Use X-SMG-Routing-Key header instead of json body and add tests (#15826)

This commit is contained in:
fzyzcjy
2025-12-25 09:58:14 -08:00
committed by GitHub
parent a355911909
commit 6ef543f99d
22 changed files with 106 additions and 107 deletions
@@ -105,7 +105,6 @@ async fn test_non_streaming_mcp_minimal_e2e_with_persistence() {
min_p: 0.0,
repetition_penalty: 1.0,
conversation: None,
routing_id: None,
};
let resp = router
@@ -329,7 +328,6 @@ fn test_responses_request_creation() {
min_p: 0.0,
repetition_penalty: 1.0,
conversation: None,
routing_id: None,
};
assert!(!request.is_stream());
@@ -374,7 +372,6 @@ fn test_responses_request_sglang_extensions() {
min_p: 0.05,
repetition_penalty: 1.1,
conversation: None,
routing_id: None,
};
// Verify SGLang extensions are present
@@ -490,7 +487,6 @@ fn test_json_serialization() {
min_p: 0.1,
repetition_penalty: 1.2,
conversation: None,
routing_id: None,
};
let json = serde_json::to_string(&request).expect("Serialization should work");
@@ -597,7 +593,6 @@ async fn test_multi_turn_loop_with_mcp() {
min_p: 0.0,
repetition_penalty: 1.0,
conversation: None,
routing_id: None,
};
// Execute the request (this should trigger the multi-turn loop)
@@ -747,7 +742,6 @@ async fn test_max_tool_calls_limit() {
min_p: 0.0,
repetition_penalty: 1.0,
conversation: None,
routing_id: None,
};
let response = router.route_responses(None, &req, None).await;
@@ -920,7 +914,6 @@ async fn test_streaming_with_mcp_tool_calls() {
min_p: 0.0,
repetition_penalty: 1.0,
conversation: None,
routing_id: None,
};
let response = router.route_responses(None, &req, None).await;
@@ -1201,7 +1194,6 @@ async fn test_streaming_multi_turn_with_mcp() {
min_p: 0.0,
repetition_penalty: 1.0,
conversation: None,
routing_id: None,
};
let response = router.route_responses(None, &req, None).await;
@@ -10,7 +10,6 @@ fn test_embedding_request_serialization_string_input() {
user: Some("user-1".to_string()),
dimensions: Some(128),
rid: Some("rid-123".to_string()),
routing_id: None,
log_metrics: None,
};
@@ -34,7 +33,6 @@ fn test_embedding_request_serialization_array_input() {
user: None,
dimensions: None,
rid: None,
routing_id: None,
log_metrics: None,
};
@@ -53,7 +51,6 @@ fn test_embedding_generation_request_trait_string() {
user: None,
dimensions: None,
rid: None,
routing_id: None,
log_metrics: None,
};
assert!(!req.is_stream());
@@ -70,7 +67,6 @@ fn test_embedding_generation_request_trait_array() {
user: None,
dimensions: None,
rid: None,
routing_id: None,
log_metrics: None,
};
assert_eq!(req.extract_text_for_routing(), "hello world");
@@ -85,7 +81,6 @@ fn test_embedding_generation_request_trait_non_text() {
user: None,
dimensions: None,
rid: None,
routing_id: None,
log_metrics: None,
};
assert_eq!(req.extract_text_for_routing(), "");
@@ -100,7 +95,6 @@ fn test_embedding_generation_request_trait_mixed_array_ignores_nested() {
user: None,
dimensions: None,
rid: None,
routing_id: None,
log_metrics: None,
};
// Only top-level string elements are extracted
-15
View File
@@ -17,7 +17,6 @@ fn test_rerank_request_serialization() {
return_documents: true,
rid: Some(StringOrArray::String("req-123".to_string())),
user: Some("user-456".to_string()),
routing_id: None,
};
let serialized = to_string(&request).unwrap();
@@ -60,7 +59,6 @@ fn test_rerank_request_validation_success() {
return_documents: true,
rid: None,
user: None,
routing_id: None,
};
assert!(request.validate().is_ok());
@@ -76,7 +74,6 @@ fn test_rerank_request_validation_empty_query() {
return_documents: true,
rid: None,
user: None,
routing_id: None,
};
let result = request.validate();
@@ -93,7 +90,6 @@ fn test_rerank_request_validation_whitespace_query() {
return_documents: true,
rid: None,
user: None,
routing_id: None,
};
let result = request.validate();
@@ -110,7 +106,6 @@ fn test_rerank_request_validation_empty_documents() {
return_documents: true,
rid: None,
user: None,
routing_id: None,
};
let result = request.validate();
@@ -127,7 +122,6 @@ fn test_rerank_request_validation_top_k_zero() {
return_documents: true,
rid: None,
user: None,
routing_id: None,
};
let result = request.validate();
@@ -144,7 +138,6 @@ fn test_rerank_request_validation_top_k_greater_than_docs() {
return_documents: true,
rid: None,
user: None,
routing_id: None,
};
// This should pass but log a warning
@@ -161,7 +154,6 @@ fn test_rerank_request_effective_top_k() {
return_documents: true,
rid: None,
user: None,
routing_id: None,
};
assert_eq!(request.effective_top_k(), 2);
@@ -177,7 +169,6 @@ fn test_rerank_request_effective_top_k_none() {
return_documents: true,
rid: None,
user: None,
routing_id: None,
};
assert_eq!(request.effective_top_k(), 3);
@@ -399,7 +390,6 @@ fn test_rerank_request_generation_request_trait() {
return_documents: true,
rid: None,
user: None,
routing_id: None,
};
assert_eq!(request.get_model(), Some("test-model"));
@@ -418,7 +408,6 @@ fn test_rerank_request_very_long_query() {
return_documents: true,
rid: None,
user: None,
routing_id: None,
};
assert!(request.validate().is_ok());
@@ -435,7 +424,6 @@ fn test_rerank_request_many_documents() {
return_documents: true,
rid: None,
user: None,
routing_id: None,
};
assert!(request.validate().is_ok());
@@ -455,7 +443,6 @@ fn test_rerank_request_special_characters() {
return_documents: true,
rid: Some(StringOrArray::String("req-🚀-123".to_string())),
user: Some("user-🎉-456".to_string()),
routing_id: None,
};
assert!(request.validate().is_ok());
@@ -474,7 +461,6 @@ fn test_rerank_request_rid_array() {
"req2".to_string(),
])),
user: None,
routing_id: None,
};
assert!(request.validate().is_ok());
@@ -529,7 +515,6 @@ fn test_full_rerank_workflow() {
return_documents: true,
rid: Some(StringOrArray::String("req-123".to_string())),
user: Some("user-456".to_string()),
routing_id: None,
};
// Validate request
@@ -89,7 +89,6 @@ fn create_minimal_completion_request() -> CompletionRequest {
return_hidden_states: false,
sampling_seed: None,
other: serde_json::Map::new(),
routing_id: None,
}
}
@@ -640,7 +639,6 @@ async fn test_unsupported_endpoints() {
return_bytes: false,
return_entropy: false,
rid: None,
routing_id: None,
};
let response = router.route_generate(None, &generate_request, None).await;