Files
gpt-load/internal/store/leader_lock.go
2025-07-13 20:28:14 +08:00

240 lines
6.2 KiB
Go

package store
import (
"context"
"crypto/rand"
"encoding/hex"
"fmt"
"sync"
"sync/atomic"
"time"
"github.com/sirupsen/logrus"
)
const (
leaderLockKey = "cluster:leader"
leaderLockTTL = 30 * time.Second
leaderRenewalInterval = 10 * time.Second
initializingLockKey = "cluster:initializing"
initializingLockTTL = 5 * time.Minute
)
const renewLockScript = `
if redis.call("get", KEYS[1]) == ARGV[1] then
return redis.call("expire", KEYS[1], ARGV[2])
else
return 0
end`
const releaseLockScript = `
if redis.call("get", KEYS[1]) == ARGV[1] then
return redis.call("del", KEYS[1])
else
return 0
end`
// LeaderLock provides a mechanism for electing a single leader in a cluster.
type LeaderLock struct {
store Store
nodeID string
isLeader atomic.Bool
stopChan chan struct{}
wg sync.WaitGroup
isSingleNode bool
}
// NewLeaderLock creates a new LeaderLock.
func NewLeaderLock(s Store) *LeaderLock {
_, isDistributed := s.(LuaScripter)
service := &LeaderLock{
store: s,
nodeID: generateNodeID(),
stopChan: make(chan struct{}),
isSingleNode: !isDistributed,
}
if service.isSingleNode {
logrus.Debug("Running in single-node mode. Assuming leadership.")
service.isLeader.Store(true)
} else {
logrus.Debug("Running in distributed mode.")
}
return service
}
// Start performs an initial leader election and starts the background leadership maintenance loop.
func (s *LeaderLock) Start() error {
if s.isSingleNode {
return nil
}
if err := s.tryToBeLeader(); err != nil {
return fmt.Errorf("initial leader election failed: %w", err)
}
s.wg.Add(1)
go s.maintainLeadershipLoop()
return nil
}
// Stop gracefully stops the leadership maintenance process.
func (s *LeaderLock) Stop() {
if s.isSingleNode {
return
}
logrus.Info("Stopping leadership maintenance process...")
close(s.stopChan)
s.wg.Wait()
if s.isLeader.Load() {
s.releaseLock()
}
logrus.Info("Leadership maintenance process stopped.")
}
// IsLeader returns true if the current node is the leader.
func (s *LeaderLock) IsLeader() bool {
return s.isLeader.Load()
}
// AcquireInitializingLock sets a temporary lock to indicate that initialization is in progress.
func (s *LeaderLock) AcquireInitializingLock() (bool, error) {
if !s.IsLeader() {
return false, nil
}
logrus.Debug("Leader acquiring initialization lock...")
return s.store.SetNX(initializingLockKey, []byte(s.nodeID), initializingLockTTL)
}
// ReleaseInitializingLock removes the initialization lock.
func (s *LeaderLock) ReleaseInitializingLock() {
if !s.IsLeader() {
return
}
logrus.Debug("Leader releasing initialization lock...")
if err := s.store.Delete(initializingLockKey); err != nil {
logrus.WithError(err).Error("Failed to release initialization lock.")
}
}
// WaitForInitializationToComplete waits until the initialization lock is released.
func (s *LeaderLock) WaitForInitializationToComplete() error {
if s.isSingleNode || s.IsLeader() {
return nil
}
logrus.Debug("Follower waiting for leader to complete initialization...")
time.Sleep(2 * time.Second)
// Use a context with timeout to prevent indefinite waiting.
ctx, cancel := context.WithTimeout(context.Background(), initializingLockTTL+1*time.Minute)
defer cancel()
ticker := time.NewTicker(2 * time.Second)
defer ticker.Stop()
// Initial check before starting the loop
exists, err := s.store.Exists(initializingLockKey)
if err != nil {
logrus.WithError(err).Warn("Initial check for initialization lock failed, will proceed to loop.")
} else if !exists {
logrus.Debug("Initialization lock not found on initial check. Assuming initialization is complete.")
return nil
}
for {
select {
case <-ctx.Done():
return fmt.Errorf("timed out waiting for leader initialization after %v", initializingLockTTL)
case <-ticker.C:
exists, err := s.store.Exists(initializingLockKey)
if err != nil {
logrus.WithError(err).Warn("Error checking initialization lock, will retry...")
continue
}
if !exists {
logrus.Debug("Initialization lock released. Follower proceeding with startup.")
return nil
}
}
}
}
// maintainLeadershipLoop is the background process that keeps trying to acquire or renew the lock.
func (s *LeaderLock) maintainLeadershipLoop() {
defer s.wg.Done()
ticker := time.NewTicker(leaderRenewalInterval)
defer ticker.Stop()
logrus.Debug("Leadership maintenance loop started.")
for {
select {
case <-ticker.C:
if err := s.tryToBeLeader(); err != nil {
logrus.WithError(err).Warn("Error during leadership maintenance cycle.")
}
case <-s.stopChan:
logrus.Info("Leadership maintenance loop stopping.")
return
}
}
}
// tryToBeLeader is an idempotent function that attempts to acquire or renew the lock.
func (s *LeaderLock) tryToBeLeader() error {
if s.isLeader.Load() {
err := s.renewLock()
if err != nil {
logrus.WithError(err).Error("Failed to renew leader lock, relinquishing leadership.")
s.isLeader.Store(false)
}
return nil
}
acquired, err := s.acquireLock()
if err != nil {
return fmt.Errorf("failed to acquire lock: %w", err)
}
if acquired {
logrus.WithField("nodeID", s.nodeID).Info("Successfully acquired leadership.")
s.isLeader.Store(true)
}
return nil
}
func (s *LeaderLock) acquireLock() (bool, error) {
return s.store.SetNX(leaderLockKey, []byte(s.nodeID), leaderLockTTL)
}
func (s *LeaderLock) renewLock() error {
luaStore := s.store.(LuaScripter)
ttlSeconds := int(leaderLockTTL.Seconds())
res, err := luaStore.Eval(renewLockScript, []string{leaderLockKey}, s.nodeID, ttlSeconds)
if err != nil {
return err
}
if i, ok := res.(int64); !ok || i == 0 {
return fmt.Errorf("failed to renew lock, another node may have taken over")
}
return nil
}
func (s *LeaderLock) releaseLock() {
luaStore := s.store.(LuaScripter)
if _, err := luaStore.Eval(releaseLockScript, []string{leaderLockKey}, s.nodeID); err != nil {
logrus.WithError(err).Error("Failed to release leader lock on shutdown.")
} else {
logrus.Info("Successfully released leader lock.")
}
}
func generateNodeID() string {
bytes := make([]byte, 16)
if _, err := rand.Read(bytes); err != nil {
return "node-" + time.Now().Format(time.RFC3339Nano)
}
return hex.EncodeToString(bytes)
}