policy-generator/cmd/policygen/cluster.go
Erik Brakkee 2373f428bb now processing relabeling rules for prometheus
obtaining all required monitoring endpoints now seems to work.
2025-01-29 21:08:42 +01:00

514 lines
15 KiB
Go

package main
import (
"context"
"fmt"
v1 "k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/labels"
"k8s.io/apimachinery/pkg/runtime/schema"
"k8s.io/apimachinery/pkg/util/json"
"k8s.io/client-go/dynamic"
"k8s.io/client-go/kubernetes"
"k8s.io/klog/v2"
"log"
"os"
"slices"
"strconv"
"strings"
)
type ServiceMonitor struct {
metav1.TypeMeta `json:",inline"`
metav1.ObjectMeta `json:"metadata,omitempty"`
Spec ServiceMonitorSpec `json:"spec"`
}
type ServiceMonitorSpec struct {
NamespaceSelector NamespaceSelector `json:"namespaceSelector"`
Selector metav1.LabelSelector `json:"selector"`
Endpoints []Endpoint `json:"endpoints"`
}
type Endpoint struct {
Path string `json:"path,omitempty"`
Port string `json:"port"`
TargetPort string `json:"targetPort"`
Scheme string `json:"scheme,omitempty"`
Relabelings []RelabelConfig `json:"relabelings,omitempty"`
}
type PodMonitor struct {
metav1.TypeMeta `json:",inline"`
metav1.ObjectMeta `json:"metadata,omitempty"`
Spec PodMonitorSpec `json:"spec"`
}
type PodMonitorSpec struct {
NamespaceSelector NamespaceSelector `json:"namespaceSelector"`
Selector metav1.LabelSelector `json:"selector"`
PodMetricsEndpoints []Endpoint `json:"podMetricsEndpoints"`
}
type NamespaceSelector struct {
Any bool `json:"any,omitempty"`
MatchNames []string `json:"matchNames,omitempty"`
}
type RelabelConfig struct {
Action string `json:"action,omitempty"`
SourceLabels []string `json:"sourceLabels,omitempty"`
Separator string `json:"separator,omitempty"`
TargetLabel string `json:"targetLabel,omitempty"`
Regex string `json:"regex,omitempty"`
Modulus uint64 `json:"modulus,omitempty"`
Replacement string `json:"replacement,omitempty"`
}
type Cluster struct {
namespaces map[string]v1.Namespace
clientset *kubernetes.Clientset
dynClient *dynamic.DynamicClient
// map of namespace to list of all pods
pods map[string][]v1.Pod
services map[string][]v1.Service
// map of namespace/podname ->: ScrapingEndpoints
endpoints map[string]MonitoringEndpoints
}
type MonitoringEndpoint struct {
Port string
Path string
}
type MonitoringEndpoints []MonitoringEndpoint
func (s MonitoringEndpoint) CompareTo(e MonitoringEndpoint) int {
if res := strings.Compare(s.Port, e.Port); res != 0 {
return res
}
return strings.Compare(s.Path, e.Path)
}
func (s *MonitoringEndpoints) Merge(endpoints MonitoringEndpoints) {
*s = append(*s, endpoints...)
slices.SortFunc(*s, func(a, b MonitoringEndpoint) int {
return a.CompareTo(b)
})
*s = slices.CompactFunc(*s, func(a, b MonitoringEndpoint) bool {
return a.CompareTo(b) == 0
})
}
func NewCluster(clientset *kubernetes.Clientset, dynClient *dynamic.DynamicClient) (*Cluster, error) {
cluster := &Cluster{
clientset: clientset,
dynClient: dynClient,
namespaces: make(map[string]v1.Namespace),
pods: make(map[string][]v1.Pod),
services: make(map[string][]v1.Service),
endpoints: make(map[string]MonitoringEndpoints),
}
nslist, err := cluster.clientset.CoreV1().Namespaces().List(context.Background(), metav1.ListOptions{})
if err != nil {
return nil, err
}
fmt.Fprintf(os.Stderr, "Getting pods, services\n")
for _, ns := range nslist.Items {
cluster.namespaces[ns.Name] = ns
}
podList, err := cluster.clientset.CoreV1().Pods("").List(context.Background(), metav1.ListOptions{})
if err != nil {
return nil, err
}
for _, pod := range podList.Items {
cluster.pods[pod.Namespace] = append(cluster.pods[pod.Namespace], pod)
}
svcList, err := cluster.clientset.CoreV1().Services("").List(context.Background(), metav1.ListOptions{})
if err != nil {
return nil, err
}
for _, svc := range svcList.Items {
cluster.services[svc.Namespace] = append(cluster.services[svc.Namespace], svc)
}
err = cluster.processServiceMonitors()
if err != nil {
return nil, err
}
err = cluster.procewssPodMonitors()
if err != nil {
return nil, err
}
podkeys := MapKeys(cluster.endpoints)
slices.Sort(podkeys)
for _, pod := range podkeys {
monitors := cluster.endpoints[pod]
slices.SortFunc(monitors,
func(m1, m2 MonitoringEndpoint) int {
return m1.CompareTo(m2)
})
monitors = slices.CompactFunc(monitors,
func(m1, m2 MonitoringEndpoint) bool {
return m1.CompareTo(m2) == 0
})
cluster.endpoints[pod] = monitors
fmt.Fprintf(os.Stderr, "monitoring endpoint: %s: %v\n", pod, monitors)
}
return cluster, nil
}
func (c *Cluster) processPodMonitor(podMonitor PodMonitor) error {
selector, err := metav1.LabelSelectorAsSelector(&podMonitor.Spec.Selector)
if err != nil {
return err
}
namespaces := c.getNamespacesForMonitor(podMonitor.Spec.NamespaceSelector, podMonitor.Namespace)
for _, endpoint := range podMonitor.Spec.PodMetricsEndpoints {
for _, relabeling := range endpoint.Relabelings {
if relabeling.Action != "keep" {
continue
}
fmt.Fprintf(os.Stderr, "WARNING: podmonitor %s/%s contains relabeling which are currently not processed\n",
podMonitor.Namespace, podMonitor.Name)
}
}
for _, namespace := range namespaces {
pods := slices.DeleteFunc(slices.Clone(c.pods[namespace]), func(pod v1.Pod) bool {
return !selector.Matches(labels.Set(pod.Labels))
})
for _, pod := range pods {
podkey := pod.Namespace + "/" + pod.Name
klog.V(3).Infof(" match %s/%s\n", pod.Namespace, pod.Name)
for _, container := range pod.Spec.Containers {
for _, endpoint := range podMonitor.Spec.PodMetricsEndpoints {
ep, err := c.getEndpointSpec(endpoint, &pod, &container, container.Ports)
if err != nil {
// TODO generate warning?
// this container did not match
continue
}
c.endpoints[podkey] = append(c.endpoints[podkey], ep...)
}
}
if len(c.endpoints[podkey]) > 0 {
klog.V(3).Infof(" POD %s/%s :%v\n", pod.Namespace, pod.Name,
c.endpoints[podkey])
}
}
}
return nil
}
func (c *Cluster) getNamespacesForMonitor(namespaceSelector NamespaceSelector, defaultNamespace string) []string {
namespaces := MapKeys(c.namespaces)
if !namespaceSelector.Any {
namespaces = namespaceSelector.MatchNames
if len(namespaces) == 0 {
namespaces = []string{defaultNamespace}
}
}
return namespaces
}
func (c *Cluster) processServiceMonitor(serviceMonitor ServiceMonitor) error {
selector, err := metav1.LabelSelectorAsSelector(&serviceMonitor.Spec.Selector)
if err != nil {
return err
}
namespaces := c.getNamespacesForMonitor(serviceMonitor.Spec.NamespaceSelector,
serviceMonitor.Namespace)
for _, endpoint := range serviceMonitor.Spec.Endpoints {
for _, relabeling := range endpoint.Relabelings {
if relabeling.Action != "keep" {
continue
}
fmt.Fprintf(os.Stderr, "WARNING: servicemonitor %s/%s contains relabeling which are currently not processed\n",
serviceMonitor.Namespace, serviceMonitor.Name)
}
}
for _, namespace := range namespaces {
svcs := slices.DeleteFunc(slices.Clone(c.services[namespace]), func(svc v1.Service) bool {
return !selector.Matches(labels.Set(svc.Labels))
})
for _, svc := range svcs {
klog.V(3).Infof(" match %s/%s\n", svc.Namespace, svc.Name)
if svc.Spec.Selector == nil {
fmt.Fprintf(os.Stderr, " service does not have a selector, skippingit, probably configured using endpoints\n")
continue
}
// Now we need to obtain the pod port.
// 1. Port specified
// service targetPort is used (equal to port if not specified)
// a. lookup pods matched by the service using the service selector
// b. path is /metics or the metric defined in the service selector
// c. port, path is defined and poss are known.
// 2. Target port specified
// a, b, c see above.
svcSelector := labels.SelectorFromSet(labels.Set(svc.Spec.Selector))
// find pods
pods := slices.DeleteFunc(slices.Clone(c.pods[svc.Namespace]), func(pod v1.Pod) bool {
return !svcSelector.Matches(labels.Set(pod.Labels))
})
for _, endpoint := range serviceMonitor.Spec.Endpoints {
for _, pod := range pods {
podkey := pod.Namespace + "/" + pod.Name
for _, container := range pod.Spec.Containers {
ep, err := c.getEndpointSpec(endpoint, &pod, &container, container.Ports)
if err != nil {
// TODO: generate warning?
// this wervice did not match
}
c.endpoints[podkey] = append(c.endpoints[podkey], ep...)
}
if len(c.endpoints[podkey]) > 0 {
klog.V(3).Infof(" POD %s/%s :%v\n", pod.Namespace, pod.Name,
c.endpoints[podkey])
}
}
}
}
}
return nil
}
type GeneralizedPort struct {
Name string
TargetPort string
Protocol string
}
func MapServicePorts(ports []v1.ServicePort) []GeneralizedPort {
return Map(ports, func(p v1.ServicePort) GeneralizedPort {
return GeneralizedPort{
Name: p.Name,
TargetPort: p.TargetPort.String(),
Protocol: string(p.Protocol),
}
})
}
func MapConteinerPorts(ports []v1.ContainerPort) []GeneralizedPort {
return Map(ports, func(p v1.ContainerPort) GeneralizedPort {
return GeneralizedPort{
Name: p.Name,
TargetPort: strconv.Itoa(int(p.ContainerPort)),
Protocol: string(p.Protocol),
}
})
}
func (c *Cluster) getEndpointSpec(endpoint Endpoint, pod *v1.Pod, container *v1.Container, ports []v1.ContainerPort) ([]MonitoringEndpoint, error) {
klog.V(4).Infof(" analyzing %s/%s container %s/%s %v\n",
endpoint.Port, endpoint.TargetPort, pod.Name, container.Name, ports)
res := []MonitoringEndpoint{}
portLoop:
for _, port := range ports {
targetPort := ""
switch {
case endpoint.Port != "":
//fmt.Fprintf(os.Stderr, "Checking %v %s\n", port.Port, endpoint.Port)
if port.Name != endpoint.Port {
continue
}
case endpoint.TargetPort != "":
if strconv.Itoa(int(port.ContainerPort)) != endpoint.TargetPort {
continue
}
}
if targetPort == "" {
if port.Name != "" {
targetPort = port.Name
} else {
targetPort = strconv.Itoa(int(port.ContainerPort))
}
}
// check relabeling configs
for _, relabeling := range endpoint.Relabelings {
match, err := MatchRelabeling(relabeling, pod, container, &port)
if err != nil {
return nil, err
}
if !match {
continue portLoop
}
}
// port matches all criteria
path := "/metrics"
if endpoint.Path != "" {
path = endpoint.Path
}
ep := MonitoringEndpoint{
Port: targetPort,
Path: path,
}
res = append(res, ep)
}
return res, nil
}
func (c *Cluster) processServiceMonitors() error {
fmt.Fprintf(os.Stderr, "Getting service monitors\n")
serviceMonitorGVR := schema.GroupVersionResource{
Group: "monitoring.coreos.com",
Version: "v1",
Resource: "servicemonitors",
}
list, err := c.dynClient.Resource(serviceMonitorGVR).
Namespace("").
List(context.Background(), metav1.ListOptions{})
if err != nil {
return err
}
for _, item := range list.Items {
var sm ServiceMonitor
bytes, _ := json.Marshal(item.Object)
if err := json.Unmarshal(bytes, &sm); err != nil {
return fmt.Errorf("Error unmarshalling %s: %w", bytes, err)
}
klog.V(2).Infof("Found servicemonitor: %s/%s\n", sm.ObjectMeta.Namespace, sm.ObjectMeta.Name)
if err := c.processServiceMonitor(sm); err != nil {
return err
}
}
return nil
}
func (c *Cluster) procewssPodMonitors() error {
fmt.Fprintf(os.Stderr, "Getting pod monitors\n")
podMonitorGVR := schema.GroupVersionResource{
Group: "monitoring.coreos.com",
Version: "v1",
Resource: "podmonitors",
}
list, err := c.dynClient.Resource(podMonitorGVR).
Namespace("").
List(context.Background(), metav1.ListOptions{})
if err != nil {
return err
}
for _, item := range list.Items {
var pm PodMonitor
bytes, _ := json.Marshal(item.Object)
if err := json.Unmarshal(bytes, &pm); err != nil {
return fmt.Errorf("Error unmarshalling %s: %w", bytes, err)
}
klog.V(2).Infof("Found podmonitor: %s/%s\n", pm.ObjectMeta.Namespace, pm.ObjectMeta.Name)
if err := c.processPodMonitor(pm); err != nil {
return err
}
}
return nil
}
func (c *Cluster) Pods(application *Application) []v1.Pod {
selector, err := metav1.LabelSelectorAsSelector(application.Selector())
if err != nil {
log.Fatalf("Error creating selector: %v", err)
}
pods := c.pods[application.Namespace.Name]
pods = slices.DeleteFunc(slices.Clone(pods), func(pod v1.Pod) bool {
return !selector.Matches(labels.Set(pod.Labels))
})
return pods
}
func (c *Cluster) ServiceAccounts(application *Application) []string {
var res []string
for _, pod := range c.Pods(application) {
if !slices.Contains(res, pod.Spec.ServiceAccountName) {
res = append(res, pod.Spec.ServiceAccountName)
}
}
return res
}
func (c *Cluster) OwnerReferences(application *Application) []string {
var ownerReferences []string
for _, pod := range c.Pods(application) {
//log.Printf(" %s %v", pod.Name, pod.OwnerReferences)
for _, ownerReference := range pod.OwnerReferences {
owner := ownerReference.Kind + "/" + ownerReference.Name
if !slices.Contains(ownerReferences, owner) {
ownerReferences = append(ownerReferences, owner)
}
}
}
return ownerReferences
}
func (c *Cluster) IsLinkerdEnabled(application *Application) bool {
pods := c.Pods(application)
ndisabled := 0
for _, pod := range pods {
if pod.Annotations["linkerd.io/inject"] == "enabled" {
return true
}
if pod.Annotations["linkerd.io/inject"] == "disabled" {
ndisabled++
}
}
if ndisabled == len(pods) {
return false
}
ns := c.namespaces[application.Namespace.Name]
return ns.Annotations["linkerd.io/inject"] == "enabled"
}
func (c *Cluster) NamespaceLIst() []v1.Namespace {
return MapValues(c.namespaces)
}
func (c *Cluster) Namespace(name string) v1.Namespace {
return c.namespaces[name]
}
func (c *Cluster) PodList(namespace string) []v1.Pod {
return c.pods[namespace]
}
func (c *Cluster) PortNumbers(application *Application) []Port {
if !c.IsLinkerdEnabled(application) {
return nil
}
tcpPorts := make(map[int]Port)
udpPorts := make(map[int]Port)
for _, pod := range c.Pods(application) {
for _, container := range pod.Spec.Containers {
if container.Name == "linkerd-proxy" {
continue
}
for _, port := range container.Ports {
switch port.Protocol {
case "TCP":
tcpPorts[int(port.ContainerPort)] = Port{
Port: strconv.Itoa(int(port.ContainerPort)),
Protocol: string(port.Protocol),
}
case "UDP":
udpPorts[int(port.ContainerPort)] = Port{
Port: strconv.Itoa(int(port.ContainerPort)),
Protocol: string(port.Protocol),
}
default:
panic(fmt.Sprintf("Unknown port type for pod %s/%s: %s",
pod.Namespace, pod.Name, port.Protocol))
}
}
}
}
res := MapValues(tcpPorts)
res = append(res, MapValues(udpPorts)...)
return res
}