heimdall/server/node.go

422 lines
10 KiB
Go
Raw Normal View History

/*
Copyright 2019 Stellar Project
Permission is hereby granted, free of charge, to any person obtaining a copy of
this software and associated documentation files (the "Software"), to deal in the
Software without restriction, including without limitation the rights to use, copy,
modify, merge, publish, distribute, sublicense, and/or sell copies of the Software,
and to permit persons to whom the Software is furnished to do so, subject to the
following conditions:
The above copyright notice and this permission notice shall be included in all copies
or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR
PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE
FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
package server
import (
"context"
"fmt"
"net/url"
"sort"
"strings"
"time"
"github.com/gogo/protobuf/proto"
"github.com/gomodule/redigo/redis"
"github.com/pkg/errors"
"github.com/sirupsen/logrus"
v1 "github.com/stellarproject/heimdall/api/v1"
)
// Nodes returns a list of known nodes
func (s *Server) Nodes(ctx context.Context, req *v1.NodesRequest) (*v1.NodesResponse, error) {
nodes, err := s.getNodes(ctx)
if err != nil {
return nil, err
}
return &v1.NodesResponse{
Nodes: nodes,
}, nil
}
func (s *Server) getNodes(ctx context.Context) ([]*v1.Node, error) {
nodeKeys, err := redis.Strings(s.local(ctx, "KEYS", s.getNodeKey("*")))
if err != nil {
return nil, err
}
var nodes []*v1.Node
for _, nodeKey := range nodeKeys {
data, err := redis.Bytes(s.local(ctx, "GET", nodeKey))
if err != nil {
return nil, err
}
var node v1.Node
if err := proto.Unmarshal(data, &node); err != nil {
return nil, err
}
nodes = append(nodes, &node)
}
return nodes, nil
}
func (s *Server) getNode(ctx context.Context, id string) (*v1.Node, error) {
data, err := redis.Bytes(s.local(ctx, "GET", s.getNodeKey(id)))
if err != nil {
return nil, err
}
var node v1.Node
if err := proto.Unmarshal(data, &node); err != nil {
return nil, err
}
return &node, nil
}
func (s *Server) configureNode() error {
ctx := context.Background()
nodes, err := s.getNodes(ctx)
if err != nil {
return err
}
// attempt to connect to existing
if len(nodes) > 0 {
for _, node := range nodes {
// ignore self
if node.ID == s.cfg.ID {
continue
}
logrus.Infof("attempting to join existing node %s", node.Addr)
c, err := s.getClient(node.Addr)
if err != nil {
logrus.Warn(err)
continue
}
r, err := c.Join(&v1.JoinRequest{
ID: s.cfg.ID,
ClusterKey: s.cfg.ClusterKey,
GRPCAddress: s.cfg.GRPCAddress,
EndpointIP: s.cfg.EndpointIP,
EndpointPort: uint64(s.cfg.EndpointPort),
InterfaceName: s.cfg.InterfaceName,
})
if err != nil {
c.Close()
logrus.Warn(err)
continue
}
// start tunnel
if err := s.updatePeerConfig(ctx, r.Node, r.Peers); err != nil {
return err
}
// wait for tunnel to come up
logrus.Infof("waiting for master %s", r.Master.ID)
if err := s.waitForMaster(ctx, r.Master); err != nil {
return err
}
if err := s.joinMaster(r.Master); err != nil {
c.Close()
logrus.Warn(err)
continue
}
logrus.Infof("waiting for redis sync with %s", r.Master.ID)
if err := s.waitForRedisSync(ctx); err != nil {
return err
}
go s.replicaMonitor()
return nil
}
}
data, err := redis.Bytes(s.local(context.Background(), "GET", masterKey))
if err != nil {
if err != redis.ErrNil {
return err
}
logrus.Infof("starting as master id=%s", s.cfg.ID)
if _, err := s.local(context.Background(), "REPLICAOF", "NO", "ONE"); err != nil {
return err
}
// start server heartbeat
logrus.Debug("starting master heartbeat")
go s.masterHeartbeat()
// reset replica settings when promoting to master
logrus.Debug("disabling replica status")
s.disableReplica()
return nil
}
var master v1.Master
if err := proto.Unmarshal(data, &master); err != nil {
return err
}
// TODO: start tunnel
if err := s.joinMaster(&master); err != nil {
return err
}
go s.replicaMonitor()
return nil
}
func (s *Server) disableReplica() error {
p, err := getPool(s.cfg.RedisURL)
if err != nil {
return err
}
s.wpool = p
// signal replica monitor to stop if started as a peer
close(s.replicaCh)
// unset peer
s.cfg.GRPCPeerAddress = ""
return nil
}
func (s *Server) replicaMonitor() {
logrus.Debugf("starting replica monitor: ttl=%s", masterHeartbeatInterval)
s.replicaCh = make(chan struct{}, 1)
t := time.NewTicker(masterHeartbeatInterval)
go func() {
for range t.C {
if _, err := redis.Bytes(s.local(context.Background(), "GET", masterKey)); err != nil {
if err == redis.ErrNil {
// skip configure until new leader election
n, err := s.getNextMaster(context.Background())
if err != nil {
logrus.Error(err)
continue
}
logrus.Debugf("replica monitor: node=%s", n)
if n == nil || n.ID != s.cfg.ID {
logrus.Debug("waiting for new master to initialize")
continue
}
logrus.Debugf("replica monitor: configuring node with master %+v", n)
if err := s.configureNode(); err != nil {
logrus.Error(err)
continue
}
return
}
logrus.Error(err)
}
}
}()
<-s.replicaCh
logrus.Debug("stopping replica monitor")
t.Stop()
}
func (s *Server) getNextMaster(ctx context.Context) (*v1.Node, error) {
nodes, err := s.getNodes(ctx)
if err != nil {
return nil, err
}
if len(nodes) == 0 {
return nil, nil
}
sort.SliceStable(nodes, func(i, j int) bool { return nodes[i].Updated.Before(nodes[j].Updated) })
return nodes[len(nodes)-1], nil
}
func (s *Server) masterHeartbeat() {
logrus.Debugf("starting master heartbeat: ttl=%s", masterHeartbeatInterval)
// initial update
ctx, cancel := context.WithTimeout(context.Background(), masterHeartbeatInterval)
defer cancel()
logrus.Infof("cluster master key=%s", s.cfg.ClusterKey)
if err := s.updateMasterInfo(ctx); err != nil {
logrus.Error(err)
}
t := time.NewTicker(masterHeartbeatInterval)
for range t.C {
if err := s.updateMasterInfo(ctx); err != nil {
logrus.Error(err)
continue
}
}
}
func (s *Server) joinMaster(m *v1.Master) error {
// configure replica
logrus.Infof("configuring node as replica of %+v", m.ID)
pool, err := getPool(s.cfg.RedisURL)
if err != nil {
return err
}
conn := pool.Get()
defer conn.Close()
logrus.Debugf("configuring redis as slave of %s", m.ID)
u, err := url.Parse(m.RedisURL)
if err != nil {
return errors.Wrap(err, "error parsing master redis url")
}
hostPort := strings.SplitN(u.Host, ":", 2)
host := hostPort[0]
port := hostPort[1]
if _, err := conn.Do("REPLICAOF", host, port); err != nil {
return err
}
s.wpool, err = getPool(m.RedisURL)
if err != nil {
return err
}
return nil
}
func (s *Server) updateMasterInfo(ctx context.Context) error {
// update master info
if _, err := s.master(ctx, "SET", clusterKey, s.cfg.ClusterKey); err != nil {
logrus.Error("updateMasterInfo.setClusterKey")
return err
}
// build redis url with gateway ip
gatewayIP, _, err := s.getNodeIP(ctx, s.cfg.ID)
if err != nil {
if err == redis.ErrNil {
logrus.Warnf("node does not have an IP assigned yet")
return nil
}
return err
}
u, err := url.Parse(s.cfg.RedisURL)
if err != nil {
return err
}
// update master redis url to gateway to serve over wireguard
u.Host = fmt.Sprintf("%s:%s", gatewayIP.String(), u.Port())
m := &v1.Master{
ID: s.cfg.ID,
GRPCAddress: s.cfg.AdvertiseGRPCAddress,
RedisURL: u.String(),
GatewayIP: gatewayIP.String(),
}
data, err := proto.Marshal(m)
if err != nil {
return errors.Wrap(err, "error marshalling master info")
}
if _, err := s.master(ctx, "SET", masterKey, data); err != nil {
return errors.Wrap(err, "error setting master info")
}
if _, err := s.master(ctx, "EXPIRE", masterKey, int(masterHeartbeatInterval.Seconds())); err != nil {
return errors.Wrap(err, "error setting expire for master info")
}
return nil
}
func (s *Server) updateNodeInfo(ctx context.Context) {
logrus.Debugf("starting node heartbeat: ttl=%s", nodeHeartbeatInterval)
t := time.NewTicker(nodeHeartbeatInterval)
for range t.C {
if err := s.updateLocalNodeInfo(ctx); err != nil {
logrus.Error(err)
continue
}
}
}
func (s *Server) updateLocalNodeInfo(ctx context.Context) error {
key := s.getNodeKey(s.cfg.ID)
keyPair, err := s.getOrCreateKeyPair(ctx, s.cfg.ID)
if err != nil {
return errors.Wrapf(err, "error getting keypair for %s", s.cfg.ID)
}
nodeIP, _, err := s.getNodeIP(ctx, s.cfg.ID)
if err != nil {
return errors.Wrapf(err, "error getting node IP for %s", s.cfg.ID)
}
node := &v1.Node{
Updated: time.Now(),
ID: s.cfg.ID,
Addr: s.cfg.GRPCAddress,
KeyPair: keyPair,
EndpointIP: s.cfg.EndpointIP,
EndpointPort: uint64(s.cfg.EndpointPort),
GatewayIP: nodeIP.String(),
InterfaceName: s.cfg.InterfaceName,
}
data, err := proto.Marshal(node)
if err != nil {
return err
}
if _, err := s.master(ctx, "SET", key, data); err != nil {
return err
}
if _, err := s.master(ctx, "EXPIRE", key, nodeHeartbeatExpiry); err != nil {
return err
}
return nil
}
func (s *Server) createNode(ctx context.Context, req *v1.JoinRequest) (*v1.Node, error) {
key := s.getNodeKey(req.ID)
keyPair, err := s.getOrCreateKeyPair(ctx, req.ID)
if err != nil {
return nil, errors.Wrapf(err, "error getting/creating keypair for %s", req.ID)
}
nodeIP, _, err := s.getNodeIP(ctx, req.ID)
if err != nil {
return nil, errors.Wrapf(err, "error getting node ip for %s", req.ID)
}
node := &v1.Node{
Updated: time.Now(),
ID: req.ID,
Addr: req.GRPCAddress,
KeyPair: keyPair,
EndpointIP: req.EndpointIP,
EndpointPort: uint64(req.EndpointPort),
GatewayIP: nodeIP.String(),
InterfaceName: req.InterfaceName,
}
data, err := proto.Marshal(node)
if err != nil {
return nil, err
}
if _, err := s.master(ctx, "SET", key, data); err != nil {
return nil, err
}
if _, err := s.master(ctx, "EXPIRE", key, nodeHeartbeatExpiry); err != nil {
return nil, err
}
return node, nil
}