Consume hash-proven recurrent replay boundaries

deploy/deepseek_v4/tp4_profile.json

@@ -4,7 +4,7 @@
   "cache_model_profile": "deepseek-v4-fp8-hma",
   "published_runtime_base": "ghcr.io/fujitsupolycom/gb10-vllm-serving@sha256:6fc26fdad81a18f0fff67ce0a05f6d90165625ea2e1cac8a6f39bfb462017028",
   "sparkcache": {
-    "source_sha256": "bc7cae86732c869ee8b2205d48ac5be6f580ee8b77a3e4ffd4c69dcd4f1bfae5"
+    "source_sha256": "01cc59bf2c45af60f02813b771a484e65829e4f464eb27aa1756ca7067c82c9f"
   },
   "model": {
     "repository": "deepseek-ai/DeepSeek-V4-Flash-0731",

deploy/glm52_35bpw/profile.json

@@ -5,7 +5,7 @@
   "published_runtime_base": "ghcr.io/fujitsupolycom/gb10-vllm-serving@sha256:6fc26fdad81a18f0fff67ce0a05f6d90165625ea2e1cac8a6f39bfb462017028",
   "base_image_requirement": "exact GLM-5.2 3.5-bpw R7 image recorded by the source container inspection",
   "sparkcache": {
-    "source_sha256": "bc7cae86732c869ee8b2205d48ac5be6f580ee8b77a3e4ffd4c69dcd4f1bfae5"
+    "source_sha256": "01cc59bf2c45af60f02813b771a484e65829e4f464eb27aa1756ca7067c82c9f"
   },
   "model": {
     "repository": "brandonmusic/GLM-5.2-EXL3-TR3v4-3.5bpw-MTP78",

patches/vllm-glm53-b12x-kda-adaptive-mtp/source-receipt.json

@@ -62,7 +62,7 @@
   ],
   "contract": {
     "path": "sparkcache/runtime_patches/vllm-kv-block-lease-contract-glm53-b12x-kda-adaptive-mtp.json",
-    "sha256": "6defde9551cbb586fd09bb2d3020495531b6573397875a767eaae1dbad126024"
+    "sha256": "45d7a92b38b836a4f829f02df85e339cfeea860e1080e4663a8340af6c125125"
   },
   "result": "All four patches apply in order to the LF source tree, and the eleven-file SparkCache contract verifies the resulting source bytes, including the live-tensor B12X KDA implementation."
 }

sparkcache/README.md

@@ -225,7 +225,13 @@ when placement completes and intentionally excludes that bookkeeping.
   `sparkcache-hybrid-page-delta/v1` codec reuses only byte-identical page
   prefixes and binds the base snapshot, layout, block counts, and semantic
   token boundaries. A boundary inside an HMA page replaces that complete page
-  while retaining earlier byte-identical pages. The
+  while retaining earlier byte-identical pages. For an aligned recurrent group,
+  vLLM may retain the replay-boundary page outside the advancing request block
+  table. Its `SchedulerOutput.recurrent_boundary_blocks` hand-off names the
+  pinned physical block by request, group, and token boundary. SparkCache uses
+  that block only after all three identities and the recurrent topology match;
+  missing or contradictory metadata skips publication rather than scanning
+  later running or speculative state. The
   `sparkcache-page-delta-manifest/v1` schema embeds its authenticated base
   graph, allowing capacity maintenance to retain shared objects after
   predecessor roots are removed. Restore reconstructs the verified full

sparkcache/runtime_patches/test_glm53_b12x_kda_adaptive_mtp_contract.py

@@ -19,6 +19,13 @@
 CONTAINERFILE = ROOT / "deploy/glm53_flash/Containerfile.b12x-kda-adaptive-mtp"
 VLLM_COMMIT = "0b67266a0f37d6146a8403fb8482403c62f412d5"
 SOURCE_ROLE = "source_built_glm53_b12x_kda_adaptive_mtp"
+RECURRENT_BOUNDARY_ROLE = "recurrent_boundary_contract"
+RECURRENT_BOUNDARY_FILES = {
+    "vllm/v1/core/kv_cache_manager.py",
+    "vllm/v1/core/sched/output.py",
+    "vllm/v1/core/sched/scheduler.py",
+    "vllm/v1/core/single_type_kv_cache_manager.py",
+}
 KDA_PATH = "vllm/model_executor/layers/mamba/gdn/kimi_gdn_linear_attn.py"
 E105_KDA_SHA256 = (
     "a879af0081f69ba8288ef909e1d69b5bbb85bdff7e5aa0d3c11ad892bfea8410"
@@ -71,12 +78,24 @@ def test_glm53_b12x_kda_adaptive_mtp_contract_attests_the_complete_sparkcache_vl
         "vllm/v1/kv_cache_interface.py",
         KDA_PATH,
     }
-    assert all(
-        set(record["accepted_sha256"]) == {SOURCE_ROLE}
-        for record in contract["files"]
-    )
+    for record in contract["files"]:
+        expected_roles = {SOURCE_ROLE}
+        if record["path"] in RECURRENT_BOUNDARY_FILES:
+            expected_roles.add(RECURRENT_BOUNDARY_ROLE)
+        assert set(record["accepted_sha256"]) == expected_roles
     assert all(record["required_symbols"] for record in contract["files"])
 
+    by_path = {record["path"]: record for record in contract["files"]}
+    assert by_path["vllm/v1/core/sched/output.py"]["accepted_sha256"][
+        RECURRENT_BOUNDARY_ROLE
+    ] == "9911b3f9d21815a185285852b5a6176e5484e1ab0ff5c30f7caaa68ea0fab543"
+    assert "SchedulerOutput.recurrent_boundary_blocks" in by_path[
+        "vllm/v1/core/sched/output.py"
+    ]["required_symbols"]
+    assert "KVCacheManager.take_recurrent_boundary_blocks" in by_path[
+        "vllm/v1/core/kv_cache_manager.py"
+    ]["required_symbols"]
+
 
 def test_glm53_b12x_kda_contract_rejects_the_e105_kda_source(
     monkeypatch: pytest.MonkeyPatch,

sparkcache/runtime_patches/vllm-kv-block-lease-contract-glm53-b12x-kda-adaptive-mtp.json

@@ -24,9 +24,10 @@
       ]
     },
     {
-      "path": "vllm/v1/core/sched/scheduler.py",
-      "accepted_sha256": {
-        "source_built_glm53_b12x_kda_adaptive_mtp": "6d397c97f31e67a75efc01b5ddd89fa58db425de14fa43965ef2d6146b6b9bdb"
+    "path": "vllm/v1/core/sched/scheduler.py",
+    "accepted_sha256": {
+      "source_built_glm53_b12x_kda_adaptive_mtp": "6d397c97f31e67a75efc01b5ddd89fa58db425de14fa43965ef2d6146b6b9bdb",
+      "recurrent_boundary_contract": "260f36ce8fabf70c193b20009ea465eea7b1b6c8e9fb72f2307a01ba8fcf7b2a"
       },
       "required_symbols": [
         "Scheduler.schedule",
@@ -39,9 +40,10 @@
       ]
     },
     {
-      "path": "vllm/v1/core/kv_cache_manager.py",
-      "accepted_sha256": {
-        "source_built_glm53_b12x_kda_adaptive_mtp": "ee03dc9ce2b720c0be6e9f572d23580ba96eff68fe3406250557e83071654af0"
+    "path": "vllm/v1/core/kv_cache_manager.py",
+    "accepted_sha256": {
+      "source_built_glm53_b12x_kda_adaptive_mtp": "ee03dc9ce2b720c0be6e9f572d23580ba96eff68fe3406250557e83071654af0",
+      "recurrent_boundary_contract": "c5b83d382c96b2bf8c466a993ed77123a14a971e2661797128533319388d0b5f"
       },
       "required_symbols": [
         "KVCacheManager.get_block_ids",
@@ -51,17 +53,20 @@
         "KVCacheManager.expire_shared_prefix_leases",
         "KVCacheManager.discard_shared_prefix_lease",
         "KVCacheManager.evict_shared_prefix_leases_until_free",
-        "KVCacheManager.take_kv_cache_block_copies"
+      "KVCacheManager.take_kv_cache_block_copies",
+      "KVCacheManager.take_recurrent_boundary_blocks"
       ]
     },
     {
-      "path": "vllm/v1/core/sched/output.py",
-      "accepted_sha256": {
-        "source_built_glm53_b12x_kda_adaptive_mtp": "65235eba652e5a3ccee18bf3cbfeac9bf4da8fb9c61e961580f612cfb7e593bc"
+    "path": "vllm/v1/core/sched/output.py",
+    "accepted_sha256": {
+      "source_built_glm53_b12x_kda_adaptive_mtp": "65235eba652e5a3ccee18bf3cbfeac9bf4da8fb9c61e961580f612cfb7e593bc",
+      "recurrent_boundary_contract": "9911b3f9d21815a185285852b5a6176e5484e1ab0ff5c30f7caaa68ea0fab543"
       },
       "required_symbols": [
         "SchedulerOutput.preempted_req_ids",
-        "SchedulerOutput.kv_cache_block_copies"
+      "SchedulerOutput.kv_cache_block_copies",
+      "SchedulerOutput.recurrent_boundary_blocks"
       ]
     },
     {
@@ -96,9 +101,10 @@
       ]
     },
     {
-      "path": "vllm/v1/core/single_type_kv_cache_manager.py",
-      "accepted_sha256": {
-        "source_built_glm53_b12x_kda_adaptive_mtp": "e4b1c5c38b63b708fd55aa40a9ab0d008b266d006a63dcfcef55890ac1371cb8"
+    "path": "vllm/v1/core/single_type_kv_cache_manager.py",
+    "accepted_sha256": {
+      "source_built_glm53_b12x_kda_adaptive_mtp": "e4b1c5c38b63b708fd55aa40a9ab0d008b266d006a63dcfcef55890ac1371cb8",
+      "recurrent_boundary_contract": "f67a1850a7e0288baaa6d42e7ec55b22b09c156720767e23acaabedcae333c8a"
       },
       "required_symbols": [
         "SingleTypeKVCacheManager.add_local_computed_blocks",
@@ -107,7 +113,8 @@
         "SingleTypeKVCacheManager.take_pending_cow_copies",
         "SingleTypeKVCacheManager.pop_blocks_for_free",
         "MambaManager.remove_skipped_blocks",
-        "MambaManager.allocate_new_blocks"
+      "MambaManager.allocate_new_blocks",
+      "SingleTypeKVCacheManager.take_pending_aligned_recurrent_boundaries"
       ]
     },
     {