address review comments

tools/server/server-models.cpp

@@ -70,8 +70,8 @@ struct server_subproc {
     }
 };
 
-struct lru_sched {
-    lru_sched(server_models & models) : models(models) {}
+struct server_lru_sched {
+    server_lru_sched(server_models & models) : models(models) {}
 
     bool has_capacity(std::unique_lock<std::mutex> & lk) {
         check_lock(lk);
@@ -412,7 +412,7 @@ server_models::server_models(
               base_params(params),
               base_env(get_environment()),
               base_preset(ctx_preset.load_from_args(argc, argv)),
-              sched(std::make_unique<lru_sched>(*this)) {
+              sched(std::make_unique<server_lru_sched>(*this)) {
     // clean up base preset
     unset_reserved_args(base_preset, true);
     // set binary path
@@ -424,7 +424,7 @@ server_models::server_models(
         LOG_WRN("using original argv[0] as fallback: %s\n", argv[0]);
     }
     load_models();
-    debug_fake_timing = !common_get_env("LLAMA_DEBUG_FAKE_TIMING").empty();
+    debug_fake_timing = !common_get_env("LLAMA_SERVER_DEBUG_FAKE_TIMING").empty();
 }
 
 server_models::~server_models() = default;

tools/server/server-models.h

@@ -106,12 +106,12 @@ struct server_model_meta {
 };
 
 struct server_models_routes;
-struct server_subproc; // defined in server-models.cpp
-struct lru_sched;      // defined in server-models.cpp
+struct server_subproc;   // defined in server-models.cpp
+struct server_lru_sched; // defined in server-models.cpp
 
 struct server_models {
     friend struct server_models_routes;
-    friend struct lru_sched;
+    friend struct server_lru_sched;
 
 private:
     struct instance_t {
@@ -198,7 +198,7 @@ private:
     common_preset base_preset; // base preset from llama-server CLI args
 
     // queue of requests waiting for a models_max slot
-    std::unique_ptr<lru_sched> sched;
+    std::unique_ptr<server_lru_sched> sched;
 
     // if true, add some delay to simulate works (useful for testing)
     bool debug_fake_timing = false;

tools/server/tests/unit/test_router.py

@@ -145,7 +145,7 @@ def test_router_models_max_evicts_lru():
     assert _get_model_status(first) == "unloaded"
 
 
-# lru_sched tests (relying on LLAMA_DEBUG_FAKE_TIMING)
+# server_lru_sched tests (relying on LLAMA_SERVER_DEBUG_FAKE_TIMING)
 
 MODEL_A = "ggml-org/tinygemma3-GGUF:Q8_0"
 MODEL_B = "ggml-org/test-model-stories260K:F32"

tools/server/tests/utils.py

@@ -134,7 +134,7 @@ class ServerProcess:
     def start(self, timeout_seconds: int = DEFAULT_HTTP_TIMEOUT) -> None:
         env = {
             **os.environ,
-            "LLAMA_DEBUG_FAKE_TIMING": "1",
+            "LLAMA_SERVER_DEBUG_FAKE_TIMING": "1",
         }
         if "LLAMA_CACHE" not in os.environ:
             env["LLAMA_CACHE"] = "tmp"