consul/test-integ/peering_commontopo/commontopo.go

449 lines
13 KiB
Go

// Copyright (c) HashiCorp, Inc.
// SPDX-License-Identifier: BUSL-1.1
package peering
import (
"context"
"fmt"
"testing"
"time"
"github.com/hashicorp/consul/api"
"github.com/hashicorp/consul/sdk/testutil/retry"
"github.com/hashicorp/consul/test/integration/consul-container/libs/utils"
"github.com/hashicorp/consul/testing/deployer/sprawl"
"github.com/hashicorp/consul/testing/deployer/sprawl/sprawltest"
"github.com/hashicorp/consul/testing/deployer/topology"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"github.com/hashicorp/consul/test-integ/topoutil"
)
// commonTopo helps create a shareable topology configured to represent
// the common denominator between tests.
//
// Use NewCommonTopo to create.
//
// Compatible suites should implement sharedTopoSuite.
//
// Style:
// - avoid referencing components using strings, prefer IDs like Service ID, etc.
// - avoid passing addresses and ports, etc. Instead, look up components in sprawl.Topology
// by ID to find a concrete type, then pass that to helper functions that know which port to use
// - minimize the surface area of information passed between setup and test code (via members)
// to those that are strictly necessary
type commonTopo struct {
//
Cfg *topology.Config
// shortcuts to corresponding entry in Cfg
DC1 *topology.Cluster
DC2 *topology.Cluster
DC3 *topology.Cluster
// set after Launch. Should be considered read-only
Sprawl *sprawl.Sprawl
Assert *topoutil.Asserter
// track per-DC services to prevent duplicates
services map[string]map[topology.ServiceID]struct{}
}
const agentlessDC = "dc2"
func NewCommonTopo(t *testing.T) *commonTopo {
t.Helper()
ct := commonTopo{}
const nServers = 3
// Make 3-server clusters in dc1 and dc2
// For simplicity, the Name and Datacenter of the clusters are the same.
// dc1 and dc2 should be symmetric.
dc1 := clusterWithJustServers("dc1", nServers)
ct.DC1 = dc1
dc2 := clusterWithJustServers("dc2", nServers)
ct.DC2 = dc2
// dc3 is a failover cluster for both dc1 and dc2
dc3 := clusterWithJustServers("dc3", 1)
// dc3 is only used for certain failover scenarios and does not need tenancies
dc3.Partitions = []*topology.Partition{{Name: "default"}}
ct.DC3 = dc3
injectTenancies(dc1)
injectTenancies(dc2)
// dc3 is only used for certain failover scenarios and does not need tenancies
dc3.Partitions = []*topology.Partition{{Name: "default"}}
ct.services = map[string]map[topology.ServiceID]struct{}{}
for _, dc := range []*topology.Cluster{dc1, dc2, dc3} {
ct.services[dc.Datacenter] = map[topology.ServiceID]struct{}{}
}
peerings := addPeerings(dc1, dc2)
peerings = append(peerings, addPeerings(dc1, dc3)...)
peerings = append(peerings, addPeerings(dc2, dc3)...)
addMeshGateways(dc1)
addMeshGateways(dc2)
addMeshGateways(dc3)
setupGlobals(dc1)
setupGlobals(dc2)
setupGlobals(dc3)
// Build final configuration
ct.Cfg = &topology.Config{
Images: utils.TargetImages(),
Networks: []*topology.Network{
{Name: dc1.Datacenter}, // "dc1" LAN
{Name: dc2.Datacenter}, // "dc2" LAN
{Name: dc3.Datacenter}, // "dc3" LAN
{Name: "wan", Type: "wan"},
},
Clusters: []*topology.Cluster{
dc1,
dc2,
dc3,
},
Peerings: peerings,
}
return &ct
}
// calls sprawltest.Launch followed by s.postLaunchChecks
func (ct *commonTopo) Launch(t *testing.T) {
if ct.Sprawl != nil {
t.Fatalf("Launch must only be called once")
}
ct.Sprawl = sprawltest.Launch(t, ct.Cfg)
ct.Assert = topoutil.NewAsserter(ct.Sprawl)
ct.postLaunchChecks(t)
}
// tests that use Relaunch might want to call this again afterwards
func (ct *commonTopo) postLaunchChecks(t *testing.T) {
t.Logf("TESTING RELATIONSHIPS: \n%s",
topology.RenderRelationships(ct.Sprawl.Topology().ComputeRelationships()),
)
// check that exports line up as expected
for _, clu := range ct.Sprawl.Topology().Clusters {
// expected exports per peer
type key struct {
peer string
partition string
namespace string
}
eepp := map[key]int{}
for _, e := range clu.InitialConfigEntries {
if e.GetKind() == api.ExportedServices {
asExport := e.(*api.ExportedServicesConfigEntry)
// do we care about the partition?
for _, svc := range asExport.Services {
for _, con := range svc.Consumers {
// do we care about con.Partition?
// TODO: surely there is code to normalize this
partition := asExport.Partition
if partition == "" {
partition = "default"
}
namespace := svc.Namespace
if namespace == "" {
namespace = "default"
}
eepp[key{peer: con.Peer, partition: partition, namespace: namespace}] += 1
}
}
}
}
cl := ct.APIClientForCluster(t, clu)
// TODO: these could probably be done in parallel
for k, v := range eepp {
retry.RunWith(&retry.Timer{Timeout: 30 * time.Second, Wait: 500 * time.Millisecond}, t, func(r *retry.R) {
peering, _, err := cl.Peerings().Read(context.Background(), k.peer, utils.CompatQueryOpts(&api.QueryOptions{
Partition: k.partition,
Namespace: k.namespace,
}))
require.Nil(r, err, "reading peering data")
require.NotNilf(r, peering, "peering not found %q", k.peer)
assert.Len(r, peering.StreamStatus.ExportedServices, v, "peering exported services")
})
}
}
if t.Failed() {
t.Fatal("failing fast: post-Launch assertions failed")
}
}
// PeerName is how you'd address a remote dc+partition locally
// as your peer name.
func LocalPeerName(clu *topology.Cluster, partition string) string {
return fmt.Sprintf("peer-%s-%s", clu.Datacenter, partition)
}
// TODO: move these to topology
// TODO: alternatively, delete it: we only use it in one place, to bundle up args
type serviceExt struct {
*topology.Service
Exports []api.ServiceConsumer
Config *api.ServiceConfigEntry
Intentions *api.ServiceIntentionsConfigEntry
}
func (ct *commonTopo) AddServiceNode(clu *topology.Cluster, svc serviceExt) *topology.Node {
clusterName := clu.Name
if _, ok := ct.services[clusterName][svc.ID]; ok {
panic(fmt.Sprintf("duplicate service %q in cluster %q", svc.ID, clusterName))
}
ct.services[clusterName][svc.ID] = struct{}{}
// TODO: inline
serviceHostnameString := func(dc string, id topology.ServiceID) string {
n := id.Name
// prepend <namespace>- and <partition>- if they are not default/empty
// avoids hostname limit of 63 chars in most cases
// TODO: this obviously isn't scalable
if id.Namespace != "default" && id.Namespace != "" {
n = id.Namespace + "-" + n
}
if id.Partition != "default" && id.Partition != "" {
n = id.Partition + "-" + n
}
n = dc + "-" + n
// TODO: experimentally, when this is larger than 63, docker can't start
// the host. confirmed by internet rumor https://gitlab.com/gitlab-org/gitlab-runner/-/issues/27763
if len(n) > 63 {
panic(fmt.Sprintf("docker hostname must not be longer than 63 chars: %q", n))
}
return n
}
nodeKind := topology.NodeKindClient
// TODO: bug in deployer somewhere; it should guard against a KindDataplane node with
// DisableServiceMesh services on it; dataplane is only for service-mesh
if !svc.DisableServiceMesh && clu.Datacenter == agentlessDC {
nodeKind = topology.NodeKindDataplane
}
node := &topology.Node{
Kind: nodeKind,
Name: serviceHostnameString(clu.Datacenter, svc.ID),
Partition: svc.ID.Partition,
Addresses: []*topology.Address{
{Network: clu.Datacenter},
},
Services: []*topology.Service{
svc.Service,
},
Cluster: clusterName,
}
clu.Nodes = append(clu.Nodes, node)
// Export if necessary
if len(svc.Exports) > 0 {
ct.ExportService(clu, svc.ID.Partition, api.ExportedService{
Name: svc.ID.Name,
Namespace: svc.ID.Namespace,
Consumers: svc.Exports,
})
}
// Add any config entries
if svc.Config != nil {
clu.InitialConfigEntries = append(clu.InitialConfigEntries, svc.Config)
}
if svc.Intentions != nil {
clu.InitialConfigEntries = append(clu.InitialConfigEntries, svc.Intentions)
}
return node
}
func (ct *commonTopo) APIClientForCluster(t *testing.T, clu *topology.Cluster) *api.Client {
cl, err := ct.Sprawl.APIClientForCluster(clu.Name, "")
require.NoError(t, err)
return cl
}
// ExportService looks for an existing ExportedServicesConfigEntry for the given partition
// and inserts svcs. If none is found, it inserts a new ExportedServicesConfigEntry.
func (ct *commonTopo) ExportService(clu *topology.Cluster, partition string, svcs ...api.ExportedService) {
var found bool
for _, ce := range clu.InitialConfigEntries {
// We check Name because it must be "default" in CE whereas Partition will be "".
if ce.GetKind() == api.ExportedServices && ce.GetName() == partition {
found = true
e := ce.(*api.ExportedServicesConfigEntry)
e.Services = append(e.Services, svcs...)
}
}
if !found {
clu.InitialConfigEntries = append(clu.InitialConfigEntries,
&api.ExportedServicesConfigEntry{
Name: partition, // this NEEDs to be "default" in CE
Partition: ConfigEntryPartition(partition),
Services: svcs,
},
)
}
}
func (ct *commonTopo) ClusterByDatacenter(t *testing.T, name string) *topology.Cluster {
t.Helper()
for _, clu := range ct.Cfg.Clusters {
if clu.Datacenter == name {
return clu
}
}
t.Fatalf("cluster %q not found", name)
return nil
}
// Since CE config entries do not contain the partition field,
// this func converts default partition to empty string.
func ConfigEntryPartition(p string) string {
if p == "default" {
return "" // make this CE friendly
}
return p
}
// DisableNode is a no-op if the node is already disabled.
func DisableNode(t *testing.T, cfg *topology.Config, clusterName string, nid topology.NodeID) *topology.Config {
changed, err := cfg.DisableNode(clusterName, nid)
require.NoError(t, err)
if changed {
t.Logf("disabling node %s in cluster %s", nid.String(), clusterName)
}
return cfg
}
// EnableNode is a no-op if the node is already enabled.
func EnableNode(t *testing.T, cfg *topology.Config, clusterName string, nid topology.NodeID) *topology.Config {
changed, err := cfg.EnableNode(clusterName, nid)
require.NoError(t, err)
if changed {
t.Logf("enabling node %s in cluster %s", nid.String(), clusterName)
}
return cfg
}
func setupGlobals(clu *topology.Cluster) {
for _, part := range clu.Partitions {
clu.InitialConfigEntries = append(clu.InitialConfigEntries,
&api.ProxyConfigEntry{
Name: api.ProxyConfigGlobal,
Kind: api.ProxyDefaults,
Partition: ConfigEntryPartition(part.Name),
MeshGateway: api.MeshGatewayConfig{
// Although we define service-defaults for most upstreams in
// this test suite, failover tests require a global mode
// because the default for peered targets is MeshGatewayModeRemote.
Mode: api.MeshGatewayModeLocal,
},
},
&api.MeshConfigEntry{
Peering: &api.PeeringMeshConfig{
PeerThroughMeshGateways: true,
},
},
)
}
}
// addMeshGateways adds a mesh gateway for every partition in the cluster.
// Assumes that the LAN network name is equal to datacenter name.
func addMeshGateways(c *topology.Cluster) {
nodeKind := topology.NodeKindClient
if c.Datacenter == agentlessDC {
nodeKind = topology.NodeKindDataplane
}
for _, p := range c.Partitions {
c.Nodes = topology.MergeSlices(c.Nodes, topoutil.NewTopologyMeshGatewaySet(
nodeKind,
p.Name,
fmt.Sprintf("%s-%s-mgw", c.Name, p.Name),
1,
[]string{c.Datacenter, "wan"},
nil,
))
}
}
func clusterWithJustServers(name string, numServers int) *topology.Cluster {
return &topology.Cluster{
Enterprise: utils.IsEnterprise(),
Name: name,
Datacenter: name,
Nodes: topoutil.NewTopologyServerSet(
name+"-server",
numServers,
[]string{name},
nil,
),
}
}
func addPeerings(acc *topology.Cluster, dial *topology.Cluster) []*topology.Peering {
peerings := []*topology.Peering{}
for _, accPart := range acc.Partitions {
for _, dialPart := range dial.Partitions {
peerings = append(peerings, &topology.Peering{
Accepting: topology.PeerCluster{
Name: acc.Datacenter,
Partition: accPart.Name,
PeerName: LocalPeerName(dial, dialPart.Name),
},
Dialing: topology.PeerCluster{
Name: dial.Datacenter,
Partition: dialPart.Name,
PeerName: LocalPeerName(acc, accPart.Name),
},
})
}
}
return peerings
}
func injectTenancies(clu *topology.Cluster) {
if !utils.IsEnterprise() {
clu.Partitions = []*topology.Partition{
{
Name: "default",
Namespaces: []string{
"default",
},
},
}
return
}
for _, part := range []string{"default", "part1"} {
clu.Partitions = append(clu.Partitions,
&topology.Partition{
Name: part,
Namespaces: []string{
"default",
"ns1",
},
},
)
}
}
// Deprecated: topoutil.NewFortioServiceWithDefaults
func NewFortioServiceWithDefaults(
cluster string,
sid topology.ServiceID,
mut func(s *topology.Service),
) *topology.Service {
return topoutil.NewFortioServiceWithDefaults(cluster, sid, topology.NodeVersionV1, mut)
}