feat(llm): 每个 LLM 配置支持自定义会话头
新增 session_header_key:非空时每次 LLM 请求带上该 HTTP 头,头值为当前 运行的 session id(chat 会话=conv-<id>,worker=exp<x>-worker-i<intent> 等)。用于某些按 session-id 头做提示缓存/粘性路由的网关。 实现:在已有的 quotaAwareTransport 里从请求 context 读取 session id (transcript.WithSessionID 挂上的)并注入,无需改 norma;同一共享 provider 也能按会话发出不同头值。DB 加 session_header_key 列(带旧库迁移),前端 LLM 配置弹窗新增输入框。
This commit is contained in:
Autumn-27 committed
1 parent
b513b656aa
commit
4319fe413f
10 files changed
+166
-36
No files matched your search
+16
-10
@@ -55,19 +55,25 @@ type LLMProfile struct {
|
||||
// reasoning tokens plus visible output. Ignored by anthropic and
|
||||
// openai-responses, which name the field themselves.
|
||||
MaxTokensField string `json:"max_tokens_field"`
|
||||
// SessionHeaderKey, when non-empty, names a custom HTTP header sent on every
|
||||
// request built from this profile; its value is the current run's session id
|
||||
// (chat conversation / worker intent). For gateways that key prompt caching
|
||||
// or sticky routing off a session-id header. "" = not sent. Maps to
|
||||
// agent.Config.SessionHeaderKey.
|
||||
SessionHeaderKey string `json:"session_header_key"`
|
||||
}
|
||||
|
||||
// profileCols is the read column list (hint variant, no api key) shared by the
|
||||
// list query; profileColsKey is the same with api_key for the single-row loads.
|
||||
const profileCols = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key_hint,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude,COALESCE(thinking_type,''),COALESCE(streaming,true),COALESCE(max_tokens,0),COALESCE(max_tokens_field,'')`
|
||||
const profileColsKey = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude,COALESCE(thinking_type,''),COALESCE(streaming,true),COALESCE(max_tokens,0),COALESCE(max_tokens_field,'')`
|
||||
const profileCols = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key_hint,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude,COALESCE(thinking_type,''),COALESCE(streaming,true),COALESCE(max_tokens,0),COALESCE(max_tokens_field,''),COALESCE(session_header_key,'')`
|
||||
const profileColsKey = `id,name,format,COALESCE(base_url,''),COALESCE(proxy,''),model,COALESCE(api_key,''),rate_per_second,rate_per_minute,context_window_k,COALESCE(reasoning_effort,''),is_default,priority,pool_exclude,COALESCE(thinking_type,''),COALESCE(streaming,true),COALESCE(max_tokens,0),COALESCE(max_tokens_field,''),COALESCE(session_header_key,'')`
|
||||
|
||||
// scanProfile reads one row in the profileCols / profileColsKey column order. The
|
||||
// 7th column lands in APIKeyHint or APIKey depending on which list the caller used.
|
||||
func scanProfile(sc interface{ Scan(...any) error }, into *string, p *LLMProfile) error {
|
||||
return sc.Scan(&p.ID, &p.Name, &p.Format, &p.BaseURL, &p.Proxy, &p.Model, into,
|
||||
&p.RatePerSecond, &p.RatePerMinute, &p.ContextWindowK, &p.ReasoningEffort, &p.IsDefault, &p.Priority, &p.PoolExclude, &p.ThinkingType, &p.Streaming,
|
||||
&p.MaxTokens, &p.MaxTokensField)
|
||||
&p.MaxTokens, &p.MaxTokensField, &p.SessionHeaderKey)
|
||||
}
|
||||
|
||||
func (d *DB) ListProfiles() ([]*LLMProfile, error) {
|
||||
@@ -140,18 +146,18 @@ func (d *DB) SaveProfile(p *LLMProfile) (int64, error) {
|
||||
}
|
||||
if p.ID == 0 {
|
||||
var id int64
|
||||
err := d.QueryRow(`INSERT INTO llm_profiles(name,format,base_url,proxy,model,api_key,api_key_hint,rate_per_second,rate_per_minute,context_window_k,reasoning_effort,priority,pool_exclude,thinking_type,streaming,max_tokens,max_tokens_field)
|
||||
VALUES ($1,$2,NULLIF($3,''),NULLIF($4,''),$5,NULLIF($6,''),NULLIF($7,''),$8,$9,$10,$11,$12,$13,$14,$15,$16,$17) RETURNING id`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.APIKey, hint, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType, p.Streaming, p.MaxTokens, p.MaxTokensField).Scan(&id)
|
||||
err := d.QueryRow(`INSERT INTO llm_profiles(name,format,base_url,proxy,model,api_key,api_key_hint,rate_per_second,rate_per_minute,context_window_k,reasoning_effort,priority,pool_exclude,thinking_type,streaming,max_tokens,max_tokens_field,session_header_key)
|
||||
VALUES ($1,$2,NULLIF($3,''),NULLIF($4,''),$5,NULLIF($6,''),NULLIF($7,''),$8,$9,$10,$11,$12,$13,$14,$15,$16,$17,NULLIF($18,'')) RETURNING id`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.APIKey, hint, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType, p.Streaming, p.MaxTokens, p.MaxTokensField, p.SessionHeaderKey).Scan(&id)
|
||||
return id, err
|
||||
}
|
||||
if p.APIKey == "" {
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,rate_per_second=$6,rate_per_minute=$7,context_window_k=$8,reasoning_effort=$9,priority=$10,pool_exclude=$11,thinking_type=$12,streaming=$13,max_tokens=$14,max_tokens_field=$15 WHERE id=$16`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType, p.Streaming, p.MaxTokens, p.MaxTokensField, p.ID)
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,rate_per_second=$6,rate_per_minute=$7,context_window_k=$8,reasoning_effort=$9,priority=$10,pool_exclude=$11,thinking_type=$12,streaming=$13,max_tokens=$14,max_tokens_field=$15,session_header_key=NULLIF($16,'') WHERE id=$17`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType, p.Streaming, p.MaxTokens, p.MaxTokensField, p.SessionHeaderKey, p.ID)
|
||||
return p.ID, err
|
||||
}
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,api_key=$6,api_key_hint=$7,rate_per_second=$8,rate_per_minute=$9,context_window_k=$10,reasoning_effort=$11,priority=$12,pool_exclude=$13,thinking_type=$14,streaming=$15,max_tokens=$16,max_tokens_field=$17 WHERE id=$18`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.APIKey, hint, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType, p.Streaming, p.MaxTokens, p.MaxTokensField, p.ID)
|
||||
_, err := d.Exec(`UPDATE llm_profiles SET name=$1,format=$2,base_url=NULLIF($3,''),proxy=NULLIF($4,''),model=$5,api_key=$6,api_key_hint=$7,rate_per_second=$8,rate_per_minute=$9,context_window_k=$10,reasoning_effort=$11,priority=$12,pool_exclude=$13,thinking_type=$14,streaming=$15,max_tokens=$16,max_tokens_field=$17,session_header_key=NULLIF($18,'') WHERE id=$19`,
|
||||
p.Name, p.Format, p.BaseURL, p.Proxy, p.Model, p.APIKey, hint, p.RatePerSecond, p.RatePerMinute, p.ContextWindowK, p.ReasoningEffort, p.Priority, p.PoolExclude, p.ThinkingType, p.Streaming, p.MaxTokens, p.MaxTokensField, p.SessionHeaderKey, p.ID)
|
||||
return p.ID, err
|
||||
}
|
||||
|
||||
|
||||
@@ -321,6 +321,9 @@ CREATE TABLE IF NOT EXISTS llm_profiles (
|
||||
-- 发 max_tokens 会被 unsupported_parameter 拒绝。
|
||||
-- anthropic(max_tokens 必填)与 openai-responses(max_output_tokens)自带字段名，不受此值影响。
|
||||
max_tokens_field TEXT NOT NULL DEFAULT '',
|
||||
-- 自定义会话头：非空时每次请求带一个该名字的 HTTP 头，头值=当前运行的 session id
|
||||
-- (chat 会话/worker 意图)。用于某些按 session-id 头做提示缓存/粘性路由的网关。''=不发送。
|
||||
session_header_key TEXT NOT NULL DEFAULT '',
|
||||
created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
||||
updated_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
);
|
||||
@@ -350,6 +353,8 @@ ALTER TABLE llm_profiles ADD CONSTRAINT llm_profiles_max_tokens_field_check
|
||||
ALTER TABLE llm_profiles DROP CONSTRAINT IF EXISTS llm_profiles_max_tokens_check;
|
||||
ALTER TABLE llm_profiles ADD CONSTRAINT llm_profiles_max_tokens_check
|
||||
CHECK (max_tokens >= 0);
|
||||
-- 自定义会话头名；补旧库。默认 '' = 不发送，旧配置行为不变。
|
||||
ALTER TABLE llm_profiles ADD COLUMN IF NOT EXISTS session_header_key TEXT NOT NULL DEFAULT '';
|
||||
|
||||
-- 思考开关字段 thinking_type，从旧的单一 reasoning_effort 语义一次性拆分而来。
|
||||
-- schema.sql 每次启动都执行，故迁移必须只跑一次：仅当该列尚不存在时才回填，
|
||||
|
||||
Reference in new issue
Block a user