| 1 | use super::*; |
| 2 | |
| 3 | #[test] |
| 4 | fn retryability_distinguishes_transient_failures_from_durable_failures() { |
| 5 | for error in [ |
| 6 | LlmError::RateLimited { |
| 7 | message: "too many requests".into(), |
| 8 | retry_after: None, |
| 9 | }, |
| 10 | LlmError::ServerError { |
| 11 | status: 500, |
| 12 | message: "internal error".into(), |
| 13 | }, |
| 14 | LlmError::NetworkError("connection refused".into()), |
| 15 | LlmError::Timeout(Duration::from_secs(30)), |
| 16 | ] { |
| 17 | assert!(error.is_retryable(), "expected transient error: {error}"); |
| 18 | } |
| 19 | for error in [ |
| 20 | LlmError::authentication_error("invalid key"), |
| 21 | LlmError::AuthorizationError("blocked".into()), |
| 22 | LlmError::InvalidRequest { |
| 23 | status: 400, |
| 24 | message: "bad json".into(), |
| 25 | }, |
| 26 | LlmError::ContentPolicyError("unsafe content".into()), |
| 27 | LlmError::ContextLengthError("too long".into()), |
| 28 | ] { |
| 29 | assert!(!error.is_retryable(), "expected durable error: {error}"); |
| 30 | } |
| 31 | } |
| 32 | |
| 33 | #[test] |
| 34 | fn http_response_boundary_classifies_status_contract() { |
| 35 | assert!(matches!( |
| 36 | LlmError::from_http_response(429, "rate limit exceeded"), |
| 37 | LlmError::RateLimited { .. } |
| 38 | )); |
| 39 | assert!(matches!( |
| 40 | LlmError::from_http_response(401, "invalid api key"), |
| 41 | LlmError::AuthenticationError(_) |
| 42 | )); |
| 43 | assert!(matches!( |
| 44 | LlmError::from_http_response(403, "forbidden"), |
| 45 | LlmError::AuthorizationError(_) |
| 46 | )); |
| 47 | assert!(matches!( |
| 48 | LlmError::from_http_response(403, "invalid api key"), |
| 49 | LlmError::AuthenticationError(_) |
| 50 | )); |
| 51 | let cancelled = LlmError::from_http_response(499, "upstream request cancelled"); |
| 52 | assert!(matches!( |
| 53 | &cancelled, |
| 54 | LlmError::ServerError { status: 499, .. } |
| 55 | )); |
| 56 | assert!(cancelled.is_retryable()); |
| 57 | assert!(matches!( |
| 58 | LlmError::from_http_response(500, "internal server error"), |
| 59 | LlmError::ServerError { status: 500, .. } |
| 60 | )); |
| 61 | assert!(matches!( |
| 62 | LlmError::from_http_response(503, "service unavailable"), |
| 63 | LlmError::ServerError { status: 503, .. } |
| 64 | )); |
| 65 | assert!(matches!( |
| 66 | LlmError::from_http_response(400, "context_length_exceeded"), |
| 67 | LlmError::ContextLengthError(_) |
| 68 | )); |
| 69 | assert!(matches!( |
| 70 | LlmError::from_http_response(400, "content_policy_violation"), |
| 71 | LlmError::ContentPolicyError(_) |
| 72 | )); |
| 73 | assert!(matches!( |
| 74 | LlmError::from_http_response(400, "invalid json"), |
| 75 | LlmError::InvalidRequest { status: 400, .. } |
| 76 | )); |
| 77 | } |
| 78 | |
| 79 | #[test] |
| 80 | fn explicit_400_402_and_429_quota_responses_are_typed_and_non_retryable() { |
| 81 | for (status, body) in [ |
| 82 | ( |
| 83 | 400, |
| 84 | r#"{"error":{"code":"insufficient_quota","message":"You exceeded your current quota"}}"#, |
| 85 | ), |
| 86 | ( |
| 87 | 429, |
| 88 | r#"{"error":{"type":"insufficient_quota","message":"Billing limit reached"}}"#, |
| 89 | ), |
| 90 | ( |
| 91 | 402, |
| 92 | r#"{"error":{"code":"billing_hard_limit_reached","message":"Payment required"}}"#, |
| 93 | ), |
| 94 | ( |
| 95 | 429, |
| 96 | "You exceeded your current quota. Please check your plan and billing details.", |
| 97 | ), |
| 98 | (429, "Account quota exhausted"), |
| 99 | ] { |
| 100 | let error = LlmError::from_http_response(status, body); |
| 101 | assert!(matches!(error, LlmError::QuotaExhausted(_))); |
| 102 | assert!(!error.is_retryable()); |
| 103 | } |
| 104 | |
| 105 | let raw = r#"{"error":{"code":"billing_hard_limit_reached","message":"Account unavailable"}}"#; |
| 106 | let safe = sanitize_http_error_body(Some("fixture"), 429, raw); |
| 107 | assert!(matches!( |
| 108 | LlmError::from_http_response(429, &safe), |
| 109 | LlmError::QuotaExhausted(_) |
| 110 | )); |
| 111 | } |
| 112 | |
| 113 | #[test] |
| 114 | fn generic_429_stays_rate_limited_and_retryable() { |
| 115 | for body in [ |
| 116 | "Too Many Requests", |
| 117 | "Rate limit on your API quota exceeded", |
| 118 | "Requests per minute quota exceeded", |
| 119 | "Quota rate limit exceeded; retry after 10 seconds", |
| 120 | ] { |
| 121 | let error = LlmError::from_http_response(429, body); |
| 122 | assert!( |
| 123 | matches!(error, LlmError::RateLimited { .. }), |
| 124 | "expected transient rate limit for {body:?}, got {error:?}" |
| 125 | ); |
| 126 | assert!(error.is_retryable()); |
| 127 | } |
| 128 | |
| 129 | let raw = r#"{"error":{"code":"RESOURCE_EXHAUSTED","message":"Rate limit on your API quota exceeded"}}"#; |
| 130 | let safe = sanitize_http_error_body(Some("fixture"), 429, raw); |
| 131 | let error = LlmError::from_http_response(429, &safe); |
| 132 | assert!(matches!(error, LlmError::RateLimited { .. })); |
| 133 | assert!(error.is_retryable()); |
| 134 | } |
| 135 | |
| 136 | #[tokio::test] |
| 137 | async fn retry_loop_stops_after_one_typed_quota_failure() { |
| 138 | let mut calls = 0; |
| 139 | let result: RetryResult<i32> = with_retry( |
| 140 | &RetryConfig::default(), |
| 141 | || { |
| 142 | calls += 1; |
| 143 | async { |
| 144 | Err(LlmError::from_http_response( |
| 145 | 429, |
| 146 | r#"{"error":{"code":"insufficient_quota"}}"#, |
| 147 | )) |
| 148 | } |
| 149 | }, |
| 150 | None, |
| 151 | ) |
| 152 | .await; |
| 153 | assert_eq!(result.unwrap_err().attempts, 1); |
| 154 | assert_eq!(calls, 1); |
| 155 | } |
| 156 | |
| 157 | #[tokio::test] |
| 158 | async fn retry_loop_stops_after_one_authentication_failure() { |
| 159 | let mut calls = 0; |
| 160 | let result: RetryResult<i32> = with_retry( |
| 161 | &RetryConfig::default(), |
| 162 | || { |
| 163 | calls += 1; |
| 164 | async { Err(LlmError::authentication_error("bad key")) } |
| 165 | }, |
| 166 | None, |
| 167 | ) |
| 168 | .await; |
| 169 | assert!(result.is_err()); |
| 170 | assert_eq!(calls, 1); |
| 171 | } |
| 172 |