Better error reporting in frontend

Signed-off-by: Antoni Baum <[email protected]>

aviary/common/backend.py

@@ -7,6 +7,12 @@
 from aviary.common.constants import TIMEOUT
 
 
+class BackendError(RuntimeError):
+    def __init__(self, *args: object, **kwargs) -> None:
+        self.response = kwargs.pop("response", None)
+        super().__init__(*args)
+
+
 def get_aviary_backend():
     """
     Establishes a connection to the Aviary backed after establishing
@@ -97,23 +103,25 @@ def __init__(self, backend_url: str, bearer: str):
 
     def models(self) -> List[str]:
         url = self.backend_url + "models"
-        resp = requests.get(url, headers=self.header, timeout=120)
+        response = requests.get(url, headers=self.header, timeout=TIMEOUT)
         try:
-            result = resp.json()
+            result = response.json()
         except requests.JSONDecodeError as e:
-            raise RuntimeError(
-                f"Error decoding JSON from {url}. Text response: {resp.text}",
+            raise BackendError(
+                f"Error decoding JSON from {url}. Text response: {response.text}",
+                response=response,
             ) from e
         return result
 
     def metadata(self, llm: str) -> Dict[str, Dict[str, Any]]:
         url = self.backend_url + "metadata/" + llm.replace("/", "--")
-        resp = requests.get(url, headers=self.header, timeout=120)
+        response = requests.get(url, headers=self.header, timeout=TIMEOUT)
         try:
-            result = resp.json()
+            result = response.json()
         except requests.JSONDecodeError as e:
-            raise RuntimeError(
-                f"Error decoding JSON from {url}. Text response: {resp.text}",
+            raise BackendError(
+                f"Error decoding JSON from {url}. Text response: {response.text}",
+                response=response,
             ) from e
         return result
 
@@ -128,8 +136,9 @@ def completions(self, prompt: str, llm: str) -> Dict[str, Union[str, float, int]
         try:
             return response.json()[llm]
         except requests.JSONDecodeError as e:
-            raise RuntimeError(
+            raise BackendError(
                 f"Error decoding JSON from {url}. Text response: {response.text}",
+                response=response,
             ) from e
 
     def batch_completions(
@@ -145,8 +154,9 @@ def batch_completions(
         try:
             return response.json()[llm]
         except requests.JSONDecodeError as e:
-            raise RuntimeError(
+            raise BackendError(
                 f"Error decoding JSON from {url}. Text response: {response.text}",
+                response=response,
             ) from e
 
 

aviary/frontend/app.py

@@ -5,10 +5,11 @@
 
 import gradio as gr
 import ray
+import requests
 from ray import serve
 from ray.serve.gradio_integrations import GradioIngress
 
-from aviary.common.backend import get_aviary_backend
+from aviary.common.backend import BackendError, get_aviary_backend
 from aviary.common.constants import (
     AVIARY_DESC,
     CSS,
@@ -75,14 +76,41 @@ def completions(prompt, llm):
 
 
 def do_query(prompt, model1, model2, model3, unused_raw=None):
-    models = [model1, model2, model3]
-    futures = [completions.remote(prompt, model) for model in models]
-    outs = ray.get(futures)
-
-    text_output = [o["generated_text"] for o in outs]
-    stats = [gen_stats(o) for o in outs]
-
-    return [*text_output, *stats, "", outs]
+    try:
+        models = [model1, model2, model3]
+        futures = [completions.remote(prompt, model) for model in models]
+        outs = ray.get(futures)
+
+        text_output = [o["generated_text"] for o in outs]
+        stats = [gen_stats(o) for o in outs]
+
+        return [*text_output, *stats, "", outs]
+    except requests.ReadTimeout as e:
+        raise gr.Error(
+            (
+                "The request timed out. This usually means the server is experiencing a higher than usual load. "
+                "Please try again in a few minutes."
+            ).replace("\n", " ")
+        ) from e
+    except BackendError as e:
+        if "timeout" in e.response or e.response.status_code in (408, 504):
+            raise gr.Error(
+                (
+                    f"The request timed out. This usually means the server is experiencing a higher than usual load. "
+                    f"Please try again in a few minutes.\nStatus code: {e.response.status_code}"
+                    f"\nResponse: {e.response.text.split('raise ')[-1]}"
+                ).replace("\n", " ")
+            ) from e
+        else:
+            raise gr.Error(
+                (
+                    f"Backend returned an error. "
+                    f"Status code: {e.response.status_code}"
+                    f"\nResponse: {e.response.text.split('raise ')[-1]}"
+                ).replace("\n", " ")
+            ) from e
+    except Exception as e:
+        raise gr.Error(f"An error occurred. Please try again.\nError: {e}") from e
 
 
 def show_results(buttons, llm_text_boxes, llm_stats):