Files
sub2api/backend/internal/service/token_refresh_service.go
T

1529 lines
47 KiB
Go
Raw Normal View History

package service
import (
"context"
"errors"
"fmt"
"log/slog"
"strings"
"sync"
"time"
"github.com/Wei-Shaw/sub2api/internal/config"
infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
"github.com/Wei-Shaw/sub2api/internal/util/logredact"
)
// tokenRefreshTempUnschedDuration token 刷新重试耗尽后临时不可调度的持续时间
const tokenRefreshTempUnschedDuration = 10 * time.Minute
const (
defaultTokenRefreshCandidatePageSize = 200
maxTokenRefreshCandidatePageSize = 1000
defaultTokenRefreshProviderConcurrency = 4
maxTokenRefreshProviderConcurrency = 32
defaultTokenRefreshProviderQPS = 2
maxTokenRefreshProviderQPS = 100
defaultTokenRefreshProviderFailureThreshold = 3
maxTokenRefreshProviderFailureThreshold = 100
defaultTokenRefreshMaxRetries = 1
maxTokenRefreshMaxRetries = 10
maxTokenRefreshRetryBackoff = 30 * time.Second
defaultTokenRefreshAttemptTimeout = 15 * time.Second
maxTokenRefreshAttemptTimeout = 5 * time.Minute
maxTokenRefreshLockSafetyMargin = 5 * time.Second
defaultTokenRefreshCycleTimeout = 4 * time.Minute
maxTokenRefreshCycleTimeout = time.Hour
defaultTokenRefreshCleanupTimeout = 2 * time.Second
)
type tokenRefreshRegistration struct {
platform string
refresher TokenRefresher
executor OAuthRefreshExecutor
}
// GrokOAuthRefreshMutationRepository protects background refresh failure
// mutations with the exact credential document used by the upstream attempt.
// This contract is intentionally Grok-only; existing provider behavior remains
// unchanged.
type GrokOAuthRefreshMutationRepository interface {
SetGrokOAuthRefreshErrorIfCredentialsUnchanged(ctx context.Context, id int64, expectedCredentials map[string]any, expectedProxyID *int64, errorMsg string) (bool, error)
SetGrokOAuthRefreshTempUnschedulableIfCredentialsUnchanged(ctx context.Context, id int64, expectedCredentials map[string]any, expectedProxyID *int64, until time.Time, reason string) (bool, error)
}
// TokenRefreshService OAuth token自动刷新服务
// 定期检查并刷新即将过期的token
type TokenRefreshService struct {
accountRepo AccountRepository
candidatePager OAuthRefreshCandidatePager
registrations []tokenRefreshRegistration
refreshPolicy BackgroundRefreshPolicy
cfg *config.TokenRefreshConfig
cacheInvalidator TokenCacheInvalidator
schedulerCache SchedulerCache // 用于同步更新调度器缓存,解决 token 刷新后缓存不一致问题
tempUnschedCache TempUnschedCache // 用于清除 Redis 中的临时不可调度缓存
refreshAPI *OAuthRefreshAPI // 统一刷新 API
runtimeBlocker AccountRuntimeBlocker
// OpenAI privacy: 刷新成功后检查并设置 training opt-out
privacyClientFactory PrivacyClientFactory
proxyRepo ProxyRepository
stopCh chan struct{}
stopOnce sync.Once
wg sync.WaitGroup
runCtx context.Context
runCancel context.CancelFunc
candidateMu sync.Mutex
afterID int64
providerMu sync.Mutex
providerGates map[string]*tokenRefreshRateGate
providerPools map[string]*tokenRefreshConcurrencyGate
// Test-only duration seam; production uses TokenRefreshConfig seconds.
attemptTimeoutOverride time.Duration
}
// NewTokenRefreshService 创建token刷新服务
func NewTokenRefreshService(
accountRepo AccountRepository,
oauthService *OAuthService,
openaiOAuthService *OpenAIOAuthService,
geminiOAuthService *GeminiOAuthService,
antigravityOAuthService *AntigravityOAuthService,
cacheInvalidator TokenCacheInvalidator,
schedulerCache SchedulerCache,
cfg *config.Config,
tempUnschedCache TempUnschedCache,
grokOAuthServices ...*GrokOAuthService,
) *TokenRefreshService {
refreshCfg := &config.TokenRefreshConfig{}
if cfg != nil {
refreshCfg = &cfg.TokenRefresh
}
runCtx, runCancel := context.WithCancel(context.Background())
s := &TokenRefreshService{
accountRepo: accountRepo,
refreshPolicy: DefaultBackgroundRefreshPolicy(),
cfg: refreshCfg,
cacheInvalidator: cacheInvalidator,
schedulerCache: schedulerCache,
tempUnschedCache: tempUnschedCache,
stopCh: make(chan struct{}),
runCtx: runCtx,
runCancel: runCancel,
}
if pager, ok := accountRepo.(OAuthRefreshCandidatePager); ok {
s.candidatePager = pager
}
openAIRefresher := NewOpenAITokenRefresher(openaiOAuthService, accountRepo)
claudeRefresher := NewClaudeTokenRefresher(oauthService)
geminiRefresher := NewGeminiTokenRefresher(geminiOAuthService)
agRefresher := NewAntigravityTokenRefresher(antigravityOAuthService)
var grokOAuthService *GrokOAuthService
if len(grokOAuthServices) > 0 {
grokOAuthService = grokOAuthServices[0]
}
grokRefresher := NewGrokTokenRefresher(grokOAuthService)
// Each provider is registered exactly once. The same registry supplies both
// execution and repository eligibility, preventing future platform drift.
s.registrations = []tokenRefreshRegistration{
{platform: PlatformAnthropic, refresher: claudeRefresher, executor: claudeRefresher},
{platform: PlatformOpenAI, refresher: openAIRefresher, executor: openAIRefresher},
{platform: PlatformGemini, refresher: geminiRefresher, executor: geminiRefresher},
{platform: PlatformAntigravity, refresher: agRefresher, executor: agRefresher},
{platform: PlatformGrok, refresher: grokRefresher, executor: grokRefresher},
}
return s
}
func (s *TokenRefreshService) eligiblePlatforms() []string {
platforms := make([]string, 0, len(s.registrations))
for _, registration := range s.registrations {
if registration.platform != "" && registration.refresher != nil {
platforms = append(platforms, registration.platform)
}
}
return platforms
}
func (s *TokenRefreshService) candidateAfterID() int64 {
s.candidateMu.Lock()
defer s.candidateMu.Unlock()
return s.afterID
}
func (s *TokenRefreshService) setCandidateAfterID(afterID int64) {
s.candidateMu.Lock()
s.afterID = afterID
s.candidateMu.Unlock()
}
// SetPrivacyDeps 注入 OpenAI privacy opt-out 所需依赖
func (s *TokenRefreshService) SetPrivacyDeps(factory PrivacyClientFactory, proxyRepo ProxyRepository) {
s.privacyClientFactory = factory
s.proxyRepo = proxyRepo
}
// SetRefreshAPI 注入统一的 OAuth 刷新 API
func (s *TokenRefreshService) SetRefreshAPI(api *OAuthRefreshAPI) {
s.refreshAPI = api
}
// SetRefreshPolicy 注入后台刷新调用侧策略(用于显式化平台/场景差异行为)。
func (s *TokenRefreshService) SetRefreshPolicy(policy BackgroundRefreshPolicy) {
s.refreshPolicy = policy
}
func (s *TokenRefreshService) SetAccountRuntimeBlocker(blocker AccountRuntimeBlocker) {
s.runtimeBlocker = blocker
}
func (s *TokenRefreshService) notifyAccountSchedulingBlocked(account *Account, until time.Time, reason string) {
if s == nil || s.runtimeBlocker == nil || account == nil {
return
}
s.runtimeBlocker.BlockAccountScheduling(account, until, reason)
}
func (s *TokenRefreshService) notifyAccountSchedulingBlockCleared(accountID int64) {
if s == nil || s.runtimeBlocker == nil || accountID <= 0 {
return
}
s.runtimeBlocker.ClearAccountSchedulingBlock(accountID)
}
// Start 启动后台刷新服务
func (s *TokenRefreshService) Start() {
if s.cfg == nil || !s.cfg.Enabled {
slog.Info("token_refresh.service_disabled")
return
}
s.wg.Add(1)
go s.refreshLoop()
slog.Info("token_refresh.service_started",
"check_interval_minutes", s.cfg.CheckIntervalMinutes,
"refresh_before_expiry_hours", s.cfg.RefreshBeforeExpiryHours,
)
}
// Stop 停止刷新服务(可安全多次调用)
func (s *TokenRefreshService) Stop() {
s.stopOnce.Do(func() {
if s.runCancel != nil {
s.runCancel()
}
close(s.stopCh)
})
s.wg.Wait()
slog.Info("token_refresh.service_stopped")
}
// refreshLoop 刷新循环
func (s *TokenRefreshService) refreshLoop() {
defer s.wg.Done()
ctx := s.runCtx
if ctx == nil {
ctx = context.Background()
}
// 计算检查间隔
checkInterval := time.Duration(s.cfg.CheckIntervalMinutes) * time.Minute
if checkInterval < time.Minute {
checkInterval = 5 * time.Minute
}
ticker := time.NewTicker(checkInterval)
defer ticker.Stop()
// 启动时立即执行一次检查
s.processRefreshContext(ctx)
for {
select {
case <-ticker.C:
s.processRefreshContext(ctx)
case <-ctx.Done():
return
case <-s.stopCh:
return
}
}
}
type tokenRefreshPageStats struct {
total int
oauth int
needsRefresh int
refreshed int
skipped int
failed int
}
type tokenRefreshProviderState struct {
service *TokenRefreshService
registration tokenRefreshRegistration
rateGate refreshAttemptGate
poolGate *tokenRefreshConcurrencyGate
mu sync.Mutex
consecutiveFailures int
tripped bool
}
type tokenRefreshRateGate struct {
mu sync.Mutex
next time.Time
interval time.Duration
}
type tokenRefreshConcurrencyGate struct {
slots chan struct{}
}
type refreshAttemptGate interface {
acquire(ctx context.Context) (release func(), err error)
}
type providerRefreshAttemptGate interface {
refreshAttemptGate
acquireRate(ctx context.Context) (release func(), err error)
}
type rateLimitedOAuthRefreshExecutor struct {
OAuthRefreshExecutor
acquireRate func(context.Context) (func(), error)
}
func (e *rateLimitedOAuthRefreshExecutor) Refresh(ctx context.Context, account *Account) (map[string]any, error) {
if e == nil || e.OAuthRefreshExecutor == nil {
return nil, errors.New("OAuth refresh executor is not configured")
}
release := func() {}
if e.acquireRate != nil {
var err error
release, err = e.acquireRate(ctx)
if err != nil {
return nil, err
}
}
defer release()
return e.OAuthRefreshExecutor.Refresh(ctx, account)
}
func newTokenRefreshRateGate(qps int) *tokenRefreshRateGate {
if qps <= 0 {
return &tokenRefreshRateGate{}
}
return newTokenRefreshRateGateWithInterval(time.Second / time.Duration(qps))
}
// newTokenRefreshRateGateWithInterval is a narrow duration seam used to test
// slot reservation and cancellation without waiting on production-scale QPS.
func newTokenRefreshRateGateWithInterval(interval time.Duration) *tokenRefreshRateGate {
return &tokenRefreshRateGate{interval: interval}
}
func (g *tokenRefreshRateGate) reserveSlot(now time.Time) time.Time {
g.mu.Lock()
defer g.mu.Unlock()
if g.next.Before(now) {
g.next = now
}
slot := g.next
g.next = g.next.Add(g.interval)
return slot
}
func (g *tokenRefreshRateGate) wait(ctx context.Context) error {
if g == nil || g.interval <= 0 {
return nil
}
slot := g.reserveSlot(time.Now())
wait := time.Until(slot)
if wait <= 0 {
return nil
}
timer := time.NewTimer(wait)
defer timer.Stop()
select {
case <-ctx.Done():
return ctx.Err()
case <-timer.C:
return nil
}
}
func (g *tokenRefreshRateGate) acquire(ctx context.Context) (func(), error) {
if err := g.wait(ctx); err != nil {
return nil, err
}
return func() {}, nil
}
func newTokenRefreshConcurrencyGate(concurrency int) *tokenRefreshConcurrencyGate {
if concurrency < 1 {
concurrency = 1
}
return &tokenRefreshConcurrencyGate{slots: make(chan struct{}, concurrency)}
}
func (g *tokenRefreshConcurrencyGate) acquire(ctx context.Context) (func(), error) {
if g == nil {
return func() {}, nil
}
select {
case g.slots <- struct{}{}:
return func() { <-g.slots }, nil
case <-ctx.Done():
return nil, ctx.Err()
}
}
func (p *tokenRefreshProviderState) isTripped() bool {
p.mu.Lock()
defer p.mu.Unlock()
return p.tripped
}
func (p *tokenRefreshProviderState) acquire(ctx context.Context) (func(), error) {
if p == nil || p.isTripped() {
return nil, errRefreshSkipped
}
release, err := p.poolGate.acquire(ctx)
if err != nil {
return nil, err
}
if p.isTripped() {
release()
return nil, errRefreshSkipped
}
return release, nil
}
func (p *tokenRefreshProviderState) acquireRate(ctx context.Context) (func(), error) {
if p == nil || p.isTripped() {
return nil, errRefreshSkipped
}
release := func() {}
if p.rateGate != nil {
var err error
release, err = p.rateGate.acquire(ctx)
if err != nil {
return nil, err
}
}
if p.isTripped() {
release()
return nil, errRefreshSkipped
}
return release, nil
}
func (p *tokenRefreshProviderState) recordResult(err error) {
p.mu.Lock()
defer p.mu.Unlock()
if err == nil {
p.consecutiveFailures = 0
return
}
if errors.Is(err, errRefreshSkipped) || errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
var attemptTimeoutErr *refreshAttemptTimeoutError
if !errors.As(err, &attemptTimeoutErr) {
return
}
}
var attemptTimeoutErr *refreshAttemptTimeoutError
if errors.As(err, &attemptTimeoutErr) {
p.consecutiveFailures++
if p.consecutiveFailures >= p.service.providerFailureThreshold() {
p.tripped = true
}
return
}
var providerErr *providerConfigurationRefreshError
if errors.As(err, &providerErr) {
p.tripped = true
return
}
var containmentErr *providerCycleContainmentRefreshError
if errors.As(err, &containmentErr) {
p.tripped = true
return
}
var permanentErr *accountPermanentRefreshError
if errors.As(err, &permanentErr) {
p.consecutiveFailures = 0
return
}
if isNonRetryableRefreshError(err) {
// A permanent account credential failure is isolated to that account and
// does not imply the provider is unhealthy.
p.consecutiveFailures = 0
return
}
p.consecutiveFailures++
if p.consecutiveFailures >= p.service.providerFailureThreshold() {
p.tripped = true
}
}
// processRefresh preserves the existing test/internal call surface while the
// production loop supplies a cancelable parent context.
func (s *TokenRefreshService) processRefresh() {
s.processRefreshContext(context.Background())
}
// processRefreshContext executes one bounded, cursor-resumable refresh cycle.
func (s *TokenRefreshService) processRefreshContext(parent context.Context) {
if parent == nil {
parent = context.Background()
}
ctx, cancel := context.WithTimeout(parent, s.cycleTimeout())
defer cancel()
pager := s.candidatePager
if pager == nil {
pager, _ = s.accountRepo.(OAuthRefreshCandidatePager)
}
if pager == nil {
slog.Error("token_refresh.candidate_pager_missing")
return
}
platforms := s.eligiblePlatforms()
if len(platforms) == 0 {
slog.Error("token_refresh.provider_registry_empty")
return
}
refreshWindow := time.Duration(s.cfg.RefreshBeforeExpiryHours * float64(time.Hour))
pageSize := s.candidatePageSize()
providerStates := make(map[string]*tokenRefreshProviderState, len(s.registrations))
for i := range s.registrations {
registration := s.registrations[i]
providerStates[registration.platform] = &tokenRefreshProviderState{
service: s,
registration: registration,
rateGate: s.providerRateGate(registration.platform),
poolGate: s.providerConcurrencyGate(registration.platform),
}
}
stats := tokenRefreshPageStats{}
afterID := s.candidateAfterID()
for {
if ctx.Err() != nil {
slog.Warn("token_refresh.cycle_stopped", "error", ctx.Err(), "resume_after_id", afterID)
break
}
page, err := pager.ListOAuthRefreshCandidatePage(ctx, OAuthRefreshPageOptions{
Platforms: platforms,
AfterID: afterID,
Limit: pageSize,
ActiveOnly: true,
IncludeSetupToken: true,
RequireRefreshToken: true,
ExcludeRetryCooldown: true,
})
if err != nil {
slog.Error("token_refresh.list_accounts_failed", "error", err, "after_id", afterID)
break
}
if page == nil {
slog.Error("token_refresh.nil_candidate_page", "after_id", afterID)
break
}
accounts := page.Accounts
if !page.HasMore && page.NextAfterID == 0 && len(accounts) == 0 {
s.setCandidateAfterID(0)
break
}
if page.NextAfterID <= afterID {
slog.Error("token_refresh.invalid_candidate_page_metadata", "after_id", afterID)
break
}
if !isStrictlyIncreasingAccountPage(accounts, afterID) {
slog.Error("token_refresh.invalid_candidate_page", "after_id", afterID, "count", len(accounts))
break
}
pageStats := s.processCandidatePage(ctx, accounts, providerStates, refreshWindow)
stats.total += pageStats.total
stats.oauth += pageStats.oauth
stats.needsRefresh += pageStats.needsRefresh
stats.refreshed += pageStats.refreshed
stats.skipped += pageStats.skipped
stats.failed += pageStats.failed
// Never advance past a partially processed page. Re-reading a page is
// safe because OAuthRefreshAPI re-reads DB state and checks expiry again.
if ctx.Err() != nil {
break
}
afterID = page.NextAfterID
s.setCandidateAfterID(afterID)
if !page.HasMore {
s.setCandidateAfterID(0)
break
}
}
if stats.needsRefresh == 0 && stats.failed == 0 {
slog.Debug("token_refresh.cycle_completed",
"total", stats.total, "oauth", stats.oauth,
"needs_refresh", stats.needsRefresh, "refreshed", stats.refreshed,
"skipped", stats.skipped, "failed", stats.failed)
} else {
slog.Info("token_refresh.cycle_completed",
"total", stats.total, "oauth", stats.oauth,
"needs_refresh", stats.needsRefresh, "refreshed", stats.refreshed,
"skipped", stats.skipped, "failed", stats.failed)
}
}
func isStrictlyIncreasingAccountPage(accounts []Account, afterID int64) bool {
previous := afterID
for i := range accounts {
if accounts[i].ID <= previous {
return false
}
previous = accounts[i].ID
}
return true
}
func (s *TokenRefreshService) processCandidatePage(
ctx context.Context,
accounts []Account,
providerStates map[string]*tokenRefreshProviderState,
refreshWindow time.Duration,
) tokenRefreshPageStats {
stats := tokenRefreshPageStats{total: len(accounts)}
groups := make(map[string][]*Account)
for i := range accounts {
account := &accounts[i]
state := providerStates[account.Platform]
if state == nil || state.registration.refresher == nil || !state.registration.refresher.CanRefresh(account) {
continue
}
stats.oauth++
if !state.registration.refresher.NeedsRefresh(account, refreshWindow) {
continue
}
stats.needsRefresh++
groups[account.Platform] = append(groups[account.Platform], account)
}
type providerResult struct {
refreshed int
skipped int
failed int
}
results := make(chan providerResult, len(groups))
var wg sync.WaitGroup
for platform, group := range groups {
state := providerStates[platform]
wg.Add(1)
go func() {
defer wg.Done()
refreshed, skipped, failed := s.processProviderAccounts(ctx, state, group, refreshWindow)
results <- providerResult{refreshed: refreshed, skipped: skipped, failed: failed}
}()
}
wg.Wait()
close(results)
for result := range results {
stats.refreshed += result.refreshed
stats.skipped += result.skipped
stats.failed += result.failed
}
return stats
}
func (s *TokenRefreshService) processProviderAccounts(
ctx context.Context,
state *tokenRefreshProviderState,
accounts []*Account,
refreshWindow time.Duration,
) (refreshed, skipped, failed int) {
if state == nil || len(accounts) == 0 {
return 0, 0, 0
}
type refreshResult struct {
accountID int64
err error
}
jobs := make(chan *Account, len(accounts))
results := make(chan refreshResult, len(accounts))
workerCount := s.providerConcurrency()
if workerCount > len(accounts) {
workerCount = len(accounts)
}
var wg sync.WaitGroup
for i := 0; i < workerCount; i++ {
wg.Add(1)
go func() {
defer wg.Done()
for account := range jobs {
if ctx.Err() != nil || state.isTripped() {
results <- refreshResult{accountID: account.ID, err: errRefreshSkipped}
continue
}
if state.isTripped() {
results <- refreshResult{accountID: account.ID, err: errRefreshSkipped}
continue
}
err := s.refreshWithRetryWithRateGate(ctx, account, state.registration.refresher, state.registration.executor, refreshWindow, state)
state.recordResult(err)
results <- refreshResult{accountID: account.ID, err: err}
}
}()
}
for _, account := range accounts {
jobs <- account
}
close(jobs)
wg.Wait()
close(results)
for result := range results {
switch {
case result.err == nil:
refreshed++
slog.Info("token_refresh.account_refreshed", "account_id", result.accountID, "platform", state.registration.platform)
case errors.Is(result.err, errRefreshSkipped):
skipped++
default:
failed++
slog.Warn("token_refresh.account_refresh_failed", "account_id", result.accountID, "platform", state.registration.platform, "error", logredact.RedactText(result.err.Error()))
}
}
return refreshed, skipped, failed
}
func (s *TokenRefreshService) candidatePageSize() int {
if s.cfg != nil && s.cfg.CandidatePageSize > 0 {
return min(s.cfg.CandidatePageSize, maxTokenRefreshCandidatePageSize)
}
return defaultTokenRefreshCandidatePageSize
}
func (s *TokenRefreshService) providerConcurrency() int {
if s.cfg != nil && s.cfg.ProviderConcurrency > 0 {
return min(s.cfg.ProviderConcurrency, maxTokenRefreshProviderConcurrency)
}
return defaultTokenRefreshProviderConcurrency
}
func (s *TokenRefreshService) providerQPS() int {
if s.cfg != nil && s.cfg.ProviderQPS > 0 {
return min(s.cfg.ProviderQPS, maxTokenRefreshProviderQPS)
}
return defaultTokenRefreshProviderQPS
}
// providerRateGate returns the process-local limiter shared by background
// cycles and admin reconciliation. Sharing it prevents concurrent entry points
// or retries from multiplying the configured per-provider request rate.
func (s *TokenRefreshService) providerRateGate(platform string) *tokenRefreshRateGate {
s.providerMu.Lock()
defer s.providerMu.Unlock()
if s.providerGates == nil {
s.providerGates = make(map[string]*tokenRefreshRateGate)
}
if gate := s.providerGates[platform]; gate != nil {
return gate
}
gate := newTokenRefreshRateGate(s.providerQPS())
s.providerGates[platform] = gate
return gate
}
// providerConcurrencyGate returns the process-local semaphore shared by every
// background cycle and admin reconciliation call for a provider. It is
// acquired and released around each upstream retry attempt, so parallel entry
// points cannot multiply ProviderConcurrency.
func (s *TokenRefreshService) providerConcurrencyGate(platform string) *tokenRefreshConcurrencyGate {
s.providerMu.Lock()
defer s.providerMu.Unlock()
if s.providerPools == nil {
s.providerPools = make(map[string]*tokenRefreshConcurrencyGate)
}
if gate := s.providerPools[platform]; gate != nil {
return gate
}
gate := newTokenRefreshConcurrencyGate(s.providerConcurrency())
s.providerPools[platform] = gate
return gate
}
func (s *TokenRefreshService) providerFailureThreshold() int {
if s.cfg != nil && s.cfg.ProviderFailureThreshold > 0 {
return min(s.cfg.ProviderFailureThreshold, maxTokenRefreshProviderFailureThreshold)
}
return defaultTokenRefreshProviderFailureThreshold
}
func (s *TokenRefreshService) attemptTimeout() time.Duration {
timeout := defaultTokenRefreshAttemptTimeout
if s.attemptTimeoutOverride > 0 {
timeout = s.attemptTimeoutOverride
} else if s.cfg != nil && s.cfg.AttemptTimeoutSeconds > 0 {
seconds := min(s.cfg.AttemptTimeoutSeconds, int(maxTokenRefreshAttemptTimeout/time.Second))
timeout = time.Duration(seconds) * time.Second
}
if s.refreshAPI != nil && s.refreshAPI.tokenCache != nil {
timeout = clampRefreshAttemptToLockLease(timeout, s.refreshAPI.lockTTL)
}
return timeout
}
func clampRefreshAttemptToLockLease(timeout, lease time.Duration) time.Duration {
if timeout <= 0 || lease <= 0 {
return timeout
}
margin := lease / 10
if margin > maxTokenRefreshLockSafetyMargin {
margin = maxTokenRefreshLockSafetyMargin
}
if margin <= 0 {
margin = time.Nanosecond
}
leaseBudget := lease - margin
if leaseBudget <= 0 {
leaseBudget = lease / 2
}
if leaseBudget > 0 && timeout > leaseBudget {
return leaseBudget
}
return timeout
}
func (s *TokenRefreshService) cycleTimeout() time.Duration {
if s.cfg != nil && s.cfg.CycleTimeoutSeconds > 0 {
seconds := min(s.cfg.CycleTimeoutSeconds, int(maxTokenRefreshCycleTimeout/time.Second))
return time.Duration(seconds) * time.Second
}
return defaultTokenRefreshCycleTimeout
}
func (s *TokenRefreshService) maxRetries() int {
if s.cfg != nil && s.cfg.MaxRetries > 0 {
return min(s.cfg.MaxRetries, maxTokenRefreshMaxRetries)
}
return defaultTokenRefreshMaxRetries
}
// refreshWithRetry 带重试的刷新
func (s *TokenRefreshService) refreshWithRetry(ctx context.Context, account *Account, refresher TokenRefresher, executor OAuthRefreshExecutor, refreshWindow time.Duration) error {
return s.refreshWithRetryWithRateGate(ctx, account, refresher, executor, refreshWindow, nil)
}
func (s *TokenRefreshService) refreshWithRetryWithRateGate(
ctx context.Context,
account *Account,
refresher TokenRefresher,
executor OAuthRefreshExecutor,
refreshWindow time.Duration,
gate refreshAttemptGate,
) error {
var lastErr error
maxRetries := s.maxRetries()
for attempt := 1; attempt <= maxRetries; attempt++ {
if err := ctx.Err(); err != nil {
return err
}
releaseAttempt := func() {}
var acquireRate func(context.Context) (func(), error)
if gate != nil {
if providerGate, ok := gate.(providerRefreshAttemptGate); ok {
var err error
releaseAttempt, err = providerGate.acquire(ctx)
if err != nil {
return err
}
acquireRate = providerGate.acquireRate
} else {
// Compatibility gates are rate-admission gates. Acquire them only
// when an upstream Refresh call is actually about to start.
acquireRate = gate.acquire
}
}
attemptCtx, cancelAttempt := context.WithTimeout(ctx, s.attemptTimeout())
var newCredentials map[string]any
var err error
shortCircuit := false
credentialsPersisted := false
// 优先使用统一 API(带分布式锁 + DB 重读保护)
if s.refreshAPI != nil && executor != nil {
actualExecutor := executor
if acquireRate != nil {
actualExecutor = &rateLimitedOAuthRefreshExecutor{
OAuthRefreshExecutor: executor,
acquireRate: acquireRate,
}
}
result, refreshErr := s.refreshAPI.RefreshIfNeeded(attemptCtx, account, actualExecutor, refreshWindow)
if result != nil && result.Account != nil {
account = result.Account
}
if refreshErr != nil {
err = refreshErr
} else if result.LockHeld {
// 锁被其他 worker 持有,由调用侧策略决定如何计数
err = s.refreshPolicy.handleLockHeld()
shortCircuit = true
} else if !result.Refreshed {
// 已被其他路径刷新,由调用侧策略决定如何计数
err = s.refreshPolicy.handleAlreadyRefreshed()
shortCircuit = true
} else {
credentialsPersisted = result.NewCredentials != nil
_ = result.NewCredentials // 统一 API 已设置 _token_version 并更新 DB,无需重复操作
}
} else {
// 降级:直接调用 refresher(兼容旧路径)
releaseRate := func() {}
if acquireRate != nil {
releaseRate, err = acquireRate(attemptCtx)
}
if err == nil {
newCredentials, err = refresher.Refresh(attemptCtx, account)
}
if releaseRate != nil {
releaseRate()
}
attemptTimedOut := errors.Is(attemptCtx.Err(), context.DeadlineExceeded) && ctx.Err() == nil
if err == nil && newCredentials != nil && !attemptTimedOut {
newCredentials["_token_version"] = time.Now().UnixMilli()
if saveErr := persistAccountCredentials(attemptCtx, s.accountRepo, account, newCredentials); saveErr != nil {
err = fmt.Errorf("failed to save credentials: %w", saveErr)
} else {
credentialsPersisted = true
}
}
}
attemptTimedOut := errors.Is(attemptCtx.Err(), context.DeadlineExceeded) && ctx.Err() == nil
cancelAttempt()
releaseAttempt()
persistedAfterAttemptDeadline := attemptTimedOut && credentialsPersisted && err == nil
if attemptTimedOut && !persistedAfterAttemptDeadline && !isProviderScopedTerminalRefreshError(err) {
cause := err
if cause == nil {
cause = context.DeadlineExceeded
}
err = &refreshAttemptTimeoutError{err: cause}
shortCircuit = false
if credentialsPersisted {
s.postRefreshStateSyncWithCleanup(ctx, account)
}
}
if shortCircuit {
return err
}
if err == nil {
if ctxErr := ctx.Err(); ctxErr != nil {
if credentialsPersisted {
s.postRefreshStateSyncWithCleanup(ctx, account)
}
return ctxErr
}
if persistedAfterAttemptDeadline {
// The provider result and exact-state CAS are already durable. Only
// the internal attempt budget elapsed while bounded detached cleanup
// completed; do not convert that success into retry/cooldown/breaker
// evidence. Publish cache state with a fresh cleanup context and stop.
s.postRefreshStateSyncWithCleanup(ctx, account)
return nil
}
s.postRefreshActions(ctx, account)
return nil
}
if ctxErr := ctx.Err(); ctxErr != nil {
if credentialsPersisted {
s.postRefreshStateSyncWithCleanup(ctx, account)
}
return ctxErr
}
if errors.Is(err, errRefreshSkipped) {
return errRefreshSkipped
}
if isProviderScopedTerminalRefreshError(err) {
return err
}
var stateUnavailableErr *oauthRefreshStateUnavailableError
if errors.As(err, &stateUnavailableErr) {
return &providerCycleContainmentRefreshError{err: err}
}
if isAmbiguousGrokEntitlementRefreshError(account, err) {
// The current Grok client labels every token-endpoint 403 as an
// entitlement denial. Without explicit entitlement evidence, contain
// the provider for this cycle instead of disabling an account on a
// possible WAF or shared provider failure.
return &providerCycleContainmentRefreshError{err: err}
}
// Provider-wide OAuth client/scope failures are not evidence that every
// account is invalid. Return a typed internal signal so the cycle contains
// the provider without mutating account state.
if isSharedProviderRefreshError(err) {
return &providerConfigurationRefreshError{err: err}
}
// 不可重试错误(invalid_grant/invalid_client 等)直接标记 error 状态并返回
if isNonRetryableRefreshError(err) {
errorMsg := "Token refresh failed (non-retryable): " + logredact.RedactText(err.Error())
isGrokOAuth := account.IsGrokOAuth()
if !isGrokOAuth {
s.notifyAccountSchedulingBlocked(account, time.Time{}, "token_refresh_non_retryable")
}
s.clearAntigravityForceTokenRefresh(ctx, account, "non_retryable")
persistentlyBlocked := false
var setErr error
if isGrokOAuth {
conditionalRepo, ok := s.accountRepo.(GrokOAuthRefreshMutationRepository)
if !ok {
return &providerConfigurationRefreshError{
err: errors.New("grok OAuth conditional refresh mutation repository is not configured"),
}
} else {
persistentlyBlocked, setErr = conditionalRepo.SetGrokOAuthRefreshErrorIfCredentialsUnchanged(
ctx,
account.ID,
account.Credentials,
account.ProxyID,
errorMsg,
)
if setErr == nil && !persistentlyBlocked {
slog.Info("token_refresh.grok_error_status_skipped_stale_credentials", "account_id", account.ID)
return errRefreshSkipped
}
}
} else {
setErr = s.accountRepo.SetError(ctx, account.ID, errorMsg)
persistentlyBlocked = setErr == nil
}
if setErr != nil {
slog.Error("token_refresh.set_error_status_failed",
"account_id", account.ID,
"error", setErr,
)
if isGrokOAuth {
return &providerCycleContainmentRefreshError{
err: fmt.Errorf("failed to conditionally persist Grok OAuth refresh failure: %w", setErr),
}
}
} else if isGrokOAuth && persistentlyBlocked {
s.notifyAccountSchedulingBlocked(account, time.Time{}, "token_refresh_non_retryable")
}
cacheInvalidationFailed := false
if account.Type == AccountTypeOAuth && (!isGrokOAuth || persistentlyBlocked) {
if s.cacheInvalidator == nil {
cacheInvalidationFailed = true
} else if invalidateErr := s.cacheInvalidator.InvalidateToken(ctx, account); invalidateErr != nil {
cacheInvalidationFailed = true
slog.Warn("token_refresh.invalidate_failed_token_cache_failed",
"account_id", account.ID,
"error", logredact.RedactText(invalidateErr.Error()),
)
}
}
return &accountPermanentRefreshError{
err: err,
persistentlyBlocked: persistentlyBlocked,
cacheInvalidationFailed: cacheInvalidationFailed,
}
}
lastErr = err
slog.Warn("token_refresh.retry_attempt_failed",
"account_id", account.ID,
"attempt", attempt,
"max_retries", maxRetries,
"error", logredact.RedactText(err.Error()),
)
// 如果还有重试机会,等待后重试
if attempt < maxRetries {
backoff := s.retryBackoff(account.ID, attempt)
if backoff > 0 {
timer := time.NewTimer(backoff)
select {
case <-ctx.Done():
timer.Stop()
return ctx.Err()
case <-timer.C:
}
}
}
}
if err := ctx.Err(); err != nil {
return err
}
// 可重试错误耗尽:临时标记账号不可调度,避免请求路径反复命中已知失败的账号
slog.Warn("token_refresh.retry_exhausted",
"account_id", account.ID,
"platform", account.Platform,
"max_retries", maxRetries,
"error", logredact.RedactText(lastErr.Error()),
)
// 设置临时不可调度 10 分钟(不标记 error,保持 status=active 让下个刷新周期能继续尝试)
until := time.Now().Add(tokenRefreshTempUnschedDuration)
reason := "token refresh retry exhausted"
if lastErr != nil {
reason += ": " + logredact.RedactText(lastErr.Error())
}
if account.IsGrokOAuth() {
conditionalRepo, ok := s.accountRepo.(GrokOAuthRefreshMutationRepository)
if !ok {
return &providerConfigurationRefreshError{
err: errors.New("grok OAuth conditional refresh mutation repository is not configured"),
}
}
applied, setErr := conditionalRepo.SetGrokOAuthRefreshTempUnschedulableIfCredentialsUnchanged(
ctx,
account.ID,
account.Credentials,
account.ProxyID,
until,
reason,
)
if setErr != nil {
slog.Warn("token_refresh.set_temp_unschedulable_failed",
"account_id", account.ID,
"error", setErr,
)
return &providerCycleContainmentRefreshError{
err: fmt.Errorf("failed to conditionally persist Grok OAuth refresh cooldown: %w", setErr),
}
} else if !applied {
slog.Info("token_refresh.grok_temp_unschedulable_skipped_stale_credentials", "account_id", account.ID)
return errRefreshSkipped
} else {
s.notifyAccountSchedulingBlocked(account, until, "token_refresh_retry_exhausted")
slog.Info("token_refresh.temp_unschedulable_set",
"account_id", account.ID,
"until", until.Format(time.RFC3339),
)
}
return lastErr
}
s.notifyAccountSchedulingBlocked(account, until, "token_refresh_retry_exhausted")
if setErr := s.accountRepo.SetTempUnschedulable(ctx, account.ID, until, reason); setErr != nil {
slog.Warn("token_refresh.set_temp_unschedulable_failed",
"account_id", account.ID,
"error", setErr,
)
} else {
slog.Info("token_refresh.temp_unschedulable_set",
"account_id", account.ID,
"until", until.Format(time.RFC3339),
)
}
return lastErr
}
func (s *TokenRefreshService) retryBackoff(accountID int64, attempt int) time.Duration {
if s.cfg == nil || s.cfg.RetryBackoffSeconds <= 0 {
return 0
}
shift := attempt - 1
if shift > 10 {
shift = 10
}
baseSeconds := min(s.cfg.RetryBackoffSeconds, int(maxTokenRefreshRetryBackoff/time.Second))
base := time.Duration(baseSeconds) * time.Second * time.Duration(1<<shift)
// Stable 75-125% jitter prevents synchronized replicas from retrying on the
// same boundaries without making tests or operations nondeterministic.
jitterPercent := int64(75) + (accountID+int64(attempt*17))%51
backoff := base * time.Duration(jitterPercent) / 100
return min(backoff, maxTokenRefreshRetryBackoff)
}
// postRefreshActions 刷新成功后的后续动作(清除错误状态、缓存失效、调度器同步等)
func (s *TokenRefreshService) postRefreshActions(ctx context.Context, account *Account) {
s.clearAntigravityForceTokenRefresh(ctx, account, "success")
// Antigravity 账户:如果之前是因为缺少 project_id 而标记为 error,现在成功获取到了,清除错误状态
if account.Platform == PlatformAntigravity &&
account.Status == StatusError &&
strings.Contains(account.ErrorMessage, "missing_project_id:") {
if clearErr := s.accountRepo.ClearError(ctx, account.ID); clearErr != nil {
slog.Warn("token_refresh.clear_account_error_failed",
"account_id", account.ID,
"error", clearErr,
)
} else {
slog.Info("token_refresh.cleared_missing_project_id_error", "account_id", account.ID)
s.notifyAccountSchedulingBlockCleared(account.ID)
}
}
// 刷新成功后清除临时不可调度状态(处理 OAuth 401 恢复场景)
if account.TempUnschedulableUntil != nil && time.Now().Before(*account.TempUnschedulableUntil) {
if clearErr := s.accountRepo.ClearTempUnschedulable(ctx, account.ID); clearErr != nil {
slog.Warn("token_refresh.clear_temp_unschedulable_failed",
"account_id", account.ID,
"error", clearErr,
)
} else {
slog.Info("token_refresh.cleared_temp_unschedulable", "account_id", account.ID)
s.notifyAccountSchedulingBlockCleared(account.ID)
}
// 同步清除 Redis 缓存,避免调度器读到过期的临时不可调度状态
if s.tempUnschedCache != nil {
if clearErr := s.tempUnschedCache.DeleteTempUnsched(ctx, account.ID); clearErr != nil {
slog.Warn("token_refresh.clear_temp_unsched_cache_failed",
"account_id", account.ID,
"error", clearErr,
)
}
}
}
s.postRefreshStateSync(ctx, account)
// OpenAI OAuth: 刷新成功后,检查是否已设置 privacy_mode,未设置则尝试关闭训练数据共享
s.ensureOpenAIPrivacy(ctx, account)
// Antigravity OAuth: 刷新成功后,检查是否已设置 privacy_mode,未设置则调用 setUserSettings
s.ensureAntigravityPrivacy(ctx, account)
// Grok: clear soft reauth flag after a successful credential refresh.
if account != nil && account.Platform == PlatformGrok && accountGrokNeedsReauth(account) {
clearGrokNeedsReauthExtra(ctx, s.accountRepo, account.ID)
}
}
func (s *TokenRefreshService) postRefreshStateSyncWithCleanup(parent context.Context, account *Account) {
cleanupParent := context.Background()
if parent != nil {
cleanupParent = context.WithoutCancel(parent)
}
ctx, cancel := context.WithTimeout(cleanupParent, defaultTokenRefreshCleanupTimeout)
defer cancel()
s.postRefreshStateSync(ctx, account)
}
func (s *TokenRefreshService) postRefreshStateSync(ctx context.Context, account *Account) {
// 对所有 OAuth 账号调用缓存失效(InvalidateToken 内部根据平台判断是否需要处理)
if s.cacheInvalidator != nil && account.Type == AccountTypeOAuth {
if err := s.cacheInvalidator.InvalidateToken(ctx, account); err != nil {
slog.Warn("token_refresh.invalidate_token_cache_failed",
"account_id", account.ID,
"error", err,
)
} else {
slog.Debug("token_refresh.token_cache_invalidated", "account_id", account.ID)
}
}
// 同步更新调度器缓存,确保调度获取的 Account 对象包含最新的 credentials
if s.schedulerCache != nil {
if err := s.schedulerCache.SetAccount(ctx, account); err != nil {
slog.Warn("token_refresh.sync_scheduler_cache_failed",
"account_id", account.ID,
"error", err,
)
} else {
slog.Debug("token_refresh.scheduler_cache_synced", "account_id", account.ID)
}
}
}
func (s *TokenRefreshService) clearAntigravityForceTokenRefresh(ctx context.Context, account *Account, outcome string) {
if s == nil || account == nil || !accountNeedsAntigravityForceTokenRefresh(account) {
return
}
updates := clearAntigravityForceTokenRefreshExtra()
if err := s.accountRepo.UpdateExtra(ctx, account.ID, updates); err != nil {
slog.Warn("token_refresh.clear_antigravity_force_refresh_failed",
"account_id", account.ID,
"outcome", outcome,
"error", err,
)
return
}
if account.Extra != nil {
for k, v := range updates {
account.Extra[k] = v
}
}
slog.Info("token_refresh.cleared_antigravity_force_refresh",
"account_id", account.ID,
"outcome", outcome,
)
}
// errRefreshSkipped 表示刷新被跳过(锁竞争或已被其他路径刷新),不计入 failed 或 refreshed
var errRefreshSkipped = fmt.Errorf("refresh skipped")
type providerConfigurationRefreshError struct {
err error
}
type providerCycleContainmentRefreshError struct {
err error
}
type accountPermanentRefreshError struct {
err error
persistentlyBlocked bool
cacheInvalidationFailed bool
}
type refreshAttemptTimeoutError struct {
err error
}
func isProviderScopedTerminalRefreshError(err error) bool {
if err == nil {
return false
}
var containmentErr *providerCycleContainmentRefreshError
if errors.As(err, &containmentErr) {
return true
}
var configurationErr *providerConfigurationRefreshError
return errors.As(err, &configurationErr)
}
func (e *refreshAttemptTimeoutError) Error() string {
return "OAuth refresh attempt timed out"
}
func (e *refreshAttemptTimeoutError) Unwrap() error {
if e == nil || e.err == nil {
return context.DeadlineExceeded
}
return e.err
}
func (e *providerConfigurationRefreshError) Error() string {
return "provider OAuth configuration rejected"
}
func (e *providerCycleContainmentRefreshError) Error() string {
return "provider OAuth failure contained for this cycle"
}
func (e *providerCycleContainmentRefreshError) Unwrap() error {
if e == nil {
return nil
}
return e.err
}
func (e *accountPermanentRefreshError) Error() string {
return "account OAuth credentials permanently rejected"
}
func (e *accountPermanentRefreshError) Unwrap() error {
if e == nil {
return nil
}
return e.err
}
func isAmbiguousGrokEntitlementRefreshError(account *Account, err error) bool {
if account == nil || !account.IsGrokOAuth() || err == nil {
return false
}
msg := strings.ToLower(err.Error())
coarseEntitlementLabel := strings.EqualFold(infraerrors.Reason(err), "GROK_OAUTH_ENTITLEMENT_DENIED") ||
strings.Contains(msg, "grok_oauth_entitlement_denied")
if !coarseEntitlementLabel {
return false
}
for _, evidence := range []string{
"subscription required",
"no active grok subscription",
"no active subscription",
"grok subscription required",
"account is not entitled",
"not entitled",
"entitlement required",
"subscription inactive",
"subscription expired",
"upgrade your plan",
} {
if strings.Contains(msg, evidence) {
return false
}
}
if bodyIndex := strings.Index(msg, "body:"); bodyIndex >= 0 {
body := msg[bodyIndex+len("body:"):]
for _, evidence := range []string{
"entitlement_denied",
"entitlement denied",
"subscription_required",
"no_active_subscription",
} {
if strings.Contains(body, evidence) {
return false
}
}
}
return true
}
func (e *providerConfigurationRefreshError) Unwrap() error {
if e == nil {
return nil
}
return e.err
}
func isSharedProviderRefreshError(err error) bool {
if err == nil {
return false
}
msg := strings.ToLower(err.Error())
for _, needle := range []string{
"invalid_client",
"unauthorized_client",
"invalid_scope",
"unknown scope",
} {
if strings.Contains(msg, needle) {
return true
}
}
return false
}
// isNonRetryableRefreshError 判断是否为不可重试的刷新错误
// 这些错误通常表示凭证已失效或配置确实缺失,需要用户重新授权
// 注意:missing_project_id 错误只在真正缺失(从未获取过)时返回,临时获取失败不会返回此错误
func isNonRetryableRefreshError(err error) bool {
if err == nil {
return false
}
msg := strings.ToLower(err.Error())
nonRetryable := []string{
"invalid_grant", // refresh_token 已失效
"invalid_refresh_token", // refresh_token 无效, team 账号工作区被删除会出现
"token_expired", // OpenAI refresh_token 已过期,需要重新授权
"app_session_terminated", // refresh_token team 账号工作区被删除
"refresh_token_reused", // OpenAI refresh_token 已被使用,必须重新授权
"refresh_token_invalidated", // OpenAI session ended; refresh token invalidated
"invalid_client", // 客户端配置错误
"unauthorized_client", // 客户端未授权
"access_denied", // 访问被拒绝
"missing_project_id", // 缺少 project_id
"no refresh token available",
"grok_oauth_entitlement_denied",
"entitlement_denied",
"invalid_scope",
"unknown scope",
"subscription required",
"no active grok subscription",
}
for _, needle := range nonRetryable {
if strings.Contains(msg, needle) {
return true
}
}
return false
}
// ensureOpenAIPrivacy 检查 OpenAI OAuth 账号是否已设置 privacy_mode
// 未设置则调用 disableOpenAITraining 并持久化结果到 Extra。
func (s *TokenRefreshService) ensureOpenAIPrivacy(ctx context.Context, account *Account) {
if account.Platform != PlatformOpenAI || account.Type != AccountTypeOAuth {
return
}
if s.privacyClientFactory == nil {
return
}
if shouldSkipOpenAIPrivacyEnsure(account.Extra) {
return
}
token, _ := account.Credentials["access_token"].(string)
if token == "" {
return
}
var proxyURL string
if account.ProxyID != nil && s.proxyRepo != nil {
if p, err := s.proxyRepo.GetByID(ctx, *account.ProxyID); err == nil && p != nil {
proxyURL = p.URL()
}
}
mode := disableOpenAITraining(ctx, s.privacyClientFactory, token, proxyURL)
if mode == "" {
return
}
if err := s.accountRepo.UpdateExtra(ctx, account.ID, map[string]any{"privacy_mode": mode}); err != nil {
slog.Warn("token_refresh.update_privacy_mode_failed",
"account_id", account.ID,
"error", err,
)
} else {
slog.Info("token_refresh.privacy_mode_set",
"account_id", account.ID,
"privacy_mode", mode,
)
}
}
// ensureAntigravityPrivacy 后台刷新中检查 Antigravity OAuth 账号隐私状态。
// 仅当 privacy_mode 已成功设置("privacy_set")时跳过;
// 未设置或之前失败("privacy_set_failed")均会重试。
func (s *TokenRefreshService) ensureAntigravityPrivacy(ctx context.Context, account *Account) {
if account.Platform != PlatformAntigravity || account.Type != AccountTypeOAuth {
return
}
if account.Extra != nil {
if mode, ok := account.Extra["privacy_mode"].(string); ok && mode == AntigravityPrivacySet {
return
}
}
token, _ := account.Credentials["access_token"].(string)
if token == "" {
return
}
projectID, _ := account.Credentials["project_id"].(string)
var proxyURL string
if account.ProxyID != nil && s.proxyRepo != nil {
if p, err := s.proxyRepo.GetByID(ctx, *account.ProxyID); err == nil && p != nil {
proxyURL = p.URL()
}
}
mode := setAntigravityPrivacy(ctx, token, projectID, proxyURL)
if mode == "" {
return
}
if err := s.accountRepo.UpdateExtra(ctx, account.ID, map[string]any{"privacy_mode": mode}); err != nil {
slog.Warn("token_refresh.update_antigravity_privacy_mode_failed",
"account_id", account.ID,
"error", err,
)
} else {
applyAntigravityPrivacyMode(account, mode)
slog.Info("token_refresh.antigravity_privacy_mode_set",
"account_id", account.ID,
"privacy_mode", mode,
)
}
}