Files
new-api/model/channel_cache.go
T
Calcium-Ion 86ac0f7745 refactor: extract protocol conversion layer into standalone relaykit module (#6369)
* test(relayconvert): add golden snapshot matrix and relaykit boundary guard

Phase 0 of the relaykit extraction plan: pin byte-level output of every
registered (from,to) request/response/stream conversion route, and
forbid kit-bound packages from growing host-only imports.

* wip(relayconvert): drop gin.Context from converter signatures; add convmeta draft

Phase 1 in progress: relayconvert now takes context.Context; host media
resolver adapts gin.Context back at the service boundary.

* refactor(relayconvert): decouple converters from RelayInfo, gin, and settings

Phase 1 of the relaykit extraction plan:
- converters now depend on convmeta.Meta (implemented by RelayInfo) instead
  of *relaycommon.RelayInfo; ClaudeConvertInfo and the format guesser move
  to convmeta with aliases left behind
- host settings reach converters via a convmeta.Options snapshot built in
  RelayInfo.ConvOptions; no more model_setting/reasoning global reads inside
  the conversion layer
- effort-suffix helpers move to service/relayconvert/reasoning (old package
  forwards); chat-to-responses upgrade policy moves to service (host routing
  logic, not conversion)
- golden conversion matrix unchanged

* test(relayconvert): tighten boundary — kit packages now free of gin/setting imports

* refactor(dto): drop gin and logger dependencies

Phase 2 (part 1): dto.Request.IsStream now takes *http.Request instead of
*gin.Context (Gemini's impl reads query/path off the std request); dto's
three logger calls become common.SysError. Boundary test allowlist is now
empty — kit-bound packages import no gin/setting/logger/model.

* refactor(kit): extract dependency-free kitutil; dto/types/relayconvert stop importing common

Phase 2 of the relaykit extraction plan:
- new service/relayconvert/kitutil holds the pure helpers the kit needs
  (JSON wrappers, pointer/string/uuid/timestamp utils, MaskSensitiveInfo,
  pluggable LogInfo/LogError hooks, Debug flag)
- dto, types, and all relayconvert packages now use kitutil; their only
  remaining internal deps are dto/types/constant
- common keeps every original symbol (MaskSensitiveInfo delegates to
  kitutil) so host code is untouched; main.go routes kit logging into
  common.SysLog/SysError and mirrors DebugEnabled
- golden conversion matrix unchanged

* refactor(kit): move EndpointType/FinishReason to types; OpenRouter dialect via Options

Kit packages (dto/types/relayconvert/reasonmap) no longer import constant:
- EndpointType and finish-reason values live in types; constant re-exports
- the OpenRouter special-case in claude->openai request conversion reads
  Options.OpenRouterDialect, set by the host from the channel type;
  InitChannelMeta invalidates the cached snapshot on channel switch

* refactor: extract relaykit submodule (dto/types/relayconvert/reasonmap)

Phase 3 of the relaykit extraction plan:
- new go module github.com/QuantumNous/new-api/relaykit containing dto
  (minus task family), types, relayconvert (with convmeta/kitutil/reasoning),
  and reasonmap; host consumes it via require + replace, go.work for dev
- task-family dto (task/suno/midjourney/video) stays in the host dto
  package; dual-consumer host files alias it as taskdto
- relaykit builds and tests standalone (GOWORK=off): no host imports,
  no gin, no DB, no settings
- golden conversion matrix unchanged

* build(docker): copy relaykit/go.mod before go mod download

The local-replace submodule's go.mod must exist inside the build context
for the main module graph to resolve.

* fix: address relaykit extraction regressions

* fix: address relaykit review regressions

* docs: document Meta nil receiver contract

* fix(relaykit): fail OpenAI→Claude conversion without max_tokens; reject negative default_max_tokens

The Claude Messages API requires max_tokens (omitting it is a 400
"Field required"), but with a nil Options.Claude.DefaultMaxTokens hook
the converters silently emitted a request the upstream is guaranteed to
reject. Both OpenAI Chat and Responses → Claude conversions now return
sharedclaude.ErrMissingMaxTokens when no path (client value, default
hook, thinking-adapter floor) supplied one. Unreachable in the host,
which always configures the hook.

Host side, claude.default_max_tokens now rejects negative values at the
option API before persisting — they would wrap into huge unsigned values
during conversion. Zero stays allowed: the current API treats
max_tokens: 0 as cache pre-warming.

* fix: make Gemini safety settings read path race-free
2026-07-27 15:56:21 +08:00

330 lines
11 KiB
Go

package model
import (
"errors"
"fmt"
"math/rand"
"sort"
"strings"
"sync"
"time"
"github.com/QuantumNous/new-api/common"
"github.com/QuantumNous/new-api/constant"
"github.com/QuantumNous/new-api/logger"
"github.com/QuantumNous/new-api/relaykit/dto"
"github.com/QuantumNous/new-api/setting/ratio_setting"
)
var group2model2channels map[string]map[string][]int // enabled channel
var channelsIDM map[int]*Channel // all channels include disabled
// channel2advancedCustomConfig caches parsed Advanced Custom (type 58) configs so
// path-aware selection avoids re-parsing JSON per request. Refreshed on full sync.
var channel2advancedCustomConfig map[int]*dto.AdvancedCustomConfig
var channelSyncLock sync.RWMutex
func InitChannelCache() {
if !common.MemoryCacheEnabled {
InvalidatePricingCache()
return
}
newChannelId2channel := make(map[int]*Channel)
newChannel2advancedCustomConfig := make(map[int]*dto.AdvancedCustomConfig)
var channels []*Channel
DB.Find(&channels)
for _, channel := range channels {
newChannelId2channel[channel.Id] = channel
if channel.Type == constant.ChannelTypeAdvancedCustom {
if config := channel.GetOtherSettings().AdvancedCustom; config != nil {
newChannel2advancedCustomConfig[channel.Id] = config
}
}
}
var abilities []*Ability
DB.Find(&abilities)
groups := make(map[string]bool)
for _, ability := range abilities {
groups[ability.Group] = true
}
newGroup2model2channels := make(map[string]map[string][]int)
for group := range groups {
newGroup2model2channels[group] = make(map[string][]int)
}
for _, channel := range channels {
if channel.Status != common.ChannelStatusEnabled {
continue // skip disabled channels
}
groups := strings.Split(channel.Group, ",")
for _, group := range groups {
models := strings.Split(channel.Models, ",")
for _, model := range models {
if _, ok := newGroup2model2channels[group][model]; !ok {
newGroup2model2channels[group][model] = make([]int, 0)
}
newGroup2model2channels[group][model] = append(newGroup2model2channels[group][model], channel.Id)
}
}
}
// sort by priority
for group, model2channels := range newGroup2model2channels {
for model, channels := range model2channels {
sort.Slice(channels, func(i, j int) bool {
return newChannelId2channel[channels[i]].GetPriority() > newChannelId2channel[channels[j]].GetPriority()
})
newGroup2model2channels[group][model] = channels
}
}
channelSyncLock.Lock()
group2model2channels = newGroup2model2channels
//channelsIDM = newChannelId2channel
for i, channel := range newChannelId2channel {
if channel.ChannelInfo.IsMultiKey {
channel.Keys = channel.GetKeys()
if channel.ChannelInfo.MultiKeyMode == constant.MultiKeyModePolling {
if oldChannel, ok := channelsIDM[i]; ok {
// 存在旧的渠道,如果是多key且轮询,保留轮询索引信息
if oldChannel.ChannelInfo.IsMultiKey && oldChannel.ChannelInfo.MultiKeyMode == constant.MultiKeyModePolling {
channel.ChannelInfo.MultiKeyPollingIndex = oldChannel.ChannelInfo.MultiKeyPollingIndex
}
}
}
}
}
channelsIDM = newChannelId2channel
channel2advancedCustomConfig = newChannel2advancedCustomConfig
channelSyncLock.Unlock()
// Lock ordering: InvalidatePricingCache acquires updatePricingLock, and
// GetPricing (holding updatePricingLock) nests channelSyncLock.RLock via
// loadPricingAdvancedCustomConfigs. channelSyncLock MUST be released before
// invalidating the pricing cache, otherwise the reversed order deadlocks.
InvalidatePricingCache()
common.SysLog("channels synced from database")
}
func SyncChannelCache(frequency int) {
for {
time.Sleep(time.Duration(frequency) * time.Second)
common.SysLog("syncing channels from database")
InitChannelCache()
}
}
func GetRandomSatisfiedChannel(group string, model string, retry int, requestPath string) (*Channel, error) {
// if memory cache is disabled, get channel directly from database
if !common.MemoryCacheEnabled {
return GetChannel(group, model, retry, requestPath)
}
channelSyncLock.RLock()
defer channelSyncLock.RUnlock()
// First, try to find channels with the exact model name.
channels := filterChannelsByRequestPathAndModel(group2model2channels[group][model], requestPath, model)
// If no channels found, try to find channels with the normalized model name.
if len(channels) == 0 {
normalizedModel := ratio_setting.FormatMatchingModelName(model)
channels = filterChannelsByRequestPathAndModel(group2model2channels[group][normalizedModel], requestPath, model)
}
if len(channels) == 0 {
return nil, nil
}
if len(channels) == 1 {
if channel, ok := channelsIDM[channels[0]]; ok {
return channel, nil
}
return nil, fmt.Errorf("数据库一致性错误,渠道# %d 不存在,请联系管理员修复", channels[0])
}
uniquePriorities := make(map[int]bool)
for _, channelId := range channels {
if channel, ok := channelsIDM[channelId]; ok {
uniquePriorities[int(channel.GetPriority())] = true
} else {
return nil, fmt.Errorf("数据库一致性错误,渠道# %d 不存在,请联系管理员修复", channelId)
}
}
var sortedUniquePriorities []int
for priority := range uniquePriorities {
sortedUniquePriorities = append(sortedUniquePriorities, priority)
}
sort.Sort(sort.Reverse(sort.IntSlice(sortedUniquePriorities)))
if retry >= len(uniquePriorities) {
retry = len(uniquePriorities) - 1
}
targetPriority := int64(sortedUniquePriorities[retry])
// get the priority for the given retry number
var sumWeight = 0
var targetChannels []*Channel
for _, channelId := range channels {
if channel, ok := channelsIDM[channelId]; ok {
if channel.GetPriority() == targetPriority {
sumWeight += channel.GetWeight()
targetChannels = append(targetChannels, channel)
}
} else {
return nil, fmt.Errorf("数据库一致性错误,渠道# %d 不存在,请联系管理员修复", channelId)
}
}
if len(targetChannels) == 0 {
return nil, errors.New(fmt.Sprintf("no channel found, group: %s, model: %s, priority: %d", group, model, targetPriority))
}
// smoothing factor and adjustment
smoothingFactor := 1
smoothingAdjustment := 0
if sumWeight == 0 {
// when all channels have weight 0, set sumWeight to the number of channels and set smoothing adjustment to 100
// each channel's effective weight = 100
sumWeight = len(targetChannels) * 100
smoothingAdjustment = 100
} else if sumWeight/len(targetChannels) < 10 {
// when the average weight is less than 10, set smoothing factor to 100
smoothingFactor = 100
}
// Calculate the total weight of all channels up to endIdx
totalWeight := sumWeight * smoothingFactor
// Generate a random value in the range [0, totalWeight)
randomWeight := rand.Intn(totalWeight)
// Find a channel based on its weight
for _, channel := range targetChannels {
randomWeight -= channel.GetWeight()*smoothingFactor + smoothingAdjustment
if randomWeight < 0 {
return channel, nil
}
}
// return null if no channel is not found
return nil, errors.New("channel not found")
}
// filterChannelsByRequestPathAndModel restricts candidates by request path and
// model. Only Advanced Custom (type 58) channels are path-checked: they are kept
// only when one of their configured routes matches requestPath and model. All
// other channel types always pass. When requestPath is empty, filtering is skipped.
// Caller must hold channelSyncLock (read lock). The cached slice is never mutated.
func filterChannelsByRequestPathAndModel(channels []int, requestPath string, model string) []int {
if requestPath == "" || len(channels) == 0 {
return channels
}
filtered := make([]int, 0, len(channels))
for _, channelId := range channels {
channel, ok := channelsIDM[channelId]
if !ok {
// keep it so the downstream consistency error is raised as before
filtered = append(filtered, channelId)
continue
}
if channel.Type != constant.ChannelTypeAdvancedCustom {
filtered = append(filtered, channelId)
continue
}
if config := channel2advancedCustomConfig[channelId]; config != nil && config.SupportsPathForModel(requestPath, model) {
filtered = append(filtered, channelId)
}
}
return filtered
}
func CacheGetChannel(id int) (*Channel, error) {
if !common.MemoryCacheEnabled {
return GetChannelById(id, true)
}
channelSyncLock.RLock()
defer channelSyncLock.RUnlock()
c, ok := channelsIDM[id]
if !ok {
return nil, fmt.Errorf("渠道# %d,已不存在", id)
}
return c, nil
}
func CacheGetChannelInfo(id int) (*ChannelInfo, error) {
if !common.MemoryCacheEnabled {
channel, err := GetChannelById(id, true)
if err != nil {
return nil, err
}
return &channel.ChannelInfo, nil
}
channelSyncLock.RLock()
defer channelSyncLock.RUnlock()
c, ok := channelsIDM[id]
if !ok {
return nil, fmt.Errorf("渠道# %d,已不存在", id)
}
return &c.ChannelInfo, nil
}
func CacheUpdateChannelStatus(id int, status int) {
if !common.MemoryCacheEnabled {
return
}
channelSyncLock.Lock()
defer channelSyncLock.Unlock()
if channel, ok := channelsIDM[id]; ok {
channel.Status = status
}
if status != common.ChannelStatusEnabled {
// delete the channel from group2model2channels
for group, model2channels := range group2model2channels {
for model, channels := range model2channels {
for i, channelId := range channels {
if channelId == id {
// remove the channel from the slice
group2model2channels[group][model] = append(channels[:i], channels[i+1:]...)
break
}
}
}
}
}
}
func CacheUpdateChannel(channel *Channel) {
if !common.MemoryCacheEnabled {
return
}
channelSyncLock.Lock()
if channel == nil {
channelSyncLock.Unlock()
return
}
if channelsIDM == nil {
channelsIDM = make(map[int]*Channel)
}
if oldChannel, ok := channelsIDM[channel.Id]; ok {
logger.LogDebug(nil, "CacheUpdateChannel before: id=%d, name=%s, status=%d, polling_index=%d", channel.Id, channel.Name, channel.Status, oldChannel.ChannelInfo.MultiKeyPollingIndex)
}
channelsIDM[channel.Id] = channel
if channel2advancedCustomConfig == nil {
channel2advancedCustomConfig = make(map[int]*dto.AdvancedCustomConfig)
}
delete(channel2advancedCustomConfig, channel.Id)
if channel.Type == constant.ChannelTypeAdvancedCustom {
if config := channel.GetOtherSettings().AdvancedCustom; config != nil {
channel2advancedCustomConfig[channel.Id] = config
}
}
logger.LogDebug(nil, "CacheUpdateChannel after: id=%d, name=%s, status=%d, polling_index=%d", channel.Id, channel.Name, channel.Status, channel.ChannelInfo.MultiKeyPollingIndex)
// Lock ordering: do NOT hold channelSyncLock while calling
// InvalidatePricingCache. GetPricing acquires updatePricingLock first and then
// channelSyncLock.RLock (via loadPricingAdvancedCustomConfigs); acquiring
// updatePricingLock while holding channelSyncLock would be an AB-BA deadlock.
channelSyncLock.Unlock()
InvalidatePricingCache()
}