fix: early stop if CPU or disk space not set when using KVBM (#2997)

Signed-off-by: Ziqi Fan <ziqif@nvidia.com>

deploy/metrics/grafana_dashboards/grafana-kvbm-dashboard.json

@@ -333,7 +333,7 @@
           "useBackend": false
         }
       ],
-      "title": "Offload Blocks",
+      "title": "Offload Blocks - Device to Host",
       "type": "timeseries"
     },
     {
@@ -730,7 +730,7 @@
           "useBackend": false
         }
       ],
-      "title": "Onboard Blocks - Disk to Host",
+      "title": "Onboard Blocks - Disk to Device",
       "type": "timeseries"
     }
   ],
@@ -749,5 +749,5 @@
   "timezone": "browser",
   "title": "KVBM Dashboard",
   "uid": "3f679257-70a5-402c-92b4-05382337b548",
-  "version": 5
+  "version": 7
 }
\ No newline at end of file

lib/bindings/python/rust/llm/block_manager/distributed/leader.rs

@@ -86,6 +86,8 @@ impl KvbmLeader {
             .build()
             .map_err(to_pyerr)?;
 
+        config.sanity_check().map_err(to_pyerr)?;
+
         let rt = drt.inner().runtime().primary();
 
         let leader =

lib/llm/src/block_manager/distributed/leader.rs

@@ -77,6 +77,35 @@ impl KvbmLeaderConfig {
     pub fn builder() -> KvbmLeaderConfigBuilder {
         KvbmLeaderConfigBuilder::default()
     }
+
+    pub fn sanity_check(&self) -> anyhow::Result<()> {
+        let cpu = &self.host_blocks_config;
+        let disk = &self.disk_blocks_config;
+        if cpu.num_blocks_overriden == 0 && cpu.cache_size_in_gb == 0.0 {
+            if disk.num_blocks_overriden == 0 && disk.cache_size_in_gb == 0.0 {
+                panic!(
+                    "KVBM Configuration Error: No CPU memory configured.\n\
+                    \n\
+                    To fix this, set one of the following environment variables:\n\
+                    • DYN_KVBM_CPU_CACHE_GB=<size_in_gb>     (e.g., DYN_KVBM_CPU_CACHE_GB=4)\n\
+                    • DYN_KVBM_CPU_CACHE_OVERRIDE_NUM_BLOCKS=<num_blocks>  (e.g., DYN_KVBM_CPU_CACHE_OVERRIDE_NUM_BLOCKS=1000)\n\
+                    \n\
+                    Example: export DYN_KVBM_CPU_CACHE_GB=4"
+                );
+            } else {
+                panic!(
+                    "KVBM Configuration Error: CPU memory must be configured before disk memory.\n\
+                    \n\
+                    To fix this, set one of the following environment variables:\n\
+                    • DYN_KVBM_CPU_CACHE_GB=<size_in_gb>     (e.g., DYN_KVBM_CPU_CACHE_GB=4)\n\
+                    • DYN_KVBM_CPU_CACHE_OVERRIDE_NUM_BLOCKS=<num_blocks>  (e.g., DYN_KVBM_CPU_CACHE_OVERRIDE_NUM_BLOCKS=1000)\n\
+                    \n\
+                    Example: export DYN_KVBM_CPU_CACHE_GB=4"
+                );
+            }
+        }
+        Ok(())
+    }
 }
 
 #[derive(Debug, Default)]