mirror of https://github.com/status-im/consul.git
143 lines
4.1 KiB
Go
143 lines
4.1 KiB
Go
// Copyright (c) HashiCorp, Inc.
|
|
// SPDX-License-Identifier: BUSL-1.1
|
|
|
|
package proxystateconverter
|
|
|
|
import (
|
|
"fmt"
|
|
|
|
"github.com/hashicorp/consul/agent/connect"
|
|
"github.com/hashicorp/consul/agent/proxycfg"
|
|
"github.com/hashicorp/consul/agent/structs"
|
|
"github.com/hashicorp/consul/envoyextensions/xdscommon"
|
|
"github.com/hashicorp/consul/proto-public/pbmesh/v2beta1/pbproxystate"
|
|
)
|
|
|
|
type discoChainTargets struct {
|
|
baseClusterName string
|
|
targets []targetInfo
|
|
failover bool
|
|
failoverPolicy structs.ServiceResolverFailoverPolicy
|
|
}
|
|
|
|
type targetInfo struct {
|
|
TargetID string
|
|
TransportSocket *pbproxystate.TransportSocket
|
|
SNI string
|
|
RootPEMs string
|
|
SpiffeIDs []string
|
|
Region *string
|
|
|
|
PrioritizeByLocality *structs.DiscoveryPrioritizeByLocality
|
|
}
|
|
|
|
type discoChainTargetGroup struct {
|
|
Targets []targetInfo
|
|
ClusterName string
|
|
}
|
|
|
|
func (ft discoChainTargets) groupedTargets() ([]discoChainTargetGroup, error) {
|
|
var targetGroups []discoChainTargetGroup
|
|
|
|
if !ft.failover {
|
|
targetGroups = append(targetGroups, discoChainTargetGroup{
|
|
ClusterName: ft.baseClusterName,
|
|
Targets: ft.targets,
|
|
})
|
|
return targetGroups, nil
|
|
}
|
|
|
|
switch ft.failoverPolicy.Mode {
|
|
case "sequential", "":
|
|
return ft.sequential()
|
|
case "order-by-locality":
|
|
return ft.orderByLocality()
|
|
default:
|
|
return targetGroups, fmt.Errorf("unexpected failover policy")
|
|
}
|
|
}
|
|
|
|
func (s *Converter) mapDiscoChainTargets(cfgSnap *proxycfg.ConfigSnapshot, chain *structs.CompiledDiscoveryChain, node *structs.DiscoveryGraphNode, upstreamConfig structs.UpstreamConfig, forMeshGateway bool) (discoChainTargets, error) {
|
|
failoverTargets := discoChainTargets{}
|
|
|
|
if node.Resolver == nil {
|
|
return discoChainTargets{}, fmt.Errorf("impossible to process a non-resolver node")
|
|
}
|
|
|
|
primaryTargetID := node.Resolver.Target
|
|
upstreamsSnapshot, err := cfgSnap.ToConfigSnapshotUpstreams()
|
|
if err != nil && !forMeshGateway {
|
|
return discoChainTargets{}, err
|
|
}
|
|
|
|
failoverTargets.baseClusterName = s.getTargetClusterName(upstreamsSnapshot, chain, primaryTargetID, forMeshGateway)
|
|
|
|
tids := []string{primaryTargetID}
|
|
failover := node.Resolver.Failover
|
|
if failover != nil && !forMeshGateway {
|
|
tids = append(tids, failover.Targets...)
|
|
failoverTargets.failover = true
|
|
if failover.Policy == nil {
|
|
failoverTargets.failoverPolicy = structs.ServiceResolverFailoverPolicy{}
|
|
} else {
|
|
failoverTargets.failoverPolicy = *failover.Policy
|
|
}
|
|
}
|
|
|
|
for _, tid := range tids {
|
|
target := chain.Targets[tid]
|
|
targetUID := proxycfg.NewUpstreamIDFromTargetID(tid)
|
|
ti := targetInfo{TargetID: tid, PrioritizeByLocality: target.PrioritizeByLocality}
|
|
|
|
configureTLS := true
|
|
if forMeshGateway {
|
|
// We only initiate TLS if we're doing an L7 proxy.
|
|
configureTLS = structs.IsProtocolHTTPLike(upstreamConfig.Protocol)
|
|
}
|
|
|
|
if !configureTLS {
|
|
failoverTargets.targets = append(failoverTargets.targets, ti)
|
|
continue
|
|
}
|
|
|
|
if targetUID.Peer != "" {
|
|
tbs, _ := upstreamsSnapshot.UpstreamPeerTrustBundles.Get(targetUID.Peer)
|
|
|
|
peerMeta, found := upstreamsSnapshot.UpstreamPeerMeta(targetUID)
|
|
if !found {
|
|
s.Logger.Warn("failed to fetch upstream peering metadata", "target", targetUID)
|
|
continue
|
|
}
|
|
ti.SNI = peerMeta.PrimarySNI()
|
|
ti.SpiffeIDs = peerMeta.SpiffeID
|
|
region := target.Locality.GetRegion()
|
|
ti.Region = ®ion
|
|
ti.RootPEMs = tbs.ConcatenatedRootPEMs()
|
|
} else {
|
|
ti.SNI = target.SNI
|
|
ti.RootPEMs = cfgSnap.RootPEMs()
|
|
ti.SpiffeIDs = []string{connect.SpiffeIDService{
|
|
Host: cfgSnap.Roots.TrustDomain,
|
|
Namespace: target.Namespace,
|
|
Partition: target.Partition,
|
|
Datacenter: target.Datacenter,
|
|
Service: target.Service,
|
|
}.URI().String()}
|
|
}
|
|
failoverTargets.targets = append(failoverTargets.targets, ti)
|
|
}
|
|
|
|
return failoverTargets, nil
|
|
}
|
|
|
|
func (ft discoChainTargets) sequential() ([]discoChainTargetGroup, error) {
|
|
var targetGroups []discoChainTargetGroup
|
|
for i, t := range ft.targets {
|
|
targetGroups = append(targetGroups, discoChainTargetGroup{
|
|
ClusterName: fmt.Sprintf("%s%d~%s", xdscommon.FailoverClusterNamePrefix, i, ft.baseClusterName),
|
|
Targets: []targetInfo{t},
|
|
})
|
|
}
|
|
return targetGroups, nil
|
|
}
|