feat(routing): consolidate scheduling strategy configuration

crates/aether-routing-core/src/policy.rs

@@ -8,7 +8,9 @@ use crate::actions::{
     RoutingAction, RoutingRulePhase, RoutingSchedulingMode, RoutingSetPriorityMode,
 };
 use crate::conditions::RoutingConditionContext;
-use crate::model::{RoutingGroupConfig, RoutingModelPolicy, RoutingPoolPolicyOverride};
+use crate::model::{
+    RoutingExecutionPolicy, RoutingGroupConfig, RoutingModelPolicy, RoutingPoolPolicyOverride,
+};
 use crate::mutations::{validate_header_patch, validate_json_patch_operations, MutationPlan};
 use crate::ranking::RankingOverlay;
 use crate::validation::validate_routing_group_config;
@@ -17,7 +19,7 @@ use crate::validation::validate_routing_group_config;
 pub enum RoutingPolicyError {
     #[error("routing group config is invalid: {0}")]
     InvalidConfig(String),
-    #[error("model is not allowed by routing group: {0}")]
+    #[error("model is not allowed by routing rule: {0}")]
     ModelNotAllowed(String),
     #[error("mutation action is invalid: {0}")]
     InvalidMutation(String),
@@ -60,6 +62,8 @@ pub struct ResolvedRoutingPolicy {
     /// See `RoutingDefaultPolicy::sticky_key_attempts`.
     #[serde(default = "default_sticky_key_attempts")]
     pub sticky_key_attempts: u32,
+    #[serde(flatten)]
+    pub execution_policy: RoutingExecutionPolicy,
     pub ranking_overlay: RankingOverlay,
     pub mutation_plan: MutationPlan,
     #[serde(default)]
@@ -75,14 +79,6 @@ pub fn resolve_routing_policy(
     validate_routing_group_config(config)
         .map_err(|error| RoutingPolicyError::InvalidConfig(error.to_string()))?;
 
-    if !model_allowed(&config.allowed_models, input.requested_model)
-        && !model_allowed(&config.allowed_models, input.resolved_model)
-    {
-        return Err(RoutingPolicyError::ModelNotAllowed(
-            input.requested_model.to_string(),
-        ));
-    }
-
     let mut policy = ResolvedRoutingPolicy {
         group_id: input.group_id.map(str::to_string),
         group_version: input.group_version,
@@ -93,6 +89,7 @@ pub fn resolve_routing_policy(
         scheduling_mode: config.default_policy.scheduling_mode,
         keep_priority_on_conversion: config.default_policy.keep_priority_on_conversion,
         sticky_key_attempts: config.default_policy.sticky_key_attempts,
+        execution_policy: config.default_policy.execution_policy,
         ranking_overlay: RankingOverlay::default(),
         mutation_plan: MutationPlan::default(),
         pool_policy_overrides: BTreeMap::new(),
@@ -322,7 +319,6 @@ mod tests {
     #[test]
     fn resolves_model_policy_and_matching_rule() {
         let config = RoutingGroupConfig {
-            allowed_models: vec!["gpt-*".to_string()],
             default_policy: RoutingDefaultPolicy::default(),
             model_policies: vec![RoutingModelPolicy {
                 model: "gpt-5".to_string(),
@@ -389,14 +385,14 @@ mod tests {
     }
 
     #[test]
-    fn empty_allowlist_keeps_default_policy_for_models_without_an_override() {
+    fn default_policy_applies_to_models_without_an_override() {
         let config = RoutingGroupConfig {
-            allowed_models: vec![],
             default_policy: RoutingDefaultPolicy {
                 priority_mode: RoutingSetPriorityMode::GlobalKey,
                 scheduling_mode: RoutingSchedulingMode::LoadBalance,
                 keep_priority_on_conversion: true,
                 sticky_key_attempts: 3,
+                execution_policy: Default::default(),
             },
             model_policies: vec![RoutingModelPolicy {
                 model: "special-model".to_string(),
@@ -471,6 +467,38 @@ mod tests {
             .is_empty());
     }
 
+    #[test]
+    fn legacy_group_model_allowlist_is_ignored() {
+        let config: RoutingGroupConfig = serde_json::from_value(json!({
+            "allowed_models": ["gpt-5"],
+            "default_policy": {
+                "priority_mode": "provider",
+                "scheduling_mode": "cache_affinity"
+            },
+            "model_policies": [],
+            "rules": []
+        }))
+        .expect("legacy routing config should remain readable");
+
+        resolve_routing_policy(
+            &config,
+            RoutingPolicyInput {
+                group_id: Some("group-1"),
+                group_version: Some(1),
+                selection_source: "system_default",
+                requested_model: "claude-sonnet",
+                resolved_model: "claude-sonnet",
+                api_format: "openai:chat",
+                user_id: None,
+                api_key_id: None,
+                headers: &json!({}),
+                body: &json!({}),
+                phase: RoutingRulePhase::ClientRequest,
+            },
+        )
+        .expect("the legacy allowlist must not reject another model");
+    }
+
     #[test]
     fn sticky_key_attempts_defaults_to_two_and_can_be_overridden_by_rule() {
         let default_config = RoutingGroupConfig::default();
@@ -541,41 +569,9 @@ mod tests {
         assert_eq!(policy.sticky_key_attempts, 1);
     }
 
-    #[test]
-    fn rejects_disallowed_model() {
-        let config = RoutingGroupConfig {
-            allowed_models: vec!["gpt-5".to_string()],
-            ..RoutingGroupConfig::default()
-        };
-
-        let err = resolve_routing_policy(
-            &config,
-            RoutingPolicyInput {
-                group_id: None,
-                group_version: None,
-                selection_source: "test",
-                requested_model: "claude",
-                resolved_model: "claude",
-                api_format: "openai:chat",
-                user_id: None,
-                api_key_id: None,
-                headers: &json!({}),
-                body: &json!({}),
-                phase: RoutingRulePhase::ClientRequest,
-            },
-        )
-        .unwrap_err();
-
-        assert_eq!(
-            err,
-            RoutingPolicyError::ModelNotAllowed("claude".to_string())
-        );
-    }
-
     #[test]
     fn restrict_model_action_rejects_matching_request() {
         let config = RoutingGroupConfig {
-            allowed_models: vec!["*".to_string()],
             rules: vec![RoutingRule {
                 id: "restrict".to_string(),
                 priority: 1,