Question 330
A scoring service works but its application logs do not show failures for malformed inputs. What observability improvement is relevant?
Explanation
Correct Answer
Serving logs identify request validation failures and model exceptions that aggregate uptime metrics may hide.
Incorrect Answers
Run representative end-to-end semantic inference tests against known cases and expected outputs. That operation would be relevant if: A deployed model's health probe succeeds but its predictions are nonsensical after a feature schema change. Its corresponding reason or verification is: Healthy infrastructure cannot guarantee correct feature interpretation or model behavior; application-level checks remain necessary. However the problem in this question is: A scoring service works but its application logs do not show failures for malformed inputs. The requirement instead calls for Capture structured request errors and scoring diagnostics in the supported endpoint logs; the question-specific check is: Serving logs identify request validation failures and model exceptions that aggregate uptime metrics may hide.
Validate the laptop's approved network route and DNS access to the private endpoint. A separate source case supports that approach when: An Azure ML online endpoint works from a VNet-hosted test runner but fails from a laptop after private access was enabled. The supporting rationale or observable result is: Private network enforcement can intentionally block clients outside the authorized path despite healthy model serving. This question describes a different issue: A scoring service works but its application logs do not show failures for malformed inputs. Address it through Capture structured request errors and scoring diagnostics in the supported endpoint logs, with the required justification or evidence: Serving logs identify request validation failures and model exceptions that aggregate uptime metrics may hide.
Use the exact active scoring URI issued for the newly created endpoint. This is useful for the distinct issue: A newly provisioned online endpoint returns 404 because the client calls an old endpoint hostname. The relevant technical rationale or evidence for that issue is: Endpoint names and addresses are specific to deployed resources; using a stale hostname cannot reach the intended service. Here the reported issue is: A scoring service works but its application logs do not show failures for malformed inputs. The relevant response is: Capture structured request errors and scoring diagnostics in the supported endpoint logs. The first result cannot verify this different correction: Serving logs identify request validation failures and model exceptions that aggregate uptime metrics may hide.