2017-10-29 09:45:21 +00:00
|
|
|
package services
|
|
|
|
|
|
|
|
import (
|
2017-11-14 18:11:21 +00:00
|
|
|
"fmt"
|
2018-02-21 01:53:32 +00:00
|
|
|
"strings"
|
2018-01-11 01:00:14 +00:00
|
|
|
"time"
|
2017-11-14 18:11:21 +00:00
|
|
|
|
2018-01-09 22:10:56 +00:00
|
|
|
"context"
|
|
|
|
|
2018-01-11 01:00:14 +00:00
|
|
|
etcdclient "github.com/coreos/etcd/client"
|
2018-02-21 01:53:32 +00:00
|
|
|
"github.com/pkg/errors"
|
2017-10-31 13:55:35 +00:00
|
|
|
"github.com/rancher/rke/docker"
|
2017-10-29 09:45:21 +00:00
|
|
|
"github.com/rancher/rke/hosts"
|
2018-01-09 22:10:56 +00:00
|
|
|
"github.com/rancher/rke/log"
|
2017-12-05 16:55:58 +00:00
|
|
|
"github.com/rancher/types/apis/management.cattle.io/v3"
|
2018-01-11 01:00:14 +00:00
|
|
|
"github.com/sirupsen/logrus"
|
2017-10-29 09:45:21 +00:00
|
|
|
)
|
|
|
|
|
2018-02-13 00:47:56 +00:00
|
|
|
const (
|
|
|
|
EtcdHealthCheckURL = "https://127.0.0.1:2379/health"
|
|
|
|
)
|
|
|
|
|
2018-03-21 17:20:58 +00:00
|
|
|
func RunEtcdPlane(ctx context.Context, etcdHosts []*hosts.Host, etcdProcessHostMap map[*hosts.Host]v3.Process, localConnDialerFactory hosts.DialerFactory, prsMap map[string]v3.PrivateRegistry, updateWorkersOnly bool, alpineImage string) error {
|
2018-01-09 22:10:56 +00:00
|
|
|
log.Infof(ctx, "[%s] Building up Etcd Plane..", ETCDRole)
|
2017-10-29 09:45:21 +00:00
|
|
|
for _, host := range etcdHosts {
|
2018-03-20 12:56:49 +00:00
|
|
|
if updateWorkersOnly {
|
|
|
|
continue
|
|
|
|
}
|
2018-02-26 23:27:54 +00:00
|
|
|
imageCfg, hostCfg, _ := GetProcessConfig(etcdProcessHostMap[host])
|
2018-03-21 17:20:58 +00:00
|
|
|
if err := docker.DoRunContainer(ctx, host.DClient, imageCfg, hostCfg, EtcdContainerName, host.Address, ETCDRole, prsMap); err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
if err := createLogLink(ctx, host, EtcdContainerName, ETCDRole, alpineImage, prsMap); err != nil {
|
2017-10-29 09:45:21 +00:00
|
|
|
return err
|
|
|
|
}
|
|
|
|
}
|
2018-01-09 22:10:56 +00:00
|
|
|
log.Infof(ctx, "[%s] Successfully started Etcd Plane..", ETCDRole)
|
2017-10-29 09:45:21 +00:00
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
2018-01-19 01:48:51 +00:00
|
|
|
func RemoveEtcdPlane(ctx context.Context, etcdHosts []*hosts.Host, force bool) error {
|
2018-01-09 22:10:56 +00:00
|
|
|
log.Infof(ctx, "[%s] Tearing down Etcd Plane..", ETCDRole)
|
2017-11-20 18:08:50 +00:00
|
|
|
for _, host := range etcdHosts {
|
2018-01-09 22:10:56 +00:00
|
|
|
err := docker.DoRemoveContainer(ctx, host.DClient, EtcdContainerName, host.Address)
|
2017-11-20 18:08:50 +00:00
|
|
|
if err != nil {
|
|
|
|
return err
|
|
|
|
}
|
2018-01-19 01:48:51 +00:00
|
|
|
if !host.IsWorker || !host.IsControl || force {
|
|
|
|
// remove unschedulable kubelet on etcd host
|
|
|
|
if err := removeKubelet(ctx, host); err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
if err := removeKubeproxy(ctx, host); err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
if err := removeNginxProxy(ctx, host); err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
if err := removeSidekick(ctx, host); err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-11-20 18:08:50 +00:00
|
|
|
}
|
2018-01-23 23:02:22 +00:00
|
|
|
log.Infof(ctx, "[%s] Successfully tore down Etcd Plane..", ETCDRole)
|
2017-11-20 18:08:50 +00:00
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
2018-02-21 01:53:32 +00:00
|
|
|
func AddEtcdMember(ctx context.Context, toAddEtcdHost *hosts.Host, etcdHosts []*hosts.Host, localConnDialerFactory hosts.DialerFactory, cert, key []byte) error {
|
|
|
|
log.Infof(ctx, "[add/%s] Adding member [etcd-%s] to etcd cluster", ETCDRole, toAddEtcdHost.HostnameOverride)
|
|
|
|
peerURL := fmt.Sprintf("https://%s:2380", toAddEtcdHost.InternalAddress)
|
2018-01-11 01:00:14 +00:00
|
|
|
added := false
|
|
|
|
for _, host := range etcdHosts {
|
2018-02-21 01:53:32 +00:00
|
|
|
if host.Address == toAddEtcdHost.Address {
|
|
|
|
continue
|
|
|
|
}
|
2018-01-16 23:10:14 +00:00
|
|
|
etcdClient, err := getEtcdClient(ctx, host, localConnDialerFactory, cert, key)
|
2018-01-11 01:00:14 +00:00
|
|
|
if err != nil {
|
|
|
|
logrus.Debugf("Failed to create etcd client for host [%s]: %v", host.Address, err)
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
memAPI := etcdclient.NewMembersAPI(etcdClient)
|
|
|
|
if _, err := memAPI.Add(ctx, peerURL); err != nil {
|
2018-02-21 01:53:32 +00:00
|
|
|
logrus.Debugf("Failed to Add etcd member [%s] from host: %v", host.Address, err)
|
2018-01-11 01:00:14 +00:00
|
|
|
continue
|
|
|
|
}
|
|
|
|
added = true
|
|
|
|
break
|
|
|
|
}
|
|
|
|
if !added {
|
2018-02-21 01:53:32 +00:00
|
|
|
return fmt.Errorf("Failed to add etcd member [etcd-%s] to etcd cluster", toAddEtcdHost.HostnameOverride)
|
2018-01-11 01:00:14 +00:00
|
|
|
}
|
2018-02-21 01:53:32 +00:00
|
|
|
log.Infof(ctx, "[add/%s] Successfully Added member [etcd-%s] to etcd cluster", ETCDRole, toAddEtcdHost.HostnameOverride)
|
2018-01-11 01:00:14 +00:00
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
2018-01-16 23:10:14 +00:00
|
|
|
func RemoveEtcdMember(ctx context.Context, etcdHost *hosts.Host, etcdHosts []*hosts.Host, localConnDialerFactory hosts.DialerFactory, cert, key []byte) error {
|
2018-01-11 01:00:14 +00:00
|
|
|
log.Infof(ctx, "[remove/%s] Removing member [etcd-%s] from etcd cluster", ETCDRole, etcdHost.HostnameOverride)
|
|
|
|
var mID string
|
|
|
|
removed := false
|
|
|
|
for _, host := range etcdHosts {
|
2018-01-16 23:10:14 +00:00
|
|
|
etcdClient, err := getEtcdClient(ctx, host, localConnDialerFactory, cert, key)
|
2018-01-11 01:00:14 +00:00
|
|
|
if err != nil {
|
|
|
|
logrus.Debugf("Failed to create etcd client for host [%s]: %v", host.Address, err)
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
memAPI := etcdclient.NewMembersAPI(etcdClient)
|
|
|
|
members, err := memAPI.List(ctx)
|
|
|
|
if err != nil {
|
|
|
|
logrus.Debugf("Failed to list etcd members from host [%s]: %v", host.Address, err)
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
for _, member := range members {
|
|
|
|
if member.Name == fmt.Sprintf("etcd-%s", etcdHost.HostnameOverride) {
|
|
|
|
mID = member.ID
|
|
|
|
break
|
|
|
|
}
|
2017-10-29 09:45:21 +00:00
|
|
|
}
|
2018-01-11 01:00:14 +00:00
|
|
|
if err := memAPI.Remove(ctx, mID); err != nil {
|
|
|
|
logrus.Debugf("Failed to list etcd members from host [%s]: %v", host.Address, err)
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
removed = true
|
|
|
|
break
|
|
|
|
}
|
|
|
|
if !removed {
|
|
|
|
return fmt.Errorf("Failed to delete etcd member [etcd-%s] from etcd cluster", etcdHost.HostnameOverride)
|
2017-10-29 09:45:21 +00:00
|
|
|
}
|
2018-01-11 01:00:14 +00:00
|
|
|
log.Infof(ctx, "[remove/%s] Successfully removed member [etcd-%s] from etcd cluster", ETCDRole, etcdHost.HostnameOverride)
|
|
|
|
return nil
|
2017-10-29 09:45:21 +00:00
|
|
|
}
|
2017-11-15 01:12:33 +00:00
|
|
|
|
2018-03-21 17:20:58 +00:00
|
|
|
func ReloadEtcdCluster(ctx context.Context, readyEtcdHosts []*hosts.Host, localConnDialerFactory hosts.DialerFactory, cert, key []byte, prsMap map[string]v3.PrivateRegistry, etcdProcessHostMap map[*hosts.Host]v3.Process, alpineImage string) error {
|
2018-02-13 00:47:56 +00:00
|
|
|
for host, process := range etcdProcessHostMap {
|
2018-02-26 23:27:54 +00:00
|
|
|
imageCfg, hostCfg, _ := GetProcessConfig(process)
|
2018-01-31 17:50:55 +00:00
|
|
|
if err := docker.DoRunContainer(ctx, host.DClient, imageCfg, hostCfg, EtcdContainerName, host.Address, ETCDRole, prsMap); err != nil {
|
2018-01-11 01:00:14 +00:00
|
|
|
return err
|
|
|
|
}
|
2018-03-21 17:20:58 +00:00
|
|
|
if err := createLogLink(ctx, host, EtcdContainerName, ETCDRole, alpineImage, prsMap); err != nil {
|
|
|
|
return err
|
|
|
|
}
|
2018-01-11 01:00:14 +00:00
|
|
|
}
|
|
|
|
time.Sleep(10 * time.Second)
|
2018-01-16 23:10:14 +00:00
|
|
|
var healthy bool
|
2018-01-11 01:00:14 +00:00
|
|
|
for _, host := range readyEtcdHosts {
|
2018-02-26 23:27:54 +00:00
|
|
|
_, _, healthCheckURL := GetProcessConfig(etcdProcessHostMap[host])
|
2018-02-13 00:47:56 +00:00
|
|
|
if healthy = isEtcdHealthy(ctx, localConnDialerFactory, host, cert, key, healthCheckURL); healthy {
|
2018-01-11 01:00:14 +00:00
|
|
|
break
|
|
|
|
}
|
|
|
|
}
|
2018-01-16 23:10:14 +00:00
|
|
|
if !healthy {
|
|
|
|
return fmt.Errorf("[etcd] Etcd Cluster is not healthy")
|
|
|
|
}
|
2018-01-11 01:00:14 +00:00
|
|
|
return nil
|
2017-11-15 01:12:33 +00:00
|
|
|
}
|
2018-02-21 01:53:32 +00:00
|
|
|
|
|
|
|
func IsEtcdMember(ctx context.Context, etcdHost *hosts.Host, etcdHosts []*hosts.Host, localConnDialerFactory hosts.DialerFactory, cert, key []byte) (bool, error) {
|
|
|
|
var listErr error
|
|
|
|
peerURL := fmt.Sprintf("https://%s:2380", etcdHost.InternalAddress)
|
|
|
|
for _, host := range etcdHosts {
|
|
|
|
if host.Address == etcdHost.Address {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
etcdClient, err := getEtcdClient(ctx, host, localConnDialerFactory, cert, key)
|
|
|
|
if err != nil {
|
|
|
|
listErr = errors.Wrapf(err, "Failed to create etcd client for host [%s]", host.Address)
|
|
|
|
logrus.Debugf("Failed to create etcd client for host [%s]: %v", host.Address, err)
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
memAPI := etcdclient.NewMembersAPI(etcdClient)
|
|
|
|
members, err := memAPI.List(ctx)
|
|
|
|
if err != nil {
|
|
|
|
listErr = errors.Wrapf(err, "Failed to create etcd client for host [%s]", host.Address)
|
|
|
|
logrus.Debugf("Failed to list etcd cluster members [%s]: %v", etcdHost.Address, err)
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
for _, member := range members {
|
|
|
|
if strings.Contains(member.PeerURLs[0], peerURL) {
|
|
|
|
logrus.Infof("[etcd] member [%s] is already part of the etcd cluster", etcdHost.Address)
|
|
|
|
return true, nil
|
|
|
|
}
|
|
|
|
}
|
2018-03-06 22:32:50 +00:00
|
|
|
// reset the list of errors to handle new hosts
|
|
|
|
listErr = nil
|
|
|
|
break
|
2018-02-21 01:53:32 +00:00
|
|
|
}
|
|
|
|
if listErr != nil {
|
|
|
|
return false, listErr
|
|
|
|
}
|
|
|
|
return false, nil
|
|
|
|
}
|