Documentation
¶
Overview ¶
internal/pkg/protocol/sync_routing.go
internal/pkg/protocol/sync_user.go
Index ¶
- type CacheEntityStats
- type FetchEntityRequest
- type FetchEntityResponse
- type ForceFullSyncResponse
- type FullSyncRequest
- type FullSyncResponse
- type GetVersionResponse
- type HeartbeatParams
- type LimiterBucketStat
- type PrivateChannelInvalidatePayload
- type RoutingMember
- type SyncPushParams
- type SyncedPrivateChannel
- type SyncedRouting
- type SyncedUser
- type UsageLogEntry
- type UsageReport
- type UserQuotaSync
- type UserRoutingMap
- type VisiblePrivateChannelSet
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type CacheEntityStats ¶
type CacheEntityStats struct {
Kind string `json:"kind,omitempty"` // "lru" | "index";旧 agent 上报空串,前端按 lru 兜底
Hits int64 `json:"hits"`
Misses int64 `json:"misses"`
Evictions int64 `json:"evictions"`
NegativeHits int64 `json:"negative_hits"`
LoadErrors int64 `json:"load_errors"`
Invalidations int64 `json:"invalidations"`
Size int `json:"size"`
Capacity int `json:"capacity"`
Extra map[string]int64 `json:"extra,omitempty"` // 索引细分,如 {"limiters":12,"bindings":30}
}
CacheEntityStats 是单个实体缓存的运行统计。 LRU 模式实体上报全字段;Full 模式实体仅 Size 有意义、其他字段为 0。
type FetchEntityRequest ¶
FetchEntityRequest 是 sync.fetchEntity 的入参。 Entity 取 events.Entity* 常量;Key 由各实体 handler 解读 (token: API key 字符串;user: id 字符串)。
type FetchEntityResponse ¶
type FetchEntityResponse struct {
Found bool `json:"found"`
Data json.RawMessage `json:"data,omitempty"`
Side json.RawMessage `json:"side,omitempty"`
}
FetchEntityResponse 是 sync.fetchEntity 的响应。 Found=false 表示 master 也未查到;调用方应进入负缓存。 Side 是可选旁路负载(如 token 响应附带的 SyncedUser),由 agent 端 fetcher 解读。
type ForceFullSyncResponse ¶
type FullSyncRequest ¶
type FullSyncResponse ¶
type GetVersionResponse ¶
type GetVersionResponse struct {
Version int64 `json:"version"`
}
type HeartbeatParams ¶
type HeartbeatParams struct {
Uptime int64 `json:"uptime"`
CachedTokens int `json:"cached_tokens"`
CachedChannels int `json:"cached_channels"`
CachedModels int `json:"cached_models"`
CachedGlobalRoutings int `json:"cached_global_routings"`
CachedUserRoutings int `json:"cached_user_routings"`
ActiveConnections int `json:"active_connections"`
Version int64 `json:"version"`
HTTPAddresses json.RawMessage `json:"http_addresses,omitempty"`
Tags string `json:"tags,omitempty"`
ProxyURL string `json:"proxy_url,omitempty"`
ListenPort int `json:"listen_port,omitempty"`
CacheStats map[string]CacheEntityStats `json:"cache_stats,omitempty"`
}
type LimiterBucketStat ¶ added in v0.0.8
type LimiterBucketStat struct {
LimiterID uint `json:"limiter_id"`
Name string `json:"name"`
Bucket string `json:"bucket"`
Metric string `json:"metric"` // concurrency | rate
KeyBy string `json:"key_by"`
Occupied int64 `json:"occupied"` // 并发:当前占用; 速率:当前窗口已用
Capacity int64 `json:"capacity"`
Waiters int64 `json:"waiters"`
WindowEndMs int64 `json:"window_end_ms"` // 速率桶本窗口复位时刻(unix ms);并发为 0
}
LimiterBucketStat 是一个 (限流器,桶) 的实时仪表读数。agent 采样 + join 规则元数据后上报。
type PrivateChannelInvalidatePayload ¶ added in v0.0.2
type PrivateChannelInvalidatePayload struct {
Action string `json:"action"` // "invalidate"
AffectedUserIDs []uint `json:"affected_user_ids"`
}
PrivateChannelInvalidatePayload is the WS push notification fired after CRUD / share changes. Master expands (owner ∪ share.targets) into the affected_user_ids list; agent only sees the post-expansion user list and invalidates each user's cache block (no incremental merge).
type RoutingMember ¶
type RoutingMember struct {
Ref string `json:"ref"`
Priority int `json:"priority"`
Weight int `json:"weight"`
}
RoutingMember 与 models.RoutingMember 字段一致。protocol 包独立定义避免反向依赖 models 包。
type SyncPushParams ¶
type SyncedPrivateChannel ¶ added in v0.0.2
type SyncedPrivateChannel struct {
models.ChannelCore
OwnerID uint `json:"owner_id"`
KeyPlaintext string `json:"key"`
Models []string `json:"models"`
ModelMapping map[string]string `json:"model_mapping"`
}
SyncedPrivateChannel is the projection of a PrivateChannel that master pushes to agent over WS sync. KeyPlaintext is the master-decrypted key; agent never holds the KEK. HeaderOverride / ProxyURL never appear here (PrivateChannel schema does not include those columns).
Scalar fields shared with models.PrivateChannel come from the embedded ChannelCore. BYOK-projection-only fields (KeyPlaintext, OwnerID, plus the downstream-friendly Models / ModelMapping types) stay on the outer struct.
type SyncedRouting ¶
type SyncedRouting struct {
ID uint `json:"id"`
Name string `json:"name"`
Scope string `json:"scope"`
UserID uint `json:"user_id"`
Members []RoutingMember `json:"members"`
Enabled bool `json:"enabled"`
}
SyncedRouting 是 master 推送给 agent 的 ModelRouting 投影。 与 models.ModelRouting 保持字段对等,但脱离 GORM 依赖(agent 端不需要持久化)。
type SyncedUser ¶
type SyncedUser struct {
ID uint `json:"id"`
GroupID uint `json:"group_id"`
Quota int64 `json:"quota"`
}
SyncedUser 是 master 推送给 agent 的精简 User 投影。 只包含鉴权和配额所需字段,不暴露 Password / Username。
type UsageLogEntry ¶
type UsageLogEntry struct {
RequestID string `json:"request_id"`
UserID uint `json:"user_id"`
TokenID uint `json:"token_id"`
ChannelID uint `json:"channel_id"`
PrivateChannelID uint `json:"private_channel_id"`
OwnerType string `json:"owner_type"`
ModelName string `json:"model_name"`
PromptTokens int `json:"prompt_tokens"`
CompletionTokens int `json:"completion_tokens"`
IsStream bool `json:"is_stream"`
Duration int `json:"duration"`
ClientIP string `json:"client_ip"`
Timestamp int64 `json:"timestamp"`
// Enhanced logging fields
TokenName string `json:"token_name"`
UpstreamModel string `json:"upstream_model"`
FirstResponseMs int `json:"first_response_ms"`
CacheReadTokens int `json:"cache_read_tokens"`
CacheWriteTokens int `json:"cache_write_tokens"`
InboundProtocol string `json:"inbound_protocol"`
OutboundProtocol string `json:"outbound_protocol"`
UseLegacy bool `json:"use_legacy"`
Status int `json:"status"`
ErrorMessage string `json:"error_message,omitempty"`
TraceData string `json:"trace_data,omitempty"`
Other string `json:"other"`
TokenSource string `json:"token_source,omitempty"`
RoutingName string `json:"routing_name,omitempty"`
// Channel affinity(渠道缓存粘性)
AffinityStatus string `json:"affinity_status,omitempty"` // "hit" | "fallback" | "none" | ""
AffinityRecorded bool `json:"affinity_recorded,omitempty"`
ErrorStage string `json:"error_stage,omitempty"`
InboundDecodeMs int `json:"inbound_decode_ms,omitempty"`
OutboundEncodeMs int `json:"outbound_encode_ms,omitempty"`
UpstreamDispatchMs int `json:"upstream_dispatch_ms,omitempty"`
UpstreamDecodeMs int `json:"upstream_decode_ms,omitempty"`
ClientEncodeMs int `json:"client_encode_ms,omitempty"`
// PriceRatio 是 relay 选中的公共 channel 在请求时刻的计费倍率快照。
// 零值 0(旧 agent 未发 / channel 未配)= 原价,settler 归一到 1.0。
PriceRatio float64 `json:"price_ratio,omitempty"`
// Free 是 relay 选中的公共 channel 在请求时刻的"免费渠道"标记快照。
// private 行不填,保持 false。settler 见 true 则四桶成本清零。
Free bool `json:"free,omitempty"`
// FallbackChain 是本次请求的候选链路(每候选一条);空表示无链路信息。
FallbackChain []models.AttemptRecord `json:"fallback_chain,omitempty"`
// AttemptTraces 是逐候选 mask 过的 header/body trace,index 对应 FallbackChain 的 Seq-1。
// AttemptIndex assigned by settler from slice order until Task 5 adds the field.
AttemptTraces []models.UsageLogTrace `json:"attempt_traces,omitempty"`
// 限流决策(请求级 limiter):Decision/WaitMs/Reason 标量 + Hits 命中明细。
// 无命中时全为零值/空,omitempty 不上报。
RateLimitDecision string `json:"rate_limit_decision,omitempty"`
RateLimitWaitMs int `json:"rate_limit_wait_ms,omitempty"`
RateLimitReason string `json:"rate_limit_reason,omitempty"`
RateLimitHits []models.RateLimitHit `json:"rate_limit_hits,omitempty"`
}
type UsageReport ¶
type UsageReport struct {
AgentID string `json:"agent_id"`
Logs []UsageLogEntry `json:"logs"`
}
type UserQuotaSync ¶ added in v0.0.8
type UserQuotaSync struct {
AgentID string `json:"agent_id"`
Users []SyncedUser `json:"users"`
}
UserQuotaSync 是 master 结算后定向回送给来源 agent 的余额更新。
type UserRoutingMap ¶
type UserRoutingMap struct {
Routings map[string]*SyncedRouting `json:"routings"`
}
UserRoutingMap 是单个用户的 routing 集合(name → routing), 按用户整体存取,避免每条 user routing 单独缓存。 放在 protocol 包是因为 loaders 与 cache 都需要引用此类型, 而 loaders 包不能反向 import cache。
type VisiblePrivateChannelSet ¶ added in v0.0.2
type VisiblePrivateChannelSet struct {
UserID uint `json:"user_id"`
Channels []SyncedPrivateChannel `json:"channels"`
}
VisiblePrivateChannelSet is the entire block returned to agent on LRU miss: all private channels visible to the requesting user, with plaintext keys injected.