Changes to new code

Signed-off-by: Shmuel Kallner <[email protected]>

Author: shmuelk

pkg/epp/scheduling/framework/plugins/multi/prefix/plugin.go

@@ -215,18 +215,18 @@ func (p *Plugin) Consumes() map[string]any {
 	return map[string]any{}
 }
 
-// PrepareRequestData hashes prompt, finds longest prefix match and stores it in pod as attribute.
-func (p *Plugin) PrepareRequestData(ctx context.Context, request *types.LLMRequest, pods []types.Pod) error {
-	hashes := hashPrompt(ctx, request, getBlockSize(pods, p.config), p.config.MaxPrefixBlocksToMatch)
+// PrepareRequestData hashes prompt, finds longest prefix match and stores it in endpoint as attribute.
+func (p *Plugin) PrepareRequestData(ctx context.Context, request *types.LLMRequest, endpoints []types.Endpoint) error {
+	hashes := hashPrompt(ctx, request, getBlockSize(endpoints, p.config), p.config.MaxPrefixBlocksToMatch)
 	state := &SchedulingContextState{
 		PrefixHashes:       hashes,
 		PrefixCacheServers: p.matchLongestPrefix(ctx, hashes),
 	}
 	total := len(state.PrefixHashes)
 
-	for _, pod := range pods {
-		matchLen := state.PrefixCacheServers[ServerID(pod.GetPod().NamespacedName)]
-		pod.Put(approximateprefix.PrefixCacheMatchInfoKey, approximateprefix.NewPrefixCacheMatchInfo(matchLen, total))
+	for _, endpoint := range endpoints {
+		matchLen := state.PrefixCacheServers[ServerID(endpoint.GetMetadata().NamespacedName)]
+		endpoint.Put(approximateprefix.PrefixCacheMatchInfoKey, approximateprefix.NewPrefixCacheMatchInfo(matchLen, total))
 	}
 	return nil
 }