fix: config should use new precise-prefix-cache-score

- we have rename prefix-cache-score to precise-prefix-cache-score in 0.3.0, configs
  need migrate from the old one to the new one with spec.
  - rename plugin name
  - remove parameters.autoTune and parameters.mode: cache_tracking and
    lruCapacityPerServer
  - move hashBlockSize, maxPrefixBlocksToMatch under indexrConfig
- for config using food-review keep old prefix-cache-scorer

Signed-off-by: Wen Zhou <wenzhou@redhat.com>

Author: zdtsw

deploy/config/epp-config.yaml

@@ -3,7 +3,7 @@
 apiVersion: inference.networking.x-k8s.io/v1alpha1
 kind: EndpointPickerConfig
 plugins:
-- type: prefix-cache-scorer
+- type: precise-prefix-cache-scorer
 - type: decode-filter
 - type: max-score-picker
 - type: single-profile-handler
@@ -12,5 +12,5 @@ schedulingProfiles:
   plugins:
   - pluginRef: decode-filter
   - pluginRef: max-score-picker
-  - pluginRef: prefix-cache-scorer
+  - pluginRef: precise-prefix-cache-scorer
     weight: 2

deploy/config/pd-epp-config.yaml

@@ -3,7 +3,7 @@ apiVersion: inference.networking.x-k8s.io/v1alpha1
 kind: EndpointPickerConfig
 plugins:
 - type: prefill-header-handler
-- type: prefix-cache-scorer
+- type: precise-prefix-cache-scorer
 - type: prefill-filter
 - type: decode-filter
 - type: max-score-picker
@@ -13,11 +13,11 @@ schedulingProfiles:
   plugins:
   - pluginRef: prefill-filter
   - pluginRef: max-score-picker
-  - pluginRef: prefix-cache-scorer
+  - pluginRef: precise-prefix-cache-scorer
     weight: 2
 - name: decode
   plugins:
   - pluginRef: decode-filter
   - pluginRef: max-score-picker
-  - pluginRef: prefix-cache-scorer
+  - pluginRef: precise-prefix-cache-scorer
     weight: 2

deploy/config/sim-epp-kvcache-config.yaml

@@ -3,14 +3,13 @@
 apiVersion: inference.networking.x-k8s.io/v1alpha1
 kind: EndpointPickerConfig
 plugins:
-- type: prefix-cache-scorer
+- type: precise-prefix-cache-scorer
   parameters:
-    mode: cache_tracking
     kvEventsConfig:
       zmqEndpoint: tcp://0.0.0.0:5557
     indexerConfig:
       prefixStoreConfig:
-        blockSize: 16 
+        blockSize: 16
       tokenProcessorConfig:
         blockSize: 16                         # must match vLLM block size if not default (16)
         hashSeed: "42"                        # must match PYTHONHASHSEED in vLLM pods
@@ -28,5 +27,5 @@ schedulingProfiles:
   plugins:
   - pluginRef: decode-filter
   - pluginRef: max-score-picker
-  - pluginRef: prefix-cache-scorer
+  - pluginRef: precise-prefix-cache-scorer
     weight: 10

deploy/config/sim-epp-no-hit-lru.yaml

@@ -3,11 +3,13 @@
 apiVersion: inference.networking.x-k8s.io/v1alpha1
 kind: EndpointPickerConfig
 plugins:
-- type: prefix-cache-scorer
+- type: precise-prefix-cache-scorer
   parameters:
-    hashBlockSize: 5
-    maxPrefixBlocksToMatch: 256
-    lruCapacityPerServer: 31250
+    indexerConfig:
+      tokenProcessorConfig:
+        blockSize: 5
+      kvBlockIndexConfig:
+        maxPrefixBlocksToMatch: 256
 - type: no-hit-lru-scorer
   parameters:
     lruSize: 2048
@@ -19,7 +21,7 @@ schedulingProfiles:
   plugins:
   - pluginRef: decode-filter
   - pluginRef: max-score-picker
-  - pluginRef: prefix-cache-scorer
+  - pluginRef: precise-prefix-cache-scorer
     weight: 2
   - pluginRef: no-hit-lru-scorer
     weight: 1

docs/architecture.md

@@ -161,11 +161,13 @@ A complete configuration might look like this:
 apiVersion: inference.networking.x-k8s.io/v1alpha1
 kind: EndpointPickerConfig
 plugins:
-- type: prefix-cache-scorer
+- type: precise-prefix-cache-scorer
   parameters:
-    hashBlockSize: 5
-    maxPrefixBlocksToMatch: 256
-    lruCapacityPerServer: 31250
+    indexerConfig:
+      tokenProcessorConfig:
+        blockSize: 5
+      kvBlockIndexConfig:
+        maxPrefixBlocksToMatch: 256
 - type: decode-filter
 - type: max-score-picker
 - type: single-profile-handler
@@ -174,7 +176,7 @@ schedulingProfiles:
   plugins:
   - pluginRef: decode-filter
   - pluginRef: max-score-picker
-  - pluginRef: prefix-cache-scorer
+  - pluginRef: precise-prefix-cache-scorer
     weight: 50
 ```
 
@@ -434,11 +436,13 @@ Example configuration:
 
 ```yaml
 plugins:
-  - type: prefix-cache-scorer
+  - type: precise-prefix-cache-scorer
     parameters:
-      hashBlockSize: 5
-      maxPrefixBlocksToMatch: 256
-      lruCapacityPerServer: 31250
+      indexerConfig:
+        tokenProcessorConfig:
+          blockSize: 5
+        kvBlockIndexConfig:
+          maxPrefixBlocksToMatch: 256
   - type: no-hit-lru-scorer
     parameters:
       lruSize: 2048
@@ -450,7 +454,7 @@ schedulingProfiles:
     plugins:
       - pluginRef: decode-filter
       - pluginRef: max-score-picker
-      - pluginRef: prefix-cache-scorer
+      - pluginRef: precise-prefix-cache-scorer
         weight: 2
       - pluginRef: no-hit-lru-scorer
         weight: 1
@@ -471,11 +475,13 @@ apiVersion: inference.networking.x-k8s.io/v1alpha1
 kind: EndpointPickerConfig
 plugins:
 - type: prefill-header-handler
-- type: prefix-cache-scorer
+- type: precise-prefix-cache-scorer
   parameters:
-    hashBlockSize: 5
-    maxPrefixBlocksToMatch: 256
-    lruCapacityPerServer: 31250
+    indexerConfig:
+      tokenProcessorConfig:
+        blockSize: 5
+      kvBlockIndexConfig:
+        maxPrefixBlocksToMatch: 256
 - type: prefill-filter
 - type: decode-filter
 - type: max-score-picker
@@ -488,13 +494,13 @@ schedulingProfiles:
   plugins:
   - pluginRef: prefill-filter
   - pluginRef: max-score-picker
-  - pluginRef: prefix-cache-scorer
+  - pluginRef: precise-prefix-cache-scorer
     weight: 50
 - name: decode
   plugins:
   - pluginRef: decode-filter
   - pluginRef: max-score-picker
-  - pluginRef: prefix-cache-scorer
+  - pluginRef: precise-prefix-cache-scorer
     weight: 50
 ```
 

docs/disagg_pd.md

@@ -168,12 +168,13 @@ plugins:
     parameters:
       label: "role"
       validValues: ["decode"]
-  - type: prefix-cache-scorer
+  - type: precise-prefix-cache-scorer
     parameters:
-      autoTune: false
-      blockSize: 5
-      maxPrefixBlocksToMatch: 256
-      lruCapacityPerServer: 31250
+      indexerConfig:
+        tokenProcessorConfig:
+          blockSize: 5
+        kvBlockIndexConfig:
+          maxPrefixBlocksToMatch: 256
   - type: max-score-picker
   - type: prefill-header-handler
   - type: pd-profile-handler
@@ -186,12 +187,12 @@ schedulingProfiles:
     plugins:
       - pluginRef: "prefill-pods"
       - pluginRef: "max-score-picker"
-      - pluginRef: "prefix-cache-scorer"
+      - pluginRef: "precise-prefix-cache-scorer"
   - name: decode
     plugins:
       - pluginRef: "decode-pods"
       - pluginRef: "max-score-picker"
-      - pluginRef: "prefix-cache-scorer"
+      - pluginRef: "precise-prefix-cache-scorer"
 ```
 
 ---