2026-01-23 17:05:09 +09:00
|
|
|
package service
|
2026-01-10 18:16:50 +09:00
|
|
|
|
2026-01-23 17:05:09 +09:00
|
|
|
import kotlinx.coroutines.CoroutineScope
|
|
|
|
|
import kotlinx.coroutines.Dispatchers
|
|
|
|
|
import kotlinx.coroutines.SupervisorJob
|
|
|
|
|
import kotlinx.coroutines.launch
|
2026-01-10 18:16:50 +09:00
|
|
|
import java.io.BufferedReader
|
2026-01-23 17:05:09 +09:00
|
|
|
import java.io.File
|
2026-01-10 18:16:50 +09:00
|
|
|
import java.io.InputStreamReader
|
2026-01-21 18:30:03 +09:00
|
|
|
import java.util.concurrent.ConcurrentHashMap
|
2026-01-10 18:16:50 +09:00
|
|
|
|
|
|
|
|
object LlamaServerManager {
|
2026-01-21 18:30:03 +09:00
|
|
|
// 포트별로 프로세스를 관리합니다.
|
|
|
|
|
private val processes = ConcurrentHashMap<Int, Process>()
|
|
|
|
|
private val scope = CoroutineScope(Dispatchers.IO + SupervisorJob())
|
|
|
|
|
init {
|
|
|
|
|
Runtime.getRuntime().addShutdownHook(Thread {
|
|
|
|
|
stopAll()
|
|
|
|
|
})
|
|
|
|
|
}
|
2026-01-10 18:16:50 +09:00
|
|
|
|
2026-01-21 18:30:03 +09:00
|
|
|
fun startServer(binPath: String, modelPath: String, port: Int, nGpuLayers: Int = 99) {
|
|
|
|
|
// 이미 해당 포트에서 실행 중이거나 모델 경로가 비었으면 무시합니다.
|
|
|
|
|
if (processes.containsKey(port) || modelPath.isBlank()) return
|
2026-01-10 18:16:50 +09:00
|
|
|
|
|
|
|
|
val command = listOf(
|
|
|
|
|
binPath,
|
|
|
|
|
"-m", modelPath,
|
2026-01-21 18:30:03 +09:00
|
|
|
"--port", port.toString(),
|
2026-01-23 17:05:09 +09:00
|
|
|
"-c", if (port == 8081) "512" else "8192", // 임베딩용은 컨텍스트가 짧아도 충분합니다.
|
2026-01-21 18:30:03 +09:00
|
|
|
"-ngl", nGpuLayers.toString(),
|
2026-02-06 17:53:17 +09:00
|
|
|
"-t", "8", // M3 Pro의 성능 코어를 고려하여 6~8개 권장
|
2026-01-21 18:30:03 +09:00
|
|
|
"--embedding" // 임베딩 기능을 활성화합니다.
|
2026-01-10 18:16:50 +09:00
|
|
|
)
|
|
|
|
|
|
|
|
|
|
scope.launch {
|
|
|
|
|
try {
|
|
|
|
|
val pb = ProcessBuilder(command)
|
2026-01-21 18:30:03 +09:00
|
|
|
|
|
|
|
|
pb.redirectErrorStream(true)
|
2026-01-10 18:16:50 +09:00
|
|
|
File(binPath).setExecutable(true)
|
|
|
|
|
|
2026-01-21 18:30:03 +09:00
|
|
|
val process = pb.start()
|
|
|
|
|
processes[port] = process
|
|
|
|
|
println("✅ AI 서버 시작 시도 (Port: $port, Model: ${File(modelPath).name})")
|
2026-01-10 18:16:50 +09:00
|
|
|
|
2026-01-21 18:30:03 +09:00
|
|
|
val reader = BufferedReader(InputStreamReader(process.inputStream))
|
2026-01-10 18:16:50 +09:00
|
|
|
var line: String?
|
|
|
|
|
while (reader.readLine().also { line = it } != null) {
|
2026-01-21 18:30:03 +09:00
|
|
|
// 로그 출력 (디버깅용)
|
2026-01-23 17:05:09 +09:00
|
|
|
// println("[Server $port] $line")
|
2026-01-21 18:30:03 +09:00
|
|
|
if (line?.contains("server is listening") == true) {
|
|
|
|
|
println("🚀 AI 서버 준비 완료 (Port: $port)")
|
2026-01-23 17:05:09 +09:00
|
|
|
if (processes.size > 1) {
|
|
|
|
|
println("[Cache] ${processes.size}")
|
|
|
|
|
RagService.active()
|
|
|
|
|
}
|
2026-01-10 18:16:50 +09:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
} catch (e: Exception) {
|
2026-01-21 18:30:03 +09:00
|
|
|
println("❌ AI 서버 실행 실패 (Port: $port): ${e.message}")
|
|
|
|
|
processes.remove(port)
|
2026-01-10 18:16:50 +09:00
|
|
|
}
|
2026-01-23 17:05:09 +09:00
|
|
|
|
2026-01-10 18:16:50 +09:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2026-01-21 18:30:03 +09:00
|
|
|
fun stopAll() {
|
|
|
|
|
processes.forEach { (port, process) ->
|
|
|
|
|
process.destroy()
|
|
|
|
|
println("🛑 AI 서버 종료 (Port: $port)")
|
|
|
|
|
}
|
|
|
|
|
processes.clear()
|
2026-01-10 18:16:50 +09:00
|
|
|
}
|
|
|
|
|
}
|