters in egress ProxyGroup.
  Previously, egress ProxyGroup only supported IPv4: the operator and
  containerboot assumed IPv4 for ClusterIP Services, EndpointSlices,
  and health check headers.

  This change introduces the following:

  - Create a per-family EndpointSlice instead of a single IPv4
    EndpointSlice.

  - Update the egress services readiness reconciler to account for
    both IPv4 and IPv6 EndpointSlices.

  - Update the pod readiness reconciler to use the primary Pod IP
    (PodIPs[0]) for readiness checks, instead of hard-coding to use
    IPv4.

  - Update the /healthz handler to return both PodIPv4Header and
    PodIPv6Header.

  - Add an IPv6 address field to egress status.

  - Update containerboot and k8s-proxy to use the new health check
    logic.

Updates tailscale/corp#41677

Change-Id: If66a3146df48c75b1e65a71632bbc9fc75feded2
Signed-off-by: Becky Pauley <becky@tailscale.com>

* cmd/{k8s-operator,containerboot}: improve dual-stack egress ProxyGroup

On dual-stack clusters, an egress ProxyGroup Service has one EndpointSlice
per IP family (IPv4 and IPv6). However, EndpointSlices were only recreated
when the ExternalName Service configuration changed, so a deleted
EndpointSlice was not recreated. The egress readiness reconciler also had
no mechanism to identify which IP families should exist (previously only
an IPv4 EndpoitSlice was required).

We now create an EndpointSlice for every IP family the ClusterIP Service
supports.

Also mark an egress Service NotReady when an EndpointSlice for an expected IP
family (derived from the Service's ClusterIPs) is missing, so a
dual-stack Service missing a family's EndpointSlice is no longer reported
Ready.

Clarify that the egress pre-shutdown and Pod readiness health checks
verify only one IP family on dual-stack clusters.

Change-Id: I35b03daf76ac817cd516e9a731770b2d85f6ee16
Signed-off-by: Becky Pauley <becky@tailscale.com>

---------

Signed-off-by: Becky Pauley <becky@tailscale.com>
2026-07-14 13:26:40 +01:00

112 lines
3.5 KiB
Go

// Copyright (c) Tailscale Inc & contributors
// SPDX-License-Identifier: BSD-3-Clause
// Package egressservices contains shared types for exposing tailnet services to
// cluster workloads.
// These are split into a separate package for consumption of
// non-Kubernetes shared libraries and binaries. Be mindful of not increasing
// dependency size for those consumers when adding anything new here.
package egressservices
import (
"encoding/json"
"net/netip"
)
const (
// KeyEgressServices is name of the proxy state Secret field that contains the
// currently applied egress proxy config.
KeyEgressServices = "egress-services"
// KeyHEPPings is the number of times an egress service health check endpoint needs to be pinged to ensure that
// each currently configured backend is hit. In practice, it depends on the number of ProxyGroup replicas.
KeyHEPPings = "hep-pings"
)
// Configs contains the desired configuration for egress services keyed by
// service name.
type Configs map[string]Config
// Config is an egress service configuration.
// TODO(irbekrm): version this?
type Config struct {
HealthCheckEndpoint string `json:"healthCheckEndpoint"`
// TailnetTarget is the target to which cluster traffic for this service
// should be proxied.
TailnetTarget TailnetTarget `json:"tailnetTarget"`
// Ports contains mappings for ports that can be accessed on the tailnet target.
Ports PortMaps `json:"ports"`
}
// TailnetTarget is the tailnet target to which traffic for the egress service
// should be proxied. Exactly one of IP or FQDN should be set.
type TailnetTarget struct {
// IP is the tailnet IP of the target.
IP string `json:"ip"`
// FQDN is the full tailnet FQDN of the target.
FQDN string `json:"fqdn"`
}
// PorMap is a mapping between match port on which proxy receives cluster
// traffic and target port where traffic received on match port should be
// fowardded to.
type PortMap struct {
Protocol string `json:"protocol"`
MatchPort uint16 `json:"matchPort"`
TargetPort uint16 `json:"targetPort"`
}
type PortMaps map[PortMap]struct{}
// PortMaps is a list of PortMap structs, however, we want to use it as a set
// with efficient lookups in code. It implements custom JSON marshalling
// methods to convert between being a list in JSON and a set (map with empty
// values) in code.
var _ json.Marshaler = &PortMaps{}
var _ json.Marshaler = PortMaps{}
var _ json.Unmarshaler = &PortMaps{}
func (p *PortMaps) UnmarshalJSON(data []byte) error {
*p = make(map[PortMap]struct{})
var v []PortMap
if err := json.Unmarshal(data, &v); err != nil {
return err
}
for _, pm := range v {
(*p)[pm] = struct{}{}
}
return nil
}
func (p PortMaps) MarshalJSON() ([]byte, error) {
v := make([]PortMap, 0, len(p))
for pm := range p {
v = append(v, pm)
}
return json.Marshal(v)
}
// Status represents the currently configured firewall rules for all egress
// services for a proxy identified by the PodIP.
type Status struct {
PodIPv4 string `json:"podIPv4"`
PodIPv6 string `json:"podIPv6,omitempty"`
// All egress service status keyed by service name.
Services map[string]*ServiceStatus `json:"services"`
}
// ServiceStatus is the currently configured firewall rules for an egress
// service.
type ServiceStatus struct {
Ports PortMaps `json:"ports"`
// TailnetTargetIPs are the tailnet target IPs that were used to
// configure these firewall rules. For a TailnetTarget with IP set, this
// is the same as IP.
TailnetTargetIPs []netip.Addr `json:"tailnetTargetIPs"`
TailnetTarget TailnetTarget `json:"tailnetTarget"`
}