fix(observability): preserve admin upstream error diagnostics

This commit is contained in:
elky
2026-09-07 10:39:48 +08:00
parent 470c59e197
commit 062e111c03
17 changed files with 690 additions and 116 deletions
+10 -5
View File
@@ -17,13 +17,13 @@ fn sanitize_request_candidate_rows(
mut candidates: Vec<StoredRequestCandidate>,
) -> Vec<StoredRequestCandidate> {
for candidate in &mut candidates {
candidate.sanitize_sensitive_diagnostics();
candidate.sanitize_for_persistence();
}
candidates
}
fn sanitize_request_candidate_row(mut candidate: StoredRequestCandidate) -> StoredRequestCandidate {
candidate.sanitize_sensitive_diagnostics();
candidate.sanitize_for_persistence();
candidate
}
@@ -1048,7 +1048,10 @@ mod request_candidate_security_tests {
fn assert_candidate_is_sanitized(candidate: &StoredRequestCandidate) {
assert_eq!(candidate.skip_reason.as_deref(), Some("unclassified_skip"));
assert_eq!(candidate.error_type.as_deref(), Some("unclassified_error"));
assert!(candidate.error_message.is_none());
assert_eq!(
candidate.error_message.as_deref(),
Some("Bearer candidate-secret")
);
assert_eq!(
candidate.extra_data,
Some(json!({"gateway_execution_runtime": true}))
@@ -1057,13 +1060,15 @@ mod request_candidate_security_tests {
candidate.required_capabilities,
Some(json!({"vision": true}))
);
assert!(!serde_json::to_string(candidate)
let mut public_candidate = candidate.clone();
public_candidate.sanitize_sensitive_diagnostics();
assert!(!serde_json::to_string(&public_candidate)
.expect("candidate should serialize")
.contains("candidate-secret"));
}
#[test]
fn gateway_candidate_boundary_sanitizes_repository_rows_and_write_results() {
fn gateway_candidate_boundary_preserves_admin_errors_and_removes_request_payloads() {
let candidate = sanitize_request_candidate_row(untrusted_candidate());
assert_candidate_is_sanitized(&candidate);
@@ -14682,9 +14682,10 @@ mod tests {
candidate_extra["upstream_response"]["status_code"],
json!(302)
);
assert!(candidate_extra["upstream_response"]
.get("headers")
.is_none());
assert_eq!(
candidate_extra["upstream_response"]["headers"]["location"],
"/"
);
assert!(candidate_extra["upstream_response"].get("body").is_none());
assert!(candidate_extra.get("client_response").is_none());
@@ -659,7 +659,7 @@ async fn admin_monitoring_trace_request_exposes_request_path_from_usage_audit()
}
#[tokio::test]
async fn admin_monitoring_trace_request_redacts_failed_candidate_response_payloads() {
async fn admin_monitoring_trace_request_exposes_failed_candidate_response_payloads() {
let mut candidate = sample_candidate(
"cand-used",
"request-1",
@@ -746,14 +746,96 @@ async fn admin_monitoring_trace_request_redacts_failed_candidate_response_payloa
json!("upstream_response")
);
assert!(extra["upstream_response"].get("headers").is_none());
assert!(extra["upstream_response"].get("body").is_none());
assert_eq!(
extra["upstream_response"]["body"]["error"]["message"],
"redirect blocked"
);
assert!(extra["upstream_response"].get("body_ref").is_none());
assert!(extra.get("client_response").is_none());
assert!(extra.get("provider_response").is_none());
}
#[tokio::test]
async fn admin_monitoring_trace_request_does_not_hydrate_ref_backed_usage_response_body() {
async fn admin_monitoring_trace_request_does_not_replace_attempt_status_with_usage_status() {
for (candidate_status, upstream_status, expected_status) in [
(Some(400), Some(400), Some(400)),
(Some(400), None, Some(400)),
(Some(502), Some(200), Some(200)),
(None, None, None),
] {
let mut candidate = sample_candidate(
"cand-used",
"request-failover-status",
0,
RequestCandidateStatus::Failed,
Some(101),
Some(33),
candidate_status,
);
if let Some(status_code) = upstream_status {
candidate.extra_data = Some(json!({
"upstream_response": {
"status_code": status_code,
"body": {"error": {"message": "sensitive upstream error"}}
}
}));
}
let request_candidates =
Arc::new(InMemoryRequestCandidateRepository::seed(vec![candidate]));
let mut usage = sample_usage(
"request-failover-status",
"provider-1",
"OpenAI",
0,
0.0,
"failed",
Some(503),
100,
);
usage.candidate_id = Some("cand-used".to_string());
usage.response_body_state = Some(UsageBodyCaptureState::Reference);
let usage_repository = Arc::new(InMemoryUsageReadRepository::seed(vec![usage]));
let data_state =
crate::data::GatewayDataState::with_request_candidate_and_usage_repository_for_tests(
request_candidates,
usage_repository,
);
let state = AppState::new()
.expect("state should build")
.with_data_state_for_tests(data_state);
let context = request_context(
http::Method::GET,
"/api/admin/monitoring/trace/request-failover-status",
);
let response = local_monitoring_response(&state, &context)
.await
.expect("handler should not error")
.expect("route should be handled locally");
assert_eq!(response.status(), http::StatusCode::OK);
let body = to_bytes(response.into_body(), usize::MAX)
.await
.expect("body should read");
let payload: serde_json::Value =
serde_json::from_slice(&body).expect("json body should parse");
let candidate = &payload["candidates"][0];
assert_eq!(candidate["status_code"], json!(candidate_status));
assert_eq!(
candidate["extra_data"]["upstream_response"]["status_code"],
json!(expected_status),
);
if upstream_status.is_some() {
assert_eq!(
candidate["extra_data"]["upstream_response"]["body"]["error"]["message"],
"sensitive upstream error"
);
}
assert!(candidate["error_message"].is_null());
}
}
#[tokio::test]
async fn admin_monitoring_trace_request_keeps_candidate_errors_without_hydrating_usage_bodies() {
let mut candidate = sample_candidate(
"cand-used",
"request-ref-body",
@@ -771,8 +853,7 @@ async fn admin_monitoring_trace_request_does_not_hydrate_ref_backed_usage_respon
"x-request-id": "stale-request-like-body"
},
"body": {
"model": "gpt-5.6-sol",
"input": [{"role": "user", "content": "request prompt"}]
"error": {"message": "candidate-specific upstream failure"}
}
}
}));
@@ -837,8 +918,14 @@ async fn admin_monitoring_trace_request_does_not_hydrate_ref_backed_usage_respon
assert_eq!(upstream_response["status_code"], json!(400));
assert_eq!(upstream_response["source"], json!("upstream_response"));
assert_eq!(upstream_response["body_state"], json!("reference"));
assert!(upstream_response.get("headers").is_none());
assert!(upstream_response.get("body").is_none());
assert_eq!(
upstream_response["headers"]["content-type"],
"text/event-stream"
);
assert_eq!(
upstream_response["body"]["error"]["message"],
"candidate-specific upstream failure"
);
assert!(upstream_response.get("body_ref").is_none());
}
@@ -1940,12 +1940,16 @@ mod tests {
#[test]
fn user_usage_active_override_uses_terminal_candidate_latency() {
let candidate = sample_candidate(
let mut candidate = sample_candidate(
RequestCandidateStatus::Success,
Some(200),
Some(9_210),
None,
);
candidate.error_message = Some("private upstream diagnostic".to_string());
candidate.extra_data = Some(json!({
"upstream_response": {"body": {"error": {"message": "private upstream diagnostic"}}}
}));
let payload =
users_me_usage_terminal_candidate_state_override(&[candidate]).expect("override");
@@ -1953,6 +1957,7 @@ mod tests {
assert_eq!(payload["status"], "completed");
assert_eq!(payload["response_time_ms"], 9_210);
assert_eq!(payload["status_code"], 200);
assert!(!payload.to_string().contains("private upstream diagnostic"));
assert_eq!(
payload["response_time_updated_at"],
"1970-01-01T00:00:10.210+00:00"
@@ -2460,15 +2460,21 @@ async fn gateway_retries_next_local_openai_chat_stream_candidate_after_retryable
failed_candidate.error_type.as_deref(),
Some("retryable_upstream_status")
);
assert!(failed_candidate.error_message.is_none());
assert!(failed_candidate.error_message.is_some());
let failed_upstream_response = failed_candidate
.extra_data
.as_ref()
.and_then(|value| value.get("upstream_response"))
.expect("failed stream candidate should keep its upstream response");
assert_eq!(failed_upstream_response["status_code"], json!(429));
assert!(failed_upstream_response.get("headers").is_none());
assert!(failed_upstream_response.get("body").is_none());
assert_eq!(
failed_upstream_response["headers"]["content-type"],
"application/json"
);
assert_eq!(
failed_upstream_response["body"]["error"]["message"],
"rate limited"
);
assert_eq!(success_candidate.status, RequestCandidateStatus::Success);
assert_eq!(success_candidate.status_code, Some(200));
assert!(success_candidate.started_at_unix_ms.is_some());
@@ -1166,15 +1166,21 @@ async fn gateway_retries_next_local_openai_chat_sync_candidate_after_auth_failur
assert_eq!(failed_candidate.retry_index, retry_index as u32);
assert_eq!(failed_candidate.status, RequestCandidateStatus::Failed);
assert_eq!(failed_candidate.status_code, Some(401));
assert!(failed_candidate.error_message.is_none());
assert!(failed_candidate.error_message.is_some());
let failed_upstream_response = failed_candidate
.extra_data
.as_ref()
.and_then(|value| value.get("upstream_response"))
.expect("failed candidate should keep its upstream response");
assert_eq!(failed_upstream_response["status_code"], json!(401));
assert!(failed_upstream_response.get("headers").is_none());
assert!(failed_upstream_response.get("body").is_none());
assert_eq!(
failed_upstream_response["headers"]["content-type"],
"application/json"
);
assert_eq!(
failed_upstream_response["body"]["error"]["message"],
"invalid auth token"
);
}
assert_eq!(stored_candidates[2].candidate_index, 1);
assert_eq!(stored_candidates[2].status, RequestCandidateStatus::Success);
+27 -9
View File
@@ -534,6 +534,24 @@ async fn gateway_exposes_request_audit_bundle_via_internal_audit_endpoint() {
#[tokio::test]
async fn gateway_exposes_request_candidate_trace_via_internal_audit_endpoint() {
let mut failed_candidate = sample_request_candidate(
"cand-2",
"req-trace-1",
1,
RequestCandidateStatus::Failed,
Some(101),
Some(37),
Some(502),
);
failed_candidate.error_message = Some("private upstream diagnostic".to_string());
failed_candidate.extra_data = Some(json!({
"upstream_response": {
"status_code": 502,
"headers": {"x-request-id": "private-upstream-id"},
"body": {"error": {"message": "private upstream diagnostic"}}
},
"error_flow": {"status_code": 502, "message": "private upstream diagnostic"}
}));
let repository = Arc::new(InMemoryRequestCandidateRepository::seed(vec![
sample_request_candidate(
"cand-1",
@@ -544,15 +562,7 @@ async fn gateway_exposes_request_candidate_trace_via_internal_audit_endpoint() {
None,
None,
),
sample_request_candidate(
"cand-2",
"req-trace-1",
1,
RequestCandidateStatus::Failed,
Some(101),
Some(37),
Some(502),
),
failed_candidate,
]));
let gateway_state = AppState::new()
@@ -582,6 +592,14 @@ async fn gateway_exposes_request_candidate_trace_via_internal_audit_endpoint() {
);
assert_eq!(payload["candidates"][0]["id"], "cand-2");
assert_eq!(payload["candidates"][0]["status"], "failed");
assert!(payload["candidates"][0]["error_message"].is_null());
assert_eq!(
payload["candidates"][0]["extra_data"]["upstream_response"]["status_code"],
502
);
let serialized = payload.to_string();
assert!(!serialized.contains("private upstream diagnostic"));
assert!(!serialized.contains("private-upstream-id"));
gateway_handle.abort();
}
@@ -473,6 +473,23 @@ async fn gateway_handles_admin_monitoring_trace_request_locally_with_trusted_adm
}),
);
let mut failed_candidate = sample_candidate(
"cand-used",
"request-1",
1,
RequestCandidateStatus::Failed,
Some(101),
Some(33),
Some(502),
);
failed_candidate.error_message = Some("private upstream diagnostic".to_string());
failed_candidate.extra_data = Some(json!({
"upstream_response": {
"status_code": 502,
"headers": {"x-request-id": "upstream-diagnostic-id"},
"body": {"error": {"message": "private upstream diagnostic"}}
}
}));
let request_candidates = Arc::new(InMemoryRequestCandidateRepository::seed(vec![
sample_candidate(
"cand-unused",
@@ -483,15 +500,7 @@ async fn gateway_handles_admin_monitoring_trace_request_locally_with_trusted_adm
None,
None,
),
sample_candidate(
"cand-used",
"request-1",
1,
RequestCandidateStatus::Failed,
Some(101),
Some(33),
Some(502),
),
failed_candidate,
]));
let provider_catalog = Arc::new(InMemoryProviderCatalogReadRepository::seed(
vec![sample_provider()],
@@ -538,6 +547,37 @@ async fn gateway_handles_admin_monitoring_trace_request_locally_with_trusted_adm
assert_eq!(payload["candidates"][0]["key_auth_type"], json!("api_key"));
assert_eq!(payload["candidates"][0]["latency_ms"], json!(33));
assert_eq!(payload["candidates"][0]["status_code"], json!(502));
assert_eq!(
payload["candidates"][0]["error_message"],
"private upstream diagnostic"
);
assert_eq!(
payload["candidates"][0]["extra_data"]["upstream_response"]["body"]["error"]["message"],
"private upstream diagnostic"
);
for role in [None, Some("user")] {
let mut request = reqwest::Client::new().get(format!(
"{gateway_url}/api/admin/monitoring/trace/request-1"
));
if let Some(role) = role {
request = request
.header(crate::constants::GATEWAY_HEADER, "rust-phase3b")
.header(TRUSTED_ADMIN_USER_ID_HEADER, "regular-user")
.header(TRUSTED_ADMIN_USER_ROLE_HEADER, role)
.header(TRUSTED_ADMIN_SESSION_ID_HEADER, "regular-session");
}
let response = request
.send()
.await
.expect("unauthorized probe should complete");
assert!(matches!(
response.status(),
StatusCode::UNAUTHORIZED | StatusCode::FORBIDDEN
));
let body = response.text().await.expect("denial body should read");
assert!(!body.contains("private upstream diagnostic"));
assert!(!body.contains("upstream-diagnostic-id"));
}
assert_eq!(*upstream_hits.lock().expect("mutex should lock"), 0);
gateway_handle.abort();
@@ -85,10 +85,7 @@ async fn admin_provider_summary_health_preserves_redacted_key_summaries() {
.into_iter()
.map(|(key_id, auth_type, auth_config, score)| {
let mut key = sample_key(key_id, "provider-openai", "openai:chat", "test")
.with_health_fields(
Some(json!({"openai:chat": {"health_score": score}})),
None,
);
.with_health_fields(Some(json!({"openai:chat": {"health_score": score}})), None);
key.auth_type = auth_type.to_string();
key.encrypted_api_key = Some("summary".to_string());
key.encrypted_auth_config = auth_config.map(ToOwned::to_owned);
@@ -115,7 +112,8 @@ async fn admin_provider_summary_health_preserves_redacted_key_summaries() {
let body = axum::body::to_bytes(response.into_body(), 1024 * 1024)
.await
.expect("summary body should read");
let payload: serde_json::Value = serde_json::from_slice(&body).expect("summary should parse");
let payload: serde_json::Value =
serde_json::from_slice(&body).expect("summary should parse");
let summary = if uri == "/api/admin/providers/summary" {
&payload["items"][0]
} else {
@@ -1308,7 +1308,13 @@ mod tests {
stored[0].error_type.as_deref(),
Some("stream_missing_terminal_event")
);
assert!(stored[0].error_message.is_none());
assert_eq!(
stored[0].error_message.as_deref(),
Some(super::STREAM_MISSING_TERMINAL_EVENT_MESSAGE)
);
let mut public_candidate = stored[0].clone();
public_candidate.sanitize_sensitive_diagnostics();
assert!(public_candidate.error_message.is_none());
}
#[tokio::test]