reduce sentry errors

bot/kodiak/entrypoints/ingest.py

@@ -87,7 +87,7 @@ async def webhook_event(
             status_code=status.HTTP_400_BAD_REQUEST,
             detail="Invalid signature: X-Hub-Signature",
         )
-    log = logger.bind(event_name=x_github_event, event=event)
+    log = logger.bind(event_name=x_github_event)
     installation_id: int | None = event.get("installation", {}).get("id")
 
     if github_event in {

bot/kodiak/errors.py

@@ -1,3 +1,11 @@
+from __future__ import annotations
+
+from collections.abc import Iterable
+from typing import Any
+
+from typing_extensions import Literal
+
+
 class RetryForSkippableChecks(Exception):
     pass
 
@@ -15,3 +23,31 @@ def __init__(self, method: str, http_status_code: int, response: bytes) -> None:
 
 class GitHubApiInternalServerError(Exception):
     pass
+
+
+def identify_github_graphql_error(
+    errors: Iterable[Any],
+) -> set[Literal["rate_limited", "internal", "not_found", "unknown"]]:
+    error_kinds = (
+        set()
+    )  # type: set[Literal["rate_limited", "internal","not_found", "unknown"]]
+    if not errors:
+        return error_kinds
+    try:
+        for error in errors:
+            if "type" in error:
+                if error["type"] == "RATE_LIMITED":
+                    error_kinds.add("rate_limited")
+                elif error["type"] == "NOT_FOUND":
+                    error_kinds.add("not_found")
+                else:
+                    error_kinds.add("unknown")
+            elif "message" in error and error["message"].startswith(
+                "Something went wrong while executing your query."
+            ):
+                error_kinds.add("internal")
+            else:
+                error_kinds.add("unknown")
+    except TypeError:
+        pass
+    return error_kinds

bot/kodiak/evaluation.py

@@ -512,7 +512,7 @@ async def set_status(msg: str, markdown_content: Optional[str] = None) -> None:
         return
 
     if api_call_retries_remaining == 0:
-        log.warning("timeout reached for api calls to GitHub")
+        log.info("api_call_timeout")
         if api_call_errors:
             first_error = api_call_errors[0]
             await set_status(

bot/kodiak/queries/__init__.py

@@ -9,6 +9,7 @@
 import httpx as http
 import jwt
 import pydantic
+import sentry_sdk
 import structlog
 import toml
 from mypy_extensions import TypedDict
@@ -17,6 +18,7 @@
 
 import kodiak.app_config as conf
 from kodiak.config import V1, MergeMethod
+from kodiak.errors import identify_github_graphql_error
 from kodiak.queries.commits import Commit, CommitConnection, GitActor
 from kodiak.queries.commits import User as PullRequestCommitUser
 from kodiak.queries.commits import get_commits
@@ -34,7 +36,7 @@ class ErrorLocation(TypedDict):
     column: int
 
 
-class GraphQLError(TypedDict):
+class GraphQLError(TypedDict, total=False):
     message: str
     locations: List[ErrorLocation]
     type: Optional[str]
@@ -857,9 +859,11 @@ async def send_query(
         log = log.bind(rate_limit=rate_limit)
         try:
             res.raise_for_status()
-        except http.HTTPError:
-            log.warning("github api request error", res=res, exc_info=True)
-            return None
+        except http.HTTPStatusError as e:
+            with sentry_sdk.configure_scope() as scope:
+                scope.set_tag("http_error_code", e.response.status_code)
+                log.warning("github api request error", res=res, exc_info=True)
+                return None
         return cast(GraphQLResponse, res.json())
 
     async def get_api_features(self) -> ApiFeatures | None:
@@ -1014,8 +1018,16 @@ async def get_event_info(self, pr_number: int) -> Optional[EventInfoResponse]:
             return None
 
         data = res.get("data")
+        errors = res.get("errors")
+        if errors is not None:
+            error_kinds = identify_github_graphql_error(errors)
+            if "unknown" in error_kinds:
+                log.warning("unknown_error_found", errors=errors)
+            else:
+                log.info("api_error")
+
         if data is None:
-            log.error("could not fetch event info", res=res)
+            log.error("no data returned in api call", res=res)
             return None
 
         repository = get_repo(data=data)

bot/kodiak/test_errors.py

@@ -0,0 +1,56 @@
+from __future__ import annotations
+
+from typing import Any
+
+import pytest
+
+from kodiak import errors
+
+
+@pytest.mark.parametrize(
+    "response,expected",
+    [
+        (
+            [
+                {
+                    "message": "Something went wrong while executing your query. Please include `904B:6354:1B60F6:57C5B5:626C9136` when reporting this issue."
+                }
+            ],
+            {"internal"},
+        ),
+        (
+            [
+                {
+                    "type": "RATE_LIMITED",
+                    "message": "API rate limit exceeded for installation ID 000001.",
+                }
+            ],
+            {"rate_limited"},
+        ),
+        (
+            [{"message": "Some unexpected error"}],
+            {"unknown"},
+        ),
+        (
+            [
+                {
+                    "locations": [{"column": "3", "line": "147"}],
+                    "message": "Could not resolve to a Repository with the name 'chdsbd/.github'.",
+                    "path": ["orgConfigRepo"],
+                    "type": "NOT_FOUND",
+                }
+            ],
+            {"not_found"},
+        ),
+        (
+            {},
+            set(),
+        ),
+        (
+            None,
+            set(),
+        ),
+    ],
+)
+def test_identify_github_graphql_error(response: Any, expected: set[str]) -> None:
+    assert errors.identify_github_graphql_error(response) == expected