Reraise Jinja2 TemplateError (#840)

* Handle jinja2.TemplateError in gateway

* Raise GatewayError in openai interface

Author: Egor-S

gateway/src/dstack/gateway/openai/clients/tgi.py

@@ -6,6 +6,7 @@
 
 import httpx
 import jinja2
+import jinja2.sandbox
 
 from dstack.gateway.errors import GatewayError
 from dstack.gateway.openai.clients import ChatCompletionsClient
@@ -31,9 +32,17 @@ def __init__(
             headers={} if host is None else {"Host": host},
             timeout=60,
         )
-        self.chat_template = jinja2.Template(chat_template)
         self.eos_token = eos_token
 
+        try:
+            jinja_env = jinja2.sandbox.ImmutableSandboxedEnvironment(
+                trim_blocks=True, lstrip_blocks=True
+            )
+            jinja_env.globals["raise_exception"] = raise_exception
+            self.chat_template = jinja_env.from_string(chat_template)
+        except jinja2.TemplateError as e:
+            raise GatewayError(f"Failed to compile chat template: {e}")
+
     async def generate(self, request: ChatCompletionsRequest) -> ChatCompletionsResponse:
         payload = self.get_payload(request)
         resp = await self.client.post("/generate", json=payload)
@@ -123,10 +132,14 @@ async def stream(self, request: ChatCompletionsRequest) -> AsyncIterator[ChatCom
                     yield chunk
 
     def get_payload(self, request: ChatCompletionsRequest) -> Dict:
-        inputs = self.chat_template.render(
-            messages=request.messages,
-            add_generation_prompt=True,
-        )
+        try:
+            inputs = self.chat_template.render(
+                messages=request.messages,
+                add_generation_prompt=True,
+            )
+        except jinja2.TemplateError as e:
+            raise GatewayError(f"Failed to render chat template: {e}")
+
         stop = ([request.stop] if isinstance(request.stop, str) else request.stop) or []
         if self.eos_token not in stop:
             stop.append(self.eos_token)
@@ -178,3 +191,7 @@ def __del__(self):
             asyncio.get_running_loop().create_task(self.aclose())
         except Exception:
             pass
+
+
+def raise_exception(message: str):
+    raise jinja2.TemplateError(message)

gateway/src/dstack/gateway/openai/routes.py

@@ -3,6 +3,7 @@
 from fastapi import APIRouter, Depends
 from fastapi.responses import StreamingResponse
 
+from dstack.gateway.errors import GatewayError
 from dstack.gateway.openai.schemas import (
     ChatCompletionsChunk,
     ChatCompletionsRequest,
@@ -25,14 +26,17 @@ async def get_models(
 async def post_chat_completions(
     project: str, body: ChatCompletionsRequest, store: Annotated[OpenAIStore, Depends(get_store)]
 ):
-    client = await store.get_chat_client(project, body.model)
-    if not body.stream:
-        return await client.generate(body)
-    else:
-        return StreamingResponse(
-            stream_chunks(client.stream(body)),
-            media_type="text/event-stream",
-        )
+    try:
+        client = await store.get_chat_client(project, body.model)
+        if not body.stream:
+            return await client.generate(body)
+        else:
+            return StreamingResponse(
+                stream_chunks(client.stream(body)),
+                media_type="text/event-stream",
+            )
+    except GatewayError as e:
+        raise e.http()
 
 
 async def stream_chunks(chunks: AsyncIterator[ChatCompletionsChunk]) -> AsyncIterator[bytes]: