You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
consul/testing/deployer/topology/topology.go

884 lines
20 KiB

// Copyright (c) HashiCorp, Inc.
// SPDX-License-Identifier: BUSL-1.1
package topology
import (
"errors"
"fmt"
"net"
"net/netip"
"reflect"
"sort"
"github.com/hashicorp/consul/api"
"github.com/hashicorp/consul/proto-public/pbresource"
)
type Topology struct {
ID string
// Images controls which specific docker images are used when running this
// node. Non-empty fields here override non-empty fields inherited from the
// general default values from DefaultImages().
Images Images
// Networks is the list of networks to create for this set of clusters.
Networks map[string]*Network
// Clusters defines the list of Consul clusters that should be created, and
// their associated workloads.
Clusters map[string]*Cluster
// Peerings defines the list of pairwise peerings that should be established
// between clusters.
Peerings []*Peering `json:",omitempty"`
// NetworkAreas defines the list of pairwise network area that should be established
// between clusters.
NetworkAreas []*NetworkArea `json:",omitempty"`
}
func (t *Topology) DigestExposedProxyPort(netName string, proxyPort int) (bool, error) {
net, ok := t.Networks[netName]
if !ok {
return false, fmt.Errorf("found output network that does not exist: %s", netName)
}
if net.ProxyPort == proxyPort {
return false, nil
}
net.ProxyPort = proxyPort
// Denormalize for UX.
for _, cluster := range t.Clusters {
for _, node := range cluster.Nodes {
for _, addr := range node.Addresses {
if addr.Network == netName {
addr.ProxyPort = proxyPort
}
}
}
}
return true, nil
}
func (t *Topology) SortedNetworks() []*Network {
var out []*Network
for _, n := range t.Networks {
out = append(out, n)
}
sort.Slice(out, func(i, j int) bool {
return out[i].Name < out[j].Name
})
return out
}
func (t *Topology) SortedClusters() []*Cluster {
var out []*Cluster
for _, c := range t.Clusters {
out = append(out, c)
}
sort.Slice(out, func(i, j int) bool {
return out[i].Name < out[j].Name
})
return out
}
type Config struct {
// Images controls which specific docker images are used when running this
// node. Non-empty fields here override non-empty fields inherited from the
// general default values from DefaultImages().
Images Images
// Networks is the list of networks to create for this set of clusters.
Networks []*Network
// Clusters defines the list of Consul clusters that should be created, and
// their associated workloads.
Clusters []*Cluster
// Peerings defines the list of pairwise peerings that should be established
// between clusters.
Peerings []*Peering
// NetworkAreas defines the list of pairwise NetworkArea that should be established
// between clusters.
NetworkAreas []*NetworkArea
}
func (c *Config) Cluster(name string) *Cluster {
for _, cluster := range c.Clusters {
if cluster.Name == name {
return cluster
}
}
return nil
}
// DisableNode is a no-op if the node is already disabled.
func (c *Config) DisableNode(clusterName string, nid NodeID) (bool, error) {
cluster := c.Cluster(clusterName)
if cluster == nil {
return false, fmt.Errorf("no such cluster: %q", clusterName)
}
for _, n := range cluster.Nodes {
if n.ID() == nid {
if n.Disabled {
return false, nil
}
n.Disabled = true
return true, nil
}
}
return false, fmt.Errorf("expected to find nodeID %q in cluster %q", nid.String(), clusterName)
}
// EnableNode is a no-op if the node is already enabled.
func (c *Config) EnableNode(clusterName string, nid NodeID) (bool, error) {
cluster := c.Cluster(clusterName)
if cluster == nil {
return false, fmt.Errorf("no such cluster: %q", clusterName)
}
for _, n := range cluster.Nodes {
if n.ID() == nid {
if !n.Disabled {
return false, nil
}
n.Disabled = false
return true, nil
}
}
return false, fmt.Errorf("expected to find nodeID %q in cluster %q", nid.String(), clusterName)
}
type Network struct {
Type string // lan/wan ; empty means lan
Name string // logical name
// computed at topology compile
DockerName string
// generated during network-and-tls
Subnet string
IPPool []string `json:"-"`
// generated during network-and-tls
ProxyAddress string `json:",omitempty"`
DNSAddress string `json:",omitempty"`
// filled in from terraform outputs after network-and-tls
ProxyPort int `json:",omitempty"`
}
func (n *Network) IsLocal() bool {
return n.Type == "" || n.Type == "lan"
}
func (n *Network) IsPublic() bool {
return n.Type == "wan"
}
func (n *Network) inheritFromExisting(existing *Network) {
n.Subnet = existing.Subnet
n.IPPool = existing.IPPool
n.ProxyAddress = existing.ProxyAddress
n.DNSAddress = existing.DNSAddress
n.ProxyPort = existing.ProxyPort
}
func (n *Network) IPByIndex(index int) string {
if index >= len(n.IPPool) {
panic(fmt.Sprintf(
"not enough ips on this network to assign index %d: %d",
len(n.IPPool), index,
))
}
return n.IPPool[index]
}
func (n *Network) SetSubnet(subnet string) (bool, error) {
if n.Subnet == subnet {
return false, nil
}
p, err := netip.ParsePrefix(subnet)
if err != nil {
return false, err
}
if !p.IsValid() {
return false, errors.New("not valid")
}
p = p.Masked()
var ipPool []string
addr := p.Addr()
for {
if !p.Contains(addr) {
break
}
ipPool = append(ipPool, addr.String())
addr = addr.Next()
}
ipPool = ipPool[2:] // skip the x.x.x.{0,1}
n.Subnet = subnet
n.IPPool = ipPool
return true, nil
}
// Cluster represents a single standalone install of Consul. This is the unit
// of what is peered when using cluster peering. Older consul installs would
// call this a datacenter.
type Cluster struct {
Name string
NetworkName string // empty assumes same as Name
// Images controls which specific docker images are used when running this
// cluster. Non-empty fields here override non-empty fields inherited from
// the enclosing Topology.
Images Images
// Enterprise marks this cluster as desiring to run Consul Enterprise
// components.
Enterprise bool `json:",omitempty"`
// Nodes is the definition of the nodes (agent-less and agent-ful).
Nodes []*Node
// Partitions is a list of tenancy configurations that should be created
// after the servers come up but before the clients and the rest of the
// topology starts.
//
// Enterprise Only.
Partitions []*Partition `json:",omitempty"`
// Datacenter defaults to "Name" if left unspecified. It lets you possibly
// create multiple peer clusters with identical datacenter names.
Datacenter string
// InitialConfigEntries is a convenience mechanism to have some config
// entries created after the servers start up but before the rest of the
// topology comes up.
InitialConfigEntries []api.ConfigEntry `json:",omitempty"`
// InitialResources is a convenience mechanism to have some resources
// created after the servers start up but before the rest of the topology
// comes up.
InitialResources []*pbresource.Resource `json:",omitempty"`
// TLSVolumeName is the docker volume name containing the various certs
// generated by 'consul tls cert create'
//
// This is generated during the networking phase and is not user specified.
TLSVolumeName string `json:",omitempty"`
// Peerings is a map of peering names to information about that peering in this cluster
//
// Denormalized during compile.
Peerings map[string]*PeerCluster `json:",omitempty"`
// Segments is a map of network segment name and the ports
Segments map[string]int
// DisableGossipEncryption disables gossip encryption on the cluster
// Default is false to enable gossip encryption
DisableGossipEncryption bool `json:",omitempty"`
}
func (c *Cluster) inheritFromExisting(existing *Cluster) {
c.TLSVolumeName = existing.TLSVolumeName
}
type Partition struct {
Name string
Namespaces []string
}
func (c *Cluster) hasPartition(p string) bool {
for _, partition := range c.Partitions {
if partition.Name == p {
return true
}
}
return false
}
func (c *Cluster) PartitionQueryOptionsList() []*api.QueryOptions {
if !c.Enterprise {
return []*api.QueryOptions{{}}
}
var out []*api.QueryOptions
for _, p := range c.Partitions {
out = append(out, &api.QueryOptions{Partition: p.Name})
}
return out
}
func (c *Cluster) ServerNodes() []*Node {
var out []*Node
for _, node := range c.SortedNodes() {
if node.Kind != NodeKindServer || node.Disabled || node.IsNewServer {
continue
}
out = append(out, node)
}
return out
}
func (c *Cluster) ServerByAddr(addr string) *Node {
expect, _, err := net.SplitHostPort(addr)
if err != nil {
return nil
}
for _, node := range c.Nodes {
if node.Kind != NodeKindServer || node.Disabled {
continue
}
if node.LocalAddress() == expect {
return node
}
}
return nil
}
func (c *Cluster) FirstServer() *Node {
for _, node := range c.Nodes {
// TODO: not sure why we check that it has 8500 exposed?
if node.IsServer() && !node.Disabled && node.ExposedPort(8500) > 0 {
return node
}
}
return nil
}
// FirstClient returns the first client agent in the cluster.
// If segment is non-empty, it will return the first client agent in that segment.
func (c *Cluster) FirstClient(segment string) *Node {
for _, node := range c.Nodes {
if node.Kind != NodeKindClient || node.Disabled {
continue
}
if segment == "" {
// return a client agent in default segment
return node
} else {
if node.Segment != nil && node.Segment.Name == segment {
return node
}
}
}
return nil
}
func (c *Cluster) ActiveNodes() []*Node {
var out []*Node
for _, node := range c.Nodes {
if !node.Disabled {
out = append(out, node)
}
}
return out
}
func (c *Cluster) SortedNodes() []*Node {
var out []*Node
out = append(out, c.Nodes...)
kindOrder := map[NodeKind]int{
NodeKindServer: 1,
NodeKindClient: 2,
NodeKindDataplane: 2,
}
sort.Slice(out, func(i, j int) bool {
ni, nj := out[i], out[j]
// servers before clients/dataplanes
ki, kj := kindOrder[ni.Kind], kindOrder[nj.Kind]
if ki < kj {
return true
} else if ki > kj {
return false
}
// lex sort by partition
if ni.Partition < nj.Partition {
return true
} else if ni.Partition > nj.Partition {
return false
}
// lex sort by name
return ni.Name < nj.Name
})
return out
}
func (c *Cluster) WorkloadByID(nid NodeID, sid ID) *Workload {
return c.NodeByID(nid).WorkloadByID(sid)
}
func (c *Cluster) WorkloadsByID(id ID) []*Workload {
id.Normalize()
var out []*Workload
for _, n := range c.Nodes {
for _, wrk := range n.Workloads {
if wrk.ID == id {
out = append(out, wrk)
}
}
}
return out
}
func (c *Cluster) NodeByID(nid NodeID) *Node {
nid.Normalize()
for _, n := range c.Nodes {
if n.ID() == nid {
return n
}
}
panic("node not found: " + nid.String())
}
type Address struct {
Network string
// denormalized at topology compile
Type string
// denormalized at topology compile
DockerNetworkName string
// generated after network-and-tls
IPAddress string
// denormalized from terraform outputs stored in the Network
ProxyPort int `json:",omitempty"`
}
func (a *Address) inheritFromExisting(existing *Address) {
a.IPAddress = existing.IPAddress
a.ProxyPort = existing.ProxyPort
}
func (a Address) IsLocal() bool {
return a.Type == "" || a.Type == "lan"
}
func (a Address) IsPublic() bool {
return a.Type == "wan"
}
type NodeKind string
const (
NodeKindUnknown NodeKind = ""
NodeKindServer NodeKind = "server"
NodeKindClient NodeKind = "client"
NodeKindDataplane NodeKind = "dataplane"
)
type NetworkSegment struct {
Name string
Port int
}
// TODO: rename pod
type Node struct {
Kind NodeKind
Partition string // will be not empty
Name string // logical name
// Images controls which specific docker images are used when running this
// node. Non-empty fields here override non-empty fields inherited from
// the enclosing Cluster.
Images Images
Disabled bool `json:",omitempty"`
Addresses []*Address
Workloads []*Workload
// denormalized at topology compile
Cluster string
Datacenter string
// computed at topology compile
Index int
// IsNewServer is true if the server joins existing cluster
IsNewServer bool
// generated during network-and-tls
TLSCertPrefix string `json:",omitempty"`
// dockerName is computed at topology compile
dockerName string
// usedPorts has keys that are computed at topology compile (internal
// ports) and values initialized to zero until terraform creates the pods
// and extracts the exposed port values from output variables.
usedPorts map[int]int // keys are from compile / values are from terraform output vars
// Meta is the node meta added to the node
Meta map[string]string
// AutopilotConfig of the server agent
AutopilotConfig map[string]string
// Network segment of the agent - applicable to client agent only
Segment *NetworkSegment
// ExtraConfig is the extra config added to the node
ExtraConfig string
}
func (n *Node) DockerName() string {
return n.dockerName
}
func (n *Node) ExposedPort(internalPort int) int {
if internalPort == 0 {
return 0
}
return n.usedPorts[internalPort]
}
func (n *Node) SortedPorts() []int {
var out []int
for internalPort := range n.usedPorts {
out = append(out, internalPort)
}
sort.Ints(out)
return out
}
func (n *Node) inheritFromExisting(existing *Node) {
n.TLSCertPrefix = existing.TLSCertPrefix
merged := existing.usedPorts
for k, vNew := range n.usedPorts {
if _, present := merged[k]; !present {
merged[k] = vNew
}
}
n.usedPorts = merged
}
func (n *Node) String() string {
return n.ID().String()
}
func (n *Node) ID() NodeID {
return NewNodeID(n.Name, n.Partition)
}
func (n *Node) CatalogID() NodeID {
return NewNodeID(n.PodName(), n.Partition)
}
func (n *Node) PodName() string {
return n.dockerName + "-pod"
}
func (n *Node) AddressByNetwork(name string) *Address {
for _, a := range n.Addresses {
if a.Network == name {
return a
}
}
return nil
}
func (n *Node) LocalAddress() string {
for _, a := range n.Addresses {
if a.IsLocal() {
if a.IPAddress == "" {
panic("node has no assigned local address: " + n.Name)
}
return a.IPAddress
}
}
panic("node has no local network")
}
func (n *Node) HasPublicAddress() bool {
for _, a := range n.Addresses {
if a.IsPublic() {
return true
}
}
return false
}
func (n *Node) LocalProxyPort() int {
for _, a := range n.Addresses {
if a.IsLocal() {
if a.ProxyPort > 0 {
return a.ProxyPort
}
panic("node has no assigned local address: " + n.Name)
}
}
panic("node has no local network")
}
func (n *Node) PublicAddress() string {
for _, a := range n.Addresses {
if a.IsPublic() {
if a.IPAddress == "" {
panic("node has no assigned public address")
}
return a.IPAddress
}
}
panic("node has no public network")
}
func (n *Node) PublicProxyPort() int {
for _, a := range n.Addresses {
if a.IsPublic() {
if a.ProxyPort > 0 {
return a.ProxyPort
}
panic("node has no assigned public address")
}
}
panic("node has no public network")
}
func (n *Node) IsServer() bool {
return n.Kind == NodeKindServer
}
func (n *Node) IsAgent() bool {
return n.Kind == NodeKindServer || n.Kind == NodeKindClient
}
func (n *Node) RunsWorkloads() bool {
return n.IsAgent() || n.IsDataplane()
}
func (n *Node) IsDataplane() bool {
return n.Kind == NodeKindDataplane
}
func (n *Node) SortedWorkloads() []*Workload {
var out []*Workload
out = append(out, n.Workloads...)
sort.Slice(out, func(i, j int) bool {
mi := out[i].IsMeshGateway
mj := out[j].IsMeshGateway
if mi && !mi {
return false
} else if !mi && mj {
return true
}
return out[i].ID.Less(out[j].ID)
})
return out
}
// DigestExposedPorts returns true if it was changed.
func (n *Node) DigestExposedPorts(ports map[int]int) bool {
if reflect.DeepEqual(n.usedPorts, ports) {
return false
}
for internalPort := range n.usedPorts {
if v, ok := ports[internalPort]; ok {
n.usedPorts[internalPort] = v
} else {
panic(fmt.Sprintf(
"cluster %q node %q port %d not found in exposed list",
n.Cluster,
n.ID(),
internalPort,
))
}
}
for _, svc := range n.Workloads {
svc.DigestExposedPorts(ports)
}
return true
}
func (n *Node) WorkloadByID(id ID) *Workload {
id.Normalize()
for _, wrk := range n.Workloads {
if wrk.ID == id {
return wrk
}
}
panic("workload not found: " + id.String())
}
type Workload struct {
ID ID
Image string
Port int `json:",omitempty"`
Disabled bool `json:",omitempty"` // TODO
// TODO: expose extra port here?
Meta map[string]string `json:",omitempty"`
// TODO(rb): re-expose this perhaps? Protocol string `json:",omitempty"` // tcp|http (empty == tcp)
CheckHTTP string `json:",omitempty"` // url; will do a GET
CheckTCP string `json:",omitempty"` // addr; will do a socket open/close
EnvoyAdminPort int
ExposedEnvoyAdminPort int `json:",omitempty"`
EnvoyPublicListenerPort int `json:",omitempty"` // agentless
Command []string `json:",omitempty"` // optional
Env []string `json:",omitempty"` // optional
DisableServiceMesh bool `json:",omitempty"`
IsMeshGateway bool `json:",omitempty"`
Upstreams []*Upstream `json:",omitempty"`
// denormalized at topology compile
Node *Node `json:"-"`
}
func (w *Workload) ExposedPort() int {
if w.Node == nil {
panic("ExposedPort cannot be called until after Compile")
}
return w.Node.ExposedPort(w.Port)
}
func (w *Workload) inheritFromExisting(existing *Workload) {
w.ExposedEnvoyAdminPort = existing.ExposedEnvoyAdminPort
}
func (w *Workload) ports() []int {
var out []int
if w.Port > 0 {
out = append(out, w.Port)
}
if w.EnvoyAdminPort > 0 {
out = append(out, w.EnvoyAdminPort)
}
if w.EnvoyPublicListenerPort > 0 {
out = append(out, w.EnvoyPublicListenerPort)
}
for _, us := range w.Upstreams {
if us.LocalPort > 0 {
out = append(out, us.LocalPort)
}
}
return out
}
func (w *Workload) HasCheck() bool {
return w.CheckTCP != "" || w.CheckHTTP != ""
}
func (w *Workload) DigestExposedPorts(ports map[int]int) {
if w.EnvoyAdminPort > 0 {
w.ExposedEnvoyAdminPort = ports[w.EnvoyAdminPort]
} else {
w.ExposedEnvoyAdminPort = 0
}
}
// Validate checks a bunch of stuff intrinsic to the definition of the workload
// itself.
func (w *Workload) Validate() error {
if w.ID.Name == "" {
return fmt.Errorf("service name is required")
}
if w.Image == "" && !w.IsMeshGateway {
return fmt.Errorf("service image is required")
}
if w.Port <= 0 {
return fmt.Errorf("service has invalid port")
}
if w.DisableServiceMesh && w.IsMeshGateway {
return fmt.Errorf("cannot disable service mesh and still run a mesh gateway")
}
if w.DisableServiceMesh && len(w.Upstreams) > 0 {
return fmt.Errorf("cannot disable service mesh and configure upstreams")
}
if w.DisableServiceMesh {
if w.EnvoyAdminPort != 0 {
return fmt.Errorf("cannot use envoy admin port without a service mesh")
}
} else {
if w.EnvoyAdminPort <= 0 {
return fmt.Errorf("envoy admin port is required")
}
}
for _, us := range w.Upstreams {
if us.ID.Name == "" {
return fmt.Errorf("upstream service name is required")
}
if us.LocalPort <= 0 {
return fmt.Errorf("upstream local port is required")
}
if us.LocalAddress != "" {
ip := net.ParseIP(us.LocalAddress)
if ip == nil {
return fmt.Errorf("upstream local address is invalid: %s", us.LocalAddress)
}
}
}
return nil
}
type Upstream struct {
ID ID
LocalAddress string `json:",omitempty"` // defaults to 127.0.0.1
LocalPort int
Peer string `json:",omitempty"`
// TODO: what about mesh gateway mode overrides?
// computed at topology compile
Cluster string `json:",omitempty"`
Peering *PeerCluster `json:",omitempty"` // this will have Link!=nil
}
type Peering struct {
Dialing PeerCluster
Accepting PeerCluster
}
// NetworkArea - a pair of clusters that are peered together
// through network area. PeerCluster type is reused here.
type NetworkArea struct {
Primary PeerCluster
Secondary PeerCluster
}
type PeerCluster struct {
Name string
Partition string
PeerName string // name to call it on this side; defaults if not specified
// computed at topology compile (pointer so it can be empty in json)
Link *PeerCluster `json:",omitempty"`
}
func (c PeerCluster) String() string {
return c.Name + ":" + c.Partition
}
func (p *Peering) String() string {
return "(" + p.Dialing.String() + ")->(" + p.Accepting.String() + ")"
}