feat: split and evict CP HiCache metadata

python/sglang/srt/mem_cache/hiradix_cache.py

@@ -1084,6 +1084,9 @@ class HiRadixCache(RadixCache):
         return parent
 
     def _evict_host_for_physical_slots(self, required_host_slots: int) -> int:
+        if required_host_slots <= 0:
+            return 0
+
         leaves = list(self.evictable_host_leaves)
         eviction_heap = [
             (self.eviction_strategy.get_priority(node), node) for node in leaves
@@ -1098,6 +1101,9 @@ class HiRadixCache(RadixCache):
             if not x.evicted:
                 continue
 
+            if not self._node_backuped(x):
+                continue
+
             if x.pin_expiry > 0 and time.monotonic() > x.pin_expiry:
                 self._clear_pin(x)
 
@@ -1124,6 +1130,9 @@ class HiRadixCache(RadixCache):
         return num_evicted
 
     def evict_host(self, num_tokens: int):
+        if self._uses_cp_hicache:
+            return self._evict_host_for_physical_slots(num_tokens)
+
         leaves = list(self.evictable_host_leaves)
         eviction_heap = [
             (self.eviction_strategy.get_priority(node), node) for node in leaves
@@ -1596,7 +1605,14 @@ class HiRadixCache(RadixCache):
         else:
             new_node.value = child.value[:split_len].clone()
             child.value = child.value[split_len:].clone()
-        if child.backuped:
+        if self._uses_cp_hicache:
+            if self._node_backuped(child):
+                new_node.cp_hicache, child.cp_hicache = child.cp_hicache.split(
+                    split_len
+                )
+                new_node.host_len = split_len
+                child.host_len = child.host_len - split_len
+        elif child.backuped:
             new_node.host_value = child.host_value[:split_len].clone()
             child.host_value = child.host_value[split_len:].clone()