easyai-ai-gateway/apps/api/internal/runner/cache_affinity.go

131 lines
3.5 KiB
Go

package runner
import (
"crypto/sha256"
"encoding/hex"
"encoding/json"
"strings"
)
type cacheAffinityKeys struct {
Primary string
Lookup []string
Record []string
}
func buildCacheAffinityKey(kind string, modelType string, body map[string]any) string {
return buildCacheAffinityKeys(kind, modelType, body).Primary
}
func buildCacheAffinityKeys(kind string, modelType string, body map[string]any) cacheAffinityKeys {
if len(body) == 0 {
return cacheAffinityKeys{}
}
for _, key := range []string{"cacheAffinityKey", "cache_affinity_key"} {
if value := strings.TrimSpace(stringFromAny(body[key])); value != "" {
key := "explicit:" + sha256Text(value)
return cacheAffinityKeys{Primary: key, Lookup: []string{key}, Record: []string{key}}
}
}
for _, key := range []string{"sessionId", "session_id", "conversationId", "conversation_id"} {
if value := strings.TrimSpace(stringFromAny(body[key])); value != "" {
key := key + ":" + sha256Text(value)
return cacheAffinityKeys{Primary: key, Lookup: []string{key}, Record: []string{key}}
}
}
if !cacheAffinityPromptHashSupported(kind, modelType) {
return cacheAffinityKeys{}
}
basePayload := map[string]any{
"kind": kind,
"modelType": modelType,
}
if tools, ok := body["tools"]; ok {
basePayload["tools"] = tools
}
if instructions, ok := body["instructions"]; ok {
basePayload["instructions"] = instructions
}
messageHashes := cacheAffinityMessageHashes(body)
if len(messageHashes) > 0 {
baseHash := stableJSONHash(basePayload)
keys := make([]string, 0, len(messageHashes))
for index := range messageHashes {
keyPayload := map[string]any{
"base": baseHash,
"messageHashes": messageHashes[:index+1],
"prefixLength": index + 1,
}
keys = append(keys, "prompt_lcp:"+stableJSONHash(keyPayload))
}
if len(keys) == 0 {
return cacheAffinityKeys{}
}
lookup := make([]string, 0, len(keys))
for index := len(keys) - 1; index >= 0; index-- {
lookup = append(lookup, keys[index])
}
return cacheAffinityKeys{
Primary: keys[len(keys)-1],
Lookup: lookup,
Record: keys,
}
}
if input, ok := body["input"]; ok {
basePayload["input"] = input
}
if len(basePayload) <= 2 {
return cacheAffinityKeys{}
}
key := "prompt:" + stableJSONHash(basePayload)
return cacheAffinityKeys{Primary: key, Lookup: []string{key}, Record: []string{key}}
}
func cacheAffinityPromptHashSupported(kind string, modelType string) bool {
switch strings.TrimSpace(kind) {
case "chat.completions", "responses":
return true
}
switch strings.TrimSpace(modelType) {
case "chat", "text_generate", "responses":
return true
default:
return false
}
}
func cacheAffinityMessageHashes(body map[string]any) []string {
if hashes := stringListFromAny(body["messageHashes"]); len(hashes) > 0 {
return hashes
}
if hashes := stringListFromAny(body["message_hashes"]); len(hashes) > 0 {
return hashes
}
messages, ok := body["messages"].([]any)
if !ok || len(messages) == 0 {
return nil
}
hashes := make([]string, 0, len(messages))
for _, message := range messages {
raw, err := json.Marshal(message)
if err != nil || len(raw) == 0 {
continue
}
hashes = append(hashes, sha256Text(string(raw)))
}
return hashes
}
func stableJSONHash(value any) string {
raw, err := json.Marshal(value)
if err != nil || len(raw) == 0 {
return ""
}
return sha256Text(string(raw))
}
func sha256Text(value string) string {
sum := sha256.Sum256([]byte(value))
return hex.EncodeToString(sum[:])
}