From 5f8bb8afda9df119ae529b4afa41ab1c7d1c1ea1 Mon Sep 17 00:00:00 2001
From: yumoqing <yumoqing@gmail.com>
Date: Sun, 14 Jun 2026 15:35:31 +0800
Subject: [PATCH] Add API endpoints (status/submit/task)

---
 app/api/status/index.dspy | 31 +++++++++++++++++++++++
 app/api/submit/index.dspy | 53 +++++++++++++++++++++++++++++++++++++++
 app/api/task/index.dspy   | 17 +++++++++++++
 app/health.dspy           |  8 +++---
 start.sh                  |  6 ++---
 5 files changed, 107 insertions(+), 8 deletions(-)
 create mode 100644 app/api/status/index.dspy
 create mode 100644 app/api/submit/index.dspy
 create mode 100644 app/api/task/index.dspy

diff --git a/app/api/status/index.dspy b/app/api/status/index.dspy
new file mode 100644
index 0000000..8e677e2
--- /dev/null
+++ b/app/api/status/index.dspy
@@ -0,0 +1,31 @@
+# -*- coding:utf-8 -*-
+# GET /api/status - ASR服务状态
+
+import subprocess
+import json
+
+result = {
+    'service': 'asr-transcription',
+    'model': 'faster-whisper-large-v3-turbo-ct2',
+    'gpu_id': 6,
+    'gpus': []
+}
+
+try:
+    out = subprocess.check_output(
+        ['nvidia-smi', '--query-gpu=index,utilization.gpu,memory.used,memory.total',
+         '--format=csv,noheader,nounits'],
+        timeout=5
+    ).decode().strip()
+    for line in out.split('\n'):
+        parts = [p.strip() for p in line.split(',')]
+        result['gpus'].append({
+            'id': int(parts[0]),
+            'util': int(parts[1]),
+            'mem_used': int(parts[2]),
+            'mem_total': int(parts[3])
+        })
+except Exception:
+    pass
+
+return json.dumps(result)
diff --git a/app/api/submit/index.dspy b/app/api/submit/index.dspy
new file mode 100644
index 0000000..1de0137
--- /dev/null
+++ b/app/api/submit/index.dspy
@@ -0,0 +1,53 @@
+# -*- coding:utf-8 -*-
+# POST /api/submit - 提交ASR转录任务
+
+import json
+import uuid
+from ahserver.serverenv import ServerEnv
+
+method = request.method
+
+if method == 'POST':
+    audio_path = params_kw.get('audio_path', '')
+    if not audio_path:
+        return json.dumps({'error': 'audio_path is required'}, ensure_ascii=False)
+
+    task_id = params_kw.get('task_id', str(uuid.uuid4()).replace("-", "")[:12])
+    language = params_kw.get('language', 'auto')
+    beam_size = params_kw.get('beam_size', 5)
+
+    payload = {
+        'task_type': 'transcribe',
+        'task_id': task_id,
+        'audio_path': audio_path,
+        'language': language,
+        'beam_size': int(beam_size)
+    }
+
+    env = ServerEnv()
+    longtasks = env.longtasks
+    if longtasks is None:
+        return json.dumps({'error': 'service not ready'}, ensure_ascii=False)
+
+    result = await longtasks.submit_task(payload)
+    real_task_id = result.get('task_id', str(result)) if isinstance(result, dict) else str(result)
+
+    return json.dumps({
+        'task_id': real_task_id,
+        'status': 'queued',
+        'audio_path': audio_path,
+        'language': language,
+        'message': 'task submitted',
+        'check_url': f'/api/task?task_id={real_task_id}'
+    }, ensure_ascii=False)
+
+else:
+    return json.dumps({
+        'usage': 'POST with JSON body',
+        'params': {
+            'audio_path': 'string (required, server path to audio file)',
+            'language': 'string (default auto, or zh/en/ja/ko etc)',
+            'beam_size': 'int (default 5)',
+            'task_id': 'string (optional, auto-generated)',
+        }
+    }, ensure_ascii=False)
diff --git a/app/api/task/index.dspy b/app/api/task/index.dspy
new file mode 100644
index 0000000..a1137df
--- /dev/null
+++ b/app/api/task/index.dspy
@@ -0,0 +1,17 @@
+# -*- coding:utf-8 -*-
+# GET /api/task?task_id=xxx - 查询任务状态
+
+import json
+from ahserver.serverenv import ServerEnv
+
+task_id = params_kw.get('task_id', '')
+if not task_id:
+    return json.dumps({'error': 'task_id is required'}, ensure_ascii=False)
+
+env = ServerEnv()
+longtasks = env.longtasks
+if longtasks is None:
+    return json.dumps({'error': 'service not ready'}, ensure_ascii=False)
+
+status = await longtasks.get_status(task_id)
+return json.dumps(status)
diff --git a/app/health.dspy b/app/health.dspy
index 9734c2a..b7028a1 100644
--- a/app/health.dspy
+++ b/app/health.dspy
@@ -1,5 +1,3 @@
-{{
-    "status": "ok",
-    "service": "asr-service",
-    "model": "faster-whisper-large-v3-turbo-ct2"
-}}
+import json
+result = {"status": "ok", "service": "$svc"}
+print(json.dumps(result))
diff --git a/start.sh b/start.sh
index 9731884..0460166 100755
--- a/start.sh
+++ b/start.sh
@@ -1,7 +1,7 @@
 #!/bin/bash
 cd /data/ymq/asr-service
-export ASR_GPU_ID=5
-export CUDA_VISIBLE_DEVICES=5
+export ASR_GPU_ID=6
+export CUDA_VISIBLE_DEVICES=6
 export PYTHONPATH=/data/ymq/asr-service
-nohup /data/ymq/demucs_venv/bin/python ah.py > nohup.out 2>&1 &
+nohup /data/ymq/wan22-service/py3/bin/python ah.py > nohup.out 2>&1 &
 echo "asr-service started, PID: $!, GPU: $ASR_GPU_ID"