From 88f8bca65792f3fba48a878226a991665e66d1b5 Mon Sep 17 00:00:00 2001
From: Omer Aplatony <omerap12@gmail.com>
Date: Tue, 14 Jul 2026 12:14:40 +0300
Subject: [PATCH 1/6] feat: set ownerReference on ModelMetadata CRs for
 automatic GC on pod deletion

Signed-off-by: Omer Aplatony <omerap12@gmail.com>
---
 .../python/modelexpress/client.py             |  6 ++++++
 .../python/modelexpress/envs.py               |  4 ++++
 .../metadata/k8s_service_client.py            |  2 ++
 .../python/modelexpress/metadata/publish.py   | 14 ++++++++++++-
 modelexpress_common/proto/p2p.proto           |  9 +++++++++
 modelexpress_common/src/envs.rs               |  6 ++++++
 modelexpress_server/src/p2p/backend.rs        |  2 ++
 .../src/p2p/backend/kubernetes.rs             | 17 ++++++++++++++++
 modelexpress_server/src/p2p/backend/memory.rs | 20 ++++++++++---------
 modelexpress_server/src/p2p/backend/redis.rs  |  2 ++
 modelexpress_server/src/p2p/service.rs        | 12 ++++++++---
 modelexpress_server/src/p2p/state.rs          | 14 ++++++++-----
 12 files changed, 90 insertions(+), 18 deletions(-)

diff --git a/modelexpress_client/python/modelexpress/client.py b/modelexpress_client/python/modelexpress/client.py
index c48ab65b..0cbb5c8b 100644
--- a/modelexpress_client/python/modelexpress/client.py
+++ b/modelexpress_client/python/modelexpress/client.py
@@ -46,6 +46,8 @@ def publish_metadata(
         identity: "p2p_pb2.SourceIdentity",
         worker: "p2p_pb2.WorkerMetadata",
         worker_id: str,
+        pod_name: str = "",
+        pod_uid: str = "",
     ) -> str:
         """Publish worker metadata and return the computed mx_source_id."""
 
@@ -167,6 +169,8 @@ def publish_metadata(
         identity: "p2p_pb2.SourceIdentity",
         worker: "p2p_pb2.WorkerMetadata",
         worker_id: str,
+        pod_name: str = "",
+        pod_uid: str = "",
     ) -> str:
         """Publish metadata for one worker so targets can discover this source.
 
@@ -176,6 +180,8 @@ def publish_metadata(
             identity=identity,
             worker=worker,
             worker_id=worker_id,
+            pod_name=pod_name,
+            pod_uid=pod_uid,
         )
         response = self.stub.PublishMetadata(request, timeout=30)
         if not response.success:
diff --git a/modelexpress_client/python/modelexpress/envs.py b/modelexpress_client/python/modelexpress/envs.py
index 64e1b27d..8e3f242c 100644
--- a/modelexpress_client/python/modelexpress/envs.py
+++ b/modelexpress_client/python/modelexpress/envs.py
@@ -105,6 +105,8 @@
     VLLM_ATTENTION_BACKEND: str
     HOSTNAME: str
     POD_NAMESPACE: str
+    POD_NAME: str
+    POD_UID: str
 
 _TRUTHY = {"1", "true", "yes", "on"}
 
@@ -213,6 +215,8 @@ def _env_float(name: str, default: float) -> float:
     "VLLM_ATTENTION_BACKEND": lambda: os.environ.get("VLLM_ATTENTION_BACKEND", "auto"),
     "HOSTNAME": lambda: os.environ.get("HOSTNAME", ""),
     "POD_NAMESPACE": lambda: os.environ.get("POD_NAMESPACE", ""),
+    "POD_NAME": lambda: os.environ.get("POD_NAME",""),
+    "POD_UID": lambda: os.environ.get("POD_UID","")
 }
 
 
diff --git a/modelexpress_client/python/modelexpress/metadata/k8s_service_client.py b/modelexpress_client/python/modelexpress/metadata/k8s_service_client.py
index 886321f5..d4259997 100644
--- a/modelexpress_client/python/modelexpress/metadata/k8s_service_client.py
+++ b/modelexpress_client/python/modelexpress/metadata/k8s_service_client.py
@@ -104,6 +104,8 @@ def publish_metadata(
         identity: "p2p_pb2.SourceIdentity",
         worker: "p2p_pb2.WorkerMetadata",
         worker_id: str,
+        pod_name: str = "",
+        pod_uid: str = "",
     ) -> str:
         """Compute mx_source_id locally - there is no central store to hit.
 
diff --git a/modelexpress_client/python/modelexpress/metadata/publish.py b/modelexpress_client/python/modelexpress/metadata/publish.py
index 52c7a476..c999a024 100644
--- a/modelexpress_client/python/modelexpress/metadata/publish.py
+++ b/modelexpress_client/python/modelexpress/metadata/publish.py
@@ -127,6 +127,9 @@ def publish_metadata_and_ready(
         f"[Worker {worker_rank}] Preparing {len(tensors)} tensors for model '{identity.model_name}'"
     )
 
+    pod_name = envs.POD_NAME
+    pod_uid = envs.POD_UID
+
     tensor_protos = build_tensor_protos(tensors, device_id, worker_rank)
 
     if _is_p2p_metadata_enabled(mx_client):
@@ -171,6 +174,8 @@ def publish_fn() -> str:
                 worker=worker,
                 worker_id=worker_id,
                 worker_rank=worker_rank,
+                pod_name=pod_name,
+                pod_uid=pod_uid,
             )
             grpc_server.set_mx_source_id(mx_source_id)
             logger.info(
@@ -204,6 +209,8 @@ def publish_fn() -> str:
                 worker=worker,
                 worker_id=worker_id,
                 worker_rank=worker_rank,
+                pod_name=pod_name,
+                pod_uid=pod_uid,
             )
             logger.info(
                 f"[Worker {worker_rank}] Published metadata to MX server "
@@ -231,13 +238,18 @@ def _publish_metadata_to_server(
     worker: "p2p_pb2.WorkerMetadata",
     worker_id: str,
     worker_rank: int,
+    pod_name: str = "",
+    pod_uid: str = "",
 ) -> str:
     """Publish metadata with bounded retries and exponential backoff."""
     last_error: grpc.RpcError | None = None
 
     for attempt in range(1, PUBLISH_METADATA_MAX_ATTEMPTS + 1):
         try:
-            return mx_client.publish_metadata(identity, worker, worker_id)
+            return mx_client.publish_metadata(
+                identity, worker, worker_id,
+                pod_name=pod_name, pod_uid=pod_uid,
+            )
         except grpc.RpcError as exc:
             if exc.code() not in PUBLISH_METADATA_RETRYABLE_STATUS_CODES:
                 raise
diff --git a/modelexpress_common/proto/p2p.proto b/modelexpress_common/proto/p2p.proto
index 85ecd9d7..435e84db 100644
--- a/modelexpress_common/proto/p2p.proto
+++ b/modelexpress_common/proto/p2p.proto
@@ -499,6 +499,15 @@ message PublishMetadataRequest {
   // Differentiates multiple replicas with identical SourceIdentity.
   // If empty, the server rejects the request.
   string worker_id = 3;
+
+  // Kubernetes pod name of the calling worker, set via the downward API.
+  // Used with pod_uid to set an ownerReference on the ModelMetadata CR
+  // so K8s garbage-collects it when the pod is deleted.
+  // When empty, no ownerReference is set.
+  string pod_name = 4;
+
+  // Kubernetes pod UID of the calling worker, set via the downward API.
+  string pod_uid = 5;
 }
 
 message PublishMetadataResponse {
diff --git a/modelexpress_common/src/envs.rs b/modelexpress_common/src/envs.rs
index 33520b91..efab43da 100644
--- a/modelexpress_common/src/envs.rs
+++ b/modelexpress_common/src/envs.rs
@@ -107,6 +107,10 @@ pub const REDIS_PORT: &str = "REDIS_PORT";
 pub const MX_METADATA_NAMESPACE: &str = "MX_METADATA_NAMESPACE";
 /// Kubernetes namespace injected via the downward API for in-cluster pods.
 pub const POD_NAMESPACE: &str = "POD_NAMESPACE";
+/// Kubernetes pod name injected via the downward API (used by clients).
+pub const POD_NAME: &str = "POD_NAME";
+/// Kubernetes pod UID injected via the downward API (used by clients).
+pub const POD_UID: &str = "POD_UID";
 
 // ── Reaper (server) ─────────────────────────────────────────────────────────
 /// Interval (seconds) between reaper scans for stale/GC worker sweeps.
@@ -324,6 +328,8 @@ mod tests {
         assert_eq!(HOME, "HOME");
         assert_eq!(USERPROFILE, "USERPROFILE");
         assert_eq!(KUBECONFIG, "KUBECONFIG");
+        assert_eq!(POD_NAME, "POD_NAME");
+        assert_eq!(POD_UID, "POD_UID");
     }
 
     #[test]
diff --git a/modelexpress_server/src/p2p/backend.rs b/modelexpress_server/src/p2p/backend.rs
index fe1a05de..c00f6408 100644
--- a/modelexpress_server/src/p2p/backend.rs
+++ b/modelexpress_server/src/p2p/backend.rs
@@ -296,6 +296,8 @@ pub trait MetadataBackend: Send + Sync {
         identity: &SourceIdentity,
         worker_id: &str,
         worker: WorkerMetadata,
+        pod_name: &str,
+        pod_uid: &str,
     ) -> MetadataResult<()>;
 
     /// Get full tensor metadata for one specific worker.
diff --git a/modelexpress_server/src/p2p/backend/kubernetes.rs b/modelexpress_server/src/p2p/backend/kubernetes.rs
index a592492f..340cb6e0 100644
--- a/modelexpress_server/src/p2p/backend/kubernetes.rs
+++ b/modelexpress_server/src/p2p/backend/kubernetes.rs
@@ -190,6 +190,8 @@ impl MetadataBackend for KubernetesBackend {
         identity: &SourceIdentity,
         worker_id: &str,
         worker: WorkerMetadata,
+        pod_name: &str,
+        pod_uid: &str,
     ) -> MetadataResult<()> {
         let source_id = crate::p2p::source_identity::compute_mx_source_id(identity);
         let source_id = source_id.as_str();
@@ -202,6 +204,7 @@ impl MetadataBackend for KubernetesBackend {
 
         // First, ensure the CR exists
         let existing = api.get_opt(&cr_name).await?;
+        let has_pod_name_and_uid = !pod_name.is_empty() && !pod_uid.is_empty();
 
         if existing.is_none() {
             let new_cr = ModelMetadata {
@@ -220,6 +223,20 @@ impl MetadataBackend for KubernetesBackend {
                         );
                         labels
                     }),
+                    owner_references: if has_pod_name_and_uid {
+                        Some(vec![
+                            k8s_openapi::apimachinery::pkg::apis::meta::v1::OwnerReference {
+                                api_version: "v1".to_string(),
+                                kind: "Pod".to_string(),
+                                name: pod_name.to_string(),
+                                uid: pod_uid.to_string(),
+                                controller: Some(false),
+                                block_owner_deletion: Some(false),
+                            },
+                        ])
+                    } else {
+                        None
+                    },
                     ..Default::default()
                 },
                 spec: ModelMetadataSpec {
diff --git a/modelexpress_server/src/p2p/backend/memory.rs b/modelexpress_server/src/p2p/backend/memory.rs
index 50b6983f..a32f9370 100644
--- a/modelexpress_server/src/p2p/backend/memory.rs
+++ b/modelexpress_server/src/p2p/backend/memory.rs
@@ -59,6 +59,8 @@ impl MetadataBackend for InMemoryMetadataBackend {
         identity: &SourceIdentity,
         worker_id: &str,
         worker: WorkerMetadata,
+        _pod_name: &str,
+        _pod_uid: &str,
     ) -> MetadataResult<()> {
         let source_id = crate::p2p::source_identity::compute_mx_source_id(identity);
         let record = WorkerRecord::from(worker);
@@ -233,7 +235,7 @@ mod tests {
         let source_id = compute_mx_source_id(&id);
 
         backend
-            .publish_metadata(&id, "w1", worker(0, SourceStatus::Ready))
+            .publish_metadata(&id, "w1", worker(0, SourceStatus::Ready), "", "")
             .await
             .expect("publish");
         let record = backend
@@ -267,7 +269,7 @@ mod tests {
 
         for rank in [2, 0, 1] {
             backend
-                .publish_metadata(&id, "w1", worker(rank, SourceStatus::Ready))
+                .publish_metadata(&id, "w1", worker(rank, SourceStatus::Ready), "", "")
                 .await
                 .expect("publish");
         }
@@ -288,11 +290,11 @@ mod tests {
         let id = identity("m");
 
         backend
-            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing))
+            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing), "", "")
             .await
             .expect("publish r0");
         backend
-            .publish_metadata(&id, "w1", worker(1, SourceStatus::Ready))
+            .publish_metadata(&id, "w1", worker(1, SourceStatus::Ready), "", "")
             .await
             .expect("publish r1");
 
@@ -332,11 +334,11 @@ mod tests {
         let source_id = compute_mx_source_id(&id);
 
         backend
-            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing))
+            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing), "", "")
             .await
             .expect("publish r0");
         backend
-            .publish_metadata(&id, "w1", worker(3, SourceStatus::Ready))
+            .publish_metadata(&id, "w1", worker(3, SourceStatus::Ready), "", "")
             .await
             .expect("publish r3");
 
@@ -361,7 +363,7 @@ mod tests {
         let source_id = compute_mx_source_id(&id);
 
         backend
-            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing))
+            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing), "", "")
             .await
             .expect("publish");
         backend
@@ -401,11 +403,11 @@ mod tests {
         let source_id = compute_mx_source_id(&id);
 
         backend
-            .publish_metadata(&id, "w1", worker(0, SourceStatus::Ready))
+            .publish_metadata(&id, "w1", worker(0, SourceStatus::Ready), "", "")
             .await
             .expect("publish w1");
         backend
-            .publish_metadata(&id, "w2", worker(0, SourceStatus::Ready))
+            .publish_metadata(&id, "w2", worker(0, SourceStatus::Ready), "", "")
             .await
             .expect("publish w2");
 
diff --git a/modelexpress_server/src/p2p/backend/redis.rs b/modelexpress_server/src/p2p/backend/redis.rs
index edbd20af..c9aee250 100644
--- a/modelexpress_server/src/p2p/backend/redis.rs
+++ b/modelexpress_server/src/p2p/backend/redis.rs
@@ -407,6 +407,8 @@ impl MetadataBackend for RedisBackend {
         identity: &SourceIdentity,
         worker_id: &str,
         worker: WorkerMetadata,
+        _pod_name: &str,
+        _pod_uid: &str,
     ) -> MetadataResult<()> {
         let source_id = crate::p2p::source_identity::compute_mx_source_id(identity);
         let mut conn = self.get_conn().await?;
diff --git a/modelexpress_server/src/p2p/service.rs b/modelexpress_server/src/p2p/service.rs
index 523b8d3e..32d4eb24 100644
--- a/modelexpress_server/src/p2p/service.rs
+++ b/modelexpress_server/src/p2p/service.rs
@@ -116,7 +116,7 @@ impl P2pService for P2pServiceImpl {
 
         match self
             .state
-            .publish_metadata(&identity, &worker_id, worker)
+            .publish_metadata(&identity, &worker_id, worker, &req.pod_name, &req.pod_uid)
             .await
         {
             Ok(()) => {
@@ -385,6 +385,7 @@ mod tests {
                 identity: None,
                 worker: None,
                 worker_id: "worker-uuid-1".to_string(),
+                ..Default::default()
             }))
             .await
             .expect("rpc")
@@ -403,6 +404,7 @@ mod tests {
                 identity: Some(id),
                 worker: None,
                 worker_id: "worker-uuid-1".to_string(),
+                ..Default::default()
             }))
             .await
             .expect("rpc")
@@ -418,6 +420,7 @@ mod tests {
                 identity: Some(test_identity()),
                 worker: None,
                 worker_id: String::new(),
+                ..Default::default()
             }))
             .await
             .expect("rpc")
@@ -431,7 +434,7 @@ mod tests {
         let mut mock = MockMetadataBackend::new();
         mock.expect_publish_metadata()
             .once()
-            .returning(|_, _, _| Ok(()));
+            .returning(|_, _, _, _, _| Ok(()));
 
         let svc = make_service(mock);
         let resp = svc
@@ -448,6 +451,7 @@ mod tests {
                     ..Default::default()
                 }),
                 worker_id: "worker-uuid-1".to_string(),
+                ..Default::default()
             }))
             .await
             .expect("rpc")
@@ -463,7 +467,7 @@ mod tests {
         let mut mock = MockMetadataBackend::new();
         mock.expect_publish_metadata()
             .once()
-            .returning(|_, _, _| Err("storage unavailable".into()));
+            .returning(|_, _, _, _, _| Err("storage unavailable".into()));
 
         let svc = make_service(mock);
         let resp = svc
@@ -478,6 +482,7 @@ mod tests {
                     ..Default::default()
                 }),
                 worker_id: "worker-uuid-1".to_string(),
+                ..Default::default()
             }))
             .await
             .expect("rpc")
@@ -649,6 +654,7 @@ mod tests {
                 identity: Some(test_identity()),
                 worker: None,
                 worker_id: "worker-uuid-1".to_string(),
+                ..Default::default()
             }))
             .await
             .expect("rpc")
diff --git a/modelexpress_server/src/p2p/state.rs b/modelexpress_server/src/p2p/state.rs
index 4a42b183..5fddc950 100644
--- a/modelexpress_server/src/p2p/state.rs
+++ b/modelexpress_server/src/p2p/state.rs
@@ -95,10 +95,12 @@ impl P2pStateManager {
         identity: &SourceIdentity,
         worker_id: &str,
         worker: WorkerMetadata,
+        pod_name: &str,
+        pod_uid: &str,
     ) -> MetadataResult<()> {
         self.get_backend()
             .await?
-            .publish_metadata(identity, worker_id, worker)
+            .publish_metadata(identity, worker_id, worker, pod_name, pod_uid)
             .await
     }
 
@@ -457,14 +459,14 @@ mod tests {
     async fn test_publish_metadata_calls_backend() {
         let mut mock = MockMetadataBackend::new();
         mock.expect_publish_metadata()
-            .withf(|identity, worker_id, worker| {
+            .withf(|identity, worker_id, worker, _, _| {
                 identity.model_name == "my-model"
                     && identity.tensor_parallel_size == 8
                     && worker_id == "a1b2c3d4"
                     && worker.worker_rank == 3
             })
             .once()
-            .returning(|_, _, _| Ok(()));
+            .returning(|_, _, _, _, _| Ok(()));
 
         let manager = P2pStateManager::with_backend(Arc::new(mock));
         manager
@@ -478,6 +480,8 @@ mod tests {
                     updated_at: 0,
                     ..Default::default()
                 },
+                "",
+                "",
             )
             .await
             .expect("publish_metadata failed");
@@ -488,12 +492,12 @@ mod tests {
         let mut mock = MockMetadataBackend::new();
         mock.expect_publish_metadata()
             .once()
-            .returning(|_, _, _| Err("storage unavailable".into()));
+            .returning(|_, _, _, _, _| Err("storage unavailable".into()));
 
         let manager = P2pStateManager::with_backend(Arc::new(mock));
         assert!(
             manager
-                .publish_metadata(&test_identity(), "a1b2c3d4", WorkerMetadata::default())
+                .publish_metadata(&test_identity(), "a1b2c3d4", WorkerMetadata::default(), "", "")
                 .await
                 .is_err()
         );

From 4d96cc39656d69ce701e8c93ff2d1d3965bdb0df Mon Sep 17 00:00:00 2001
From: Omer Aplatony <omerap12@gmail.com>
Date: Tue, 14 Jul 2026 13:41:29 +0300
Subject: [PATCH 2/6] update proto

Signed-off-by: Omer Aplatony <omerap12@gmail.com>
---
 .../python/modelexpress/p2p_pb2.py            | 56 +++++++++----------
 1 file changed, 28 insertions(+), 28 deletions(-)

diff --git a/modelexpress_client/python/modelexpress/p2p_pb2.py b/modelexpress_client/python/modelexpress/p2p_pb2.py
index faaaec2c..654da8cc 100644
--- a/modelexpress_client/python/modelexpress/p2p_pb2.py
+++ b/modelexpress_client/python/modelexpress/p2p_pb2.py
@@ -27,7 +27,7 @@
 
 
 
-DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\tp2p.proto\x12\x11model_express.p2p\"\xe7\x04\n\x0eSourceIdentity\x12\x12\n\nmx_version\x18\x01 \x01(\t\x12\x37\n\x0emx_source_type\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\nmodel_name\x18\x03 \x01(\t\x12>\n\x11\x62\x61\x63kend_framework\x18\x04 \x01(\x0e\x32#.model_express.p2p.BackendFramework\x12\x1c\n\x14tensor_parallel_size\x18\x05 \x01(\r\x12\x1e\n\x16pipeline_parallel_size\x18\x06 \x01(\r\x12\x1c\n\x14\x65xpert_parallel_size\x18\x07 \x01(\r\x12\r\n\x05\x64type\x18\x08 \x01(\t\x12\x14\n\x0cquantization\x18\t \x01(\t\x12P\n\x10\x65xtra_parameters\x18\n \x03(\x0b\x32\x36.model_express.p2p.SourceIdentity.ExtraParametersEntry\x12\x10\n\x08revision\x18\x0b \x01(\t\x12!\n\x19\x62\x61\x63kend_framework_version\x18\x0c \x01(\t\x12\x15\n\rtorch_version\x18\r \x01(\t\x12\x14\n\x0c\x63uda_version\x18\x0e \x01(\t\x12\x16\n\x0etriton_version\x18\x0f \x01(\t\x12\x10\n\x08gpu_arch\x18\x10 \x01(\t\x12\x1d\n\x15\x63ompile_config_digest\x18\x11 \x01(\t\x1a\x36\n\x14\x45xtraParametersEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\t:\x02\x38\x01\"^\n\x10TensorDescriptor\x12\x0c\n\x04name\x18\x01 \x01(\t\x12\x0c\n\x04\x61\x64\x64r\x18\x02 \x01(\x04\x12\x0c\n\x04size\x18\x03 \x01(\x04\x12\x11\n\tdevice_id\x18\x04 \x01(\r\x12\r\n\x05\x64type\x18\x05 \x01(\t\"}\n\x16\x41rtifactSourceMetadata\x12\x13\n\x0b\x61rtifact_id\x18\x01 \x01(\t\x12\x12\n\ntotal_size\x18\x02 \x01(\x04\x12\x12\n\nfile_count\x18\x03 \x01(\r\x12\x13\n\x0b\x63hunk_count\x18\x04 \x01(\r\x12\x11\n\tnode_rank\x18\x05 \x01(\r\"L\n\x14TensorSourceMetadata\x12\x34\n\x07tensors\x18\x01 \x03(\x0b\x32#.model_express.p2p.TensorDescriptor\"\xeb\x01\n\x10\x41rtifactManifest\x12\x18\n\x10manifest_version\x18\x01 \x01(\r\x12\x37\n\x0emx_source_type\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\nchunk_size\x18\x03 \x01(\x04\x12\x36\n\x05\x66iles\x18\x04 \x03(\x0b\x32\'.model_express.p2p.ArtifactManifestFile\x12\x38\n\x06\x63hunks\x18\x05 \x03(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\"X\n\x14\x41rtifactManifestFile\x12\x12\n\nfile_index\x18\x01 \x01(\r\x12\x0c\n\x04path\x18\x02 \x01(\t\x12\x0c\n\x04size\x18\x03 \x01(\x04\x12\x10\n\x08\x63hecksum\x18\x04 \x01(\t\"w\n\x15\x41rtifactManifestChunk\x12\x13\n\x0b\x63hunk_index\x18\x01 \x01(\r\x12\x12\n\nfile_index\x18\x02 \x01(\r\x12\x13\n\x0b\x66ile_offset\x18\x03 \x01(\x04\x12\x0e\n\x06length\x18\x04 \x01(\x04\x12\x10\n\x08\x63hecksum\x18\x05 \x01(\t\"\xf3\x03\n\x0eWorkerMetadata\x12\x13\n\x0bworker_rank\x18\x01 \x01(\r\x12\x17\n\rnixl_metadata\x18\x02 \x01(\x0cH\x00\x12$\n\x1atransfer_engine_session_id\x18\n \x01(\tH\x00\x12\x38\n\x07tensors\x18\x03 \x03(\x0b\x32#.model_express.p2p.TensorDescriptorB\x02\x18\x01\x12/\n\x06status\x18\x04 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatus\x12\x12\n\nupdated_at\x18\x05 \x01(\x03\x12\x19\n\x11metadata_endpoint\x18\x06 \x01(\t\x12\x12\n\nagent_name\x18\x07 \x01(\t\x12\x1c\n\x14worker_grpc_endpoint\x18\x08 \x01(\t\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\t \x01(\t\x12@\n\rtensor_source\x18\x14 \x01(\x0b\x32\'.model_express.p2p.TensorSourceMetadataH\x01\x12\x44\n\x0f\x61rtifact_source\x18\x15 \x01(\x0b\x32).model_express.p2p.ArtifactSourceMetadataH\x01\x42\x12\n\x10\x62\x61\x63kend_metadataB\x10\n\x0esource_payload\"V\n\x18GetTensorManifestRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x16\n\tworker_id\x18\x02 \x01(\tH\x00\x88\x01\x01\x42\x0c\n\n_worker_id\"\xe6\x01\n\x19GetTensorManifestResponse\x12\x34\n\x07tensors\x18\x01 \x03(\x0b\x32#.model_express.p2p.TensorDescriptor\x12\x14\n\x0cmx_source_id\x18\x02 \x01(\t\x12\x19\n\x11metadata_endpoint\x18\x03 \x01(\t\x12\x12\n\nagent_name\x18\x04 \x01(\t\x12\x13\n\x0bworker_rank\x18\x05 \x01(\r\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\x06 \x01(\t\x12\x16\n\tworker_id\x18\x07 \x01(\tH\x00\x88\x01\x01\x42\x0c\n\n_worker_id\"M\n GetArtifactManifestHeaderRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\"\xee\x02\n!GetArtifactManifestHeaderResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x18\n\x10manifest_version\x18\x03 \x01(\r\x12\x37\n\x0emx_source_type\x18\x04 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\ntotal_size\x18\x05 \x01(\x04\x12\x12\n\nfile_count\x18\x06 \x01(\r\x12\x13\n\x0b\x63hunk_count\x18\x07 \x01(\r\x12\x12\n\nchunk_size\x18\x08 \x01(\x04\x12\x19\n\x11metadata_endpoint\x18\t \x01(\t\x12\x12\n\nagent_name\x18\n \x01(\t\x12\x13\n\x0bworker_rank\x18\x0b \x01(\r\x12\x36\n\x05\x66iles\x18\x0c \x03(\x0b\x32\'.model_express.p2p.ArtifactManifestFile\"|\n GetArtifactManifestChunksRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x19\n\x11start_chunk_index\x18\x03 \x01(\r\x12\x12\n\nmax_chunks\x18\x04 \x01(\r\"\xbc\x01\n!GetArtifactManifestChunksResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x19\n\x11start_chunk_index\x18\x03 \x01(\r\x12\x38\n\x06\x63hunks\x18\x04 \x03(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\x12\x17\n\x0fnext_page_token\x18\x05 \x01(\t\"R\n\x1f\x41rtifactChunkTransferDescriptor\x12\x0c\n\x04\x61\x64\x64r\x18\x01 \x01(\x04\x12\x0e\n\x06length\x18\x02 \x01(\x04\x12\x11\n\tdevice_id\x18\x03 \x01(\r\"]\n\x1bPrepareArtifactChunkRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x13\n\x0b\x63hunk_index\x18\x03 \x01(\r\"\xf1\x01\n\x1cPrepareArtifactChunkResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x10\n\x08lease_id\x18\x03 \x01(\t\x12\x37\n\x05\x63hunk\x18\x04 \x01(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\x12\x42\n\x06source\x18\x05 \x01(\x0b\x32\x32.model_express.p2p.ArtifactChunkTransferDescriptor\x12\x17\n\x0fsource_metadata\x18\x06 \x01(\x0c\"Z\n\x1bReleaseArtifactChunkRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x10\n\x08lease_id\x18\x03 \x01(\t\"\x82\x01\n\x1cReleaseArtifactChunkResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x37\n\x05\x63hunk\x18\x03 \x01(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\"\x93\x01\n\x16PublishMetadataRequest\x12\x33\n\x08identity\x18\x01 \x01(\x0b\x32!.model_express.p2p.SourceIdentity\x12\x31\n\x06worker\x18\x02 \x01(\x0b\x32!.model_express.p2p.WorkerMetadata\x12\x11\n\tworker_id\x18\x03 \x01(\t\"d\n\x17PublishMetadataResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\x12\x14\n\x0cmx_source_id\x18\x03 \x01(\t\x12\x11\n\tworker_id\x18\x04 \x01(\t\"z\n\x11SourceInstanceRef\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x11\n\tworker_id\x18\x02 \x01(\t\x12\x12\n\nmodel_name\x18\x03 \x01(\t\x12\x13\n\x0bworker_rank\x18\x04 \x01(\r\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\x05 \x01(\t\"\x98\x01\n\x12ListSourcesRequest\x12\x33\n\x08identity\x18\x01 \x01(\x0b\x32!.model_express.p2p.SourceIdentity\x12;\n\rstatus_filter\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatusH\x00\x88\x01\x01\x42\x10\n\x0e_status_filter\"N\n\x13ListSourcesResponse\x12\x37\n\tinstances\x18\x01 \x03(\x0b\x32$.model_express.p2p.SourceInstanceRef\"=\n\x12GetMetadataRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x11\n\tworker_id\x18\x02 \x01(\t\"\x80\x01\n\x13GetMetadataResponse\x12\r\n\x05\x66ound\x18\x01 \x01(\x08\x12\x31\n\x06worker\x18\x02 \x01(\x0b\x32!.model_express.p2p.WorkerMetadata\x12\x14\n\x0cmx_source_id\x18\x03 \x01(\t\x12\x11\n\tworker_id\x18\x04 \x01(\t\"\x84\x01\n\x13UpdateStatusRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0bworker_rank\x18\x02 \x01(\r\x12/\n\x06status\x18\x03 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatus\x12\x11\n\tworker_id\x18\x04 \x01(\t\"8\n\x14UpdateStatusResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t*\x8a\x01\n\x10\x42\x61\x63kendFramework\x12\x1d\n\x19\x42\x41\x43KEND_FRAMEWORK_UNKNOWN\x10\x00\x12\x1a\n\x16\x42\x41\x43KEND_FRAMEWORK_VLLM\x10\x01\x12\x1c\n\x18\x42\x41\x43KEND_FRAMEWORK_SGLANG\x10\x02\x12\x1d\n\x19\x42\x41\x43KEND_FRAMEWORK_TRT_LLM\x10\x03*\xba\x02\n\x0cMxSourceType\x12\x1a\n\x16MX_SOURCE_TYPE_WEIGHTS\x10\x00\x12\x17\n\x13MX_SOURCE_TYPE_LORA\x10\x01\x12\x1d\n\x19MX_SOURCE_TYPE_CUDA_GRAPH\x10\x02\x12&\n\"MX_SOURCE_TYPE_TORCH_COMPILE_CACHE\x10\x03\x12\x1f\n\x1bMX_SOURCE_TYPE_TRITON_CACHE\x10\x04\x12\"\n\x1eMX_SOURCE_TYPE_DEEP_GEMM_CACHE\x10\x05\x12!\n\x1dMX_SOURCE_TYPE_TILELANG_CACHE\x10\x06\x12!\n\x1dMX_SOURCE_TYPE_CUTE_DSL_CACHE\x10\x07\x12#\n\x1fMX_SOURCE_TYPE_FLASHINFER_CACHE\x10\x08*{\n\x0cSourceStatus\x12\x19\n\x15SOURCE_STATUS_UNKNOWN\x10\x00\x12\x1e\n\x1aSOURCE_STATUS_INITIALIZING\x10\x01\x12\x17\n\x13SOURCE_STATUS_READY\x10\x02\x12\x17\n\x13SOURCE_STATUS_STALE\x10\x03\x32\x93\x03\n\nP2pService\x12h\n\x0fPublishMetadata\x12).model_express.p2p.PublishMetadataRequest\x1a*.model_express.p2p.PublishMetadataResponse\x12\\\n\x0bListSources\x12%.model_express.p2p.ListSourcesRequest\x1a&.model_express.p2p.ListSourcesResponse\x12\\\n\x0bGetMetadata\x12%.model_express.p2p.GetMetadataRequest\x1a&.model_express.p2p.GetMetadataResponse\x12_\n\x0cUpdateStatus\x12&.model_express.p2p.UpdateStatusRequest\x1a\'.model_express.p2p.UpdateStatusResponse2\x83\x05\n\rWorkerService\x12n\n\x11GetTensorManifest\x12+.model_express.p2p.GetTensorManifestRequest\x1a,.model_express.p2p.GetTensorManifestResponse\x12\x86\x01\n\x19GetArtifactManifestHeader\x12\x33.model_express.p2p.GetArtifactManifestHeaderRequest\x1a\x34.model_express.p2p.GetArtifactManifestHeaderResponse\x12\x86\x01\n\x19GetArtifactManifestChunks\x12\x33.model_express.p2p.GetArtifactManifestChunksRequest\x1a\x34.model_express.p2p.GetArtifactManifestChunksResponse\x12w\n\x14PrepareArtifactChunk\x12..model_express.p2p.PrepareArtifactChunkRequest\x1a/.model_express.p2p.PrepareArtifactChunkResponse\x12w\n\x14ReleaseArtifactChunk\x12..model_express.p2p.ReleaseArtifactChunkRequest\x1a/.model_express.p2p.ReleaseArtifactChunkResponseb\x06proto3')
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\tp2p.proto\x12\x11model_express.p2p\"\xe7\x04\n\x0eSourceIdentity\x12\x12\n\nmx_version\x18\x01 \x01(\t\x12\x37\n\x0emx_source_type\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\nmodel_name\x18\x03 \x01(\t\x12>\n\x11\x62\x61\x63kend_framework\x18\x04 \x01(\x0e\x32#.model_express.p2p.BackendFramework\x12\x1c\n\x14tensor_parallel_size\x18\x05 \x01(\r\x12\x1e\n\x16pipeline_parallel_size\x18\x06 \x01(\r\x12\x1c\n\x14\x65xpert_parallel_size\x18\x07 \x01(\r\x12\r\n\x05\x64type\x18\x08 \x01(\t\x12\x14\n\x0cquantization\x18\t \x01(\t\x12P\n\x10\x65xtra_parameters\x18\n \x03(\x0b\x32\x36.model_express.p2p.SourceIdentity.ExtraParametersEntry\x12\x10\n\x08revision\x18\x0b \x01(\t\x12!\n\x19\x62\x61\x63kend_framework_version\x18\x0c \x01(\t\x12\x15\n\rtorch_version\x18\r \x01(\t\x12\x14\n\x0c\x63uda_version\x18\x0e \x01(\t\x12\x16\n\x0etriton_version\x18\x0f \x01(\t\x12\x10\n\x08gpu_arch\x18\x10 \x01(\t\x12\x1d\n\x15\x63ompile_config_digest\x18\x11 \x01(\t\x1a\x36\n\x14\x45xtraParametersEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\t:\x02\x38\x01\"^\n\x10TensorDescriptor\x12\x0c\n\x04name\x18\x01 \x01(\t\x12\x0c\n\x04\x61\x64\x64r\x18\x02 \x01(\x04\x12\x0c\n\x04size\x18\x03 \x01(\x04\x12\x11\n\tdevice_id\x18\x04 \x01(\r\x12\r\n\x05\x64type\x18\x05 \x01(\t\"}\n\x16\x41rtifactSourceMetadata\x12\x13\n\x0b\x61rtifact_id\x18\x01 \x01(\t\x12\x12\n\ntotal_size\x18\x02 \x01(\x04\x12\x12\n\nfile_count\x18\x03 \x01(\r\x12\x13\n\x0b\x63hunk_count\x18\x04 \x01(\r\x12\x11\n\tnode_rank\x18\x05 \x01(\r\"L\n\x14TensorSourceMetadata\x12\x34\n\x07tensors\x18\x01 \x03(\x0b\x32#.model_express.p2p.TensorDescriptor\"\xeb\x01\n\x10\x41rtifactManifest\x12\x18\n\x10manifest_version\x18\x01 \x01(\r\x12\x37\n\x0emx_source_type\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\nchunk_size\x18\x03 \x01(\x04\x12\x36\n\x05\x66iles\x18\x04 \x03(\x0b\x32\'.model_express.p2p.ArtifactManifestFile\x12\x38\n\x06\x63hunks\x18\x05 \x03(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\"X\n\x14\x41rtifactManifestFile\x12\x12\n\nfile_index\x18\x01 \x01(\r\x12\x0c\n\x04path\x18\x02 \x01(\t\x12\x0c\n\x04size\x18\x03 \x01(\x04\x12\x10\n\x08\x63hecksum\x18\x04 \x01(\t\"w\n\x15\x41rtifactManifestChunk\x12\x13\n\x0b\x63hunk_index\x18\x01 \x01(\r\x12\x12\n\nfile_index\x18\x02 \x01(\r\x12\x13\n\x0b\x66ile_offset\x18\x03 \x01(\x04\x12\x0e\n\x06length\x18\x04 \x01(\x04\x12\x10\n\x08\x63hecksum\x18\x05 \x01(\t\"\xf3\x03\n\x0eWorkerMetadata\x12\x13\n\x0bworker_rank\x18\x01 \x01(\r\x12\x17\n\rnixl_metadata\x18\x02 \x01(\x0cH\x00\x12$\n\x1atransfer_engine_session_id\x18\n \x01(\tH\x00\x12\x38\n\x07tensors\x18\x03 \x03(\x0b\x32#.model_express.p2p.TensorDescriptorB\x02\x18\x01\x12/\n\x06status\x18\x04 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatus\x12\x12\n\nupdated_at\x18\x05 \x01(\x03\x12\x19\n\x11metadata_endpoint\x18\x06 \x01(\t\x12\x12\n\nagent_name\x18\x07 \x01(\t\x12\x1c\n\x14worker_grpc_endpoint\x18\x08 \x01(\t\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\t \x01(\t\x12@\n\rtensor_source\x18\x14 \x01(\x0b\x32\'.model_express.p2p.TensorSourceMetadataH\x01\x12\x44\n\x0f\x61rtifact_source\x18\x15 \x01(\x0b\x32).model_express.p2p.ArtifactSourceMetadataH\x01\x42\x12\n\x10\x62\x61\x63kend_metadataB\x10\n\x0esource_payload\"V\n\x18GetTensorManifestRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x16\n\tworker_id\x18\x02 \x01(\tH\x00\x88\x01\x01\x42\x0c\n\n_worker_id\"\xe6\x01\n\x19GetTensorManifestResponse\x12\x34\n\x07tensors\x18\x01 \x03(\x0b\x32#.model_express.p2p.TensorDescriptor\x12\x14\n\x0cmx_source_id\x18\x02 \x01(\t\x12\x19\n\x11metadata_endpoint\x18\x03 \x01(\t\x12\x12\n\nagent_name\x18\x04 \x01(\t\x12\x13\n\x0bworker_rank\x18\x05 \x01(\r\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\x06 \x01(\t\x12\x16\n\tworker_id\x18\x07 \x01(\tH\x00\x88\x01\x01\x42\x0c\n\n_worker_id\"M\n GetArtifactManifestHeaderRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\"\xee\x02\n!GetArtifactManifestHeaderResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x18\n\x10manifest_version\x18\x03 \x01(\r\x12\x37\n\x0emx_source_type\x18\x04 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\ntotal_size\x18\x05 \x01(\x04\x12\x12\n\nfile_count\x18\x06 \x01(\r\x12\x13\n\x0b\x63hunk_count\x18\x07 \x01(\r\x12\x12\n\nchunk_size\x18\x08 \x01(\x04\x12\x19\n\x11metadata_endpoint\x18\t \x01(\t\x12\x12\n\nagent_name\x18\n \x01(\t\x12\x13\n\x0bworker_rank\x18\x0b \x01(\r\x12\x36\n\x05\x66iles\x18\x0c \x03(\x0b\x32\'.model_express.p2p.ArtifactManifestFile\"|\n GetArtifactManifestChunksRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x19\n\x11start_chunk_index\x18\x03 \x01(\r\x12\x12\n\nmax_chunks\x18\x04 \x01(\r\"\xbc\x01\n!GetArtifactManifestChunksResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x19\n\x11start_chunk_index\x18\x03 \x01(\r\x12\x38\n\x06\x63hunks\x18\x04 \x03(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\x12\x17\n\x0fnext_page_token\x18\x05 \x01(\t\"R\n\x1f\x41rtifactChunkTransferDescriptor\x12\x0c\n\x04\x61\x64\x64r\x18\x01 \x01(\x04\x12\x0e\n\x06length\x18\x02 \x01(\x04\x12\x11\n\tdevice_id\x18\x03 \x01(\r\"]\n\x1bPrepareArtifactChunkRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x13\n\x0b\x63hunk_index\x18\x03 \x01(\r\"\xf1\x01\n\x1cPrepareArtifactChunkResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x10\n\x08lease_id\x18\x03 \x01(\t\x12\x37\n\x05\x63hunk\x18\x04 \x01(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\x12\x42\n\x06source\x18\x05 \x01(\x0b\x32\x32.model_express.p2p.ArtifactChunkTransferDescriptor\x12\x17\n\x0fsource_metadata\x18\x06 \x01(\x0c\"Z\n\x1bReleaseArtifactChunkRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x10\n\x08lease_id\x18\x03 \x01(\t\"\x82\x01\n\x1cReleaseArtifactChunkResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x37\n\x05\x63hunk\x18\x03 \x01(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\"\xb6\x01\n\x16PublishMetadataRequest\x12\x33\n\x08identity\x18\x01 \x01(\x0b\x32!.model_express.p2p.SourceIdentity\x12\x31\n\x06worker\x18\x02 \x01(\x0b\x32!.model_express.p2p.WorkerMetadata\x12\x11\n\tworker_id\x18\x03 \x01(\t\x12\x10\n\x08pod_name\x18\x04 \x01(\t\x12\x0f\n\x07pod_uid\x18\x05 \x01(\t\"d\n\x17PublishMetadataResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\x12\x14\n\x0cmx_source_id\x18\x03 \x01(\t\x12\x11\n\tworker_id\x18\x04 \x01(\t\"z\n\x11SourceInstanceRef\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x11\n\tworker_id\x18\x02 \x01(\t\x12\x12\n\nmodel_name\x18\x03 \x01(\t\x12\x13\n\x0bworker_rank\x18\x04 \x01(\r\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\x05 \x01(\t\"\x98\x01\n\x12ListSourcesRequest\x12\x33\n\x08identity\x18\x01 \x01(\x0b\x32!.model_express.p2p.SourceIdentity\x12;\n\rstatus_filter\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatusH\x00\x88\x01\x01\x42\x10\n\x0e_status_filter\"N\n\x13ListSourcesResponse\x12\x37\n\tinstances\x18\x01 \x03(\x0b\x32$.model_express.p2p.SourceInstanceRef\"=\n\x12GetMetadataRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x11\n\tworker_id\x18\x02 \x01(\t\"\x80\x01\n\x13GetMetadataResponse\x12\r\n\x05\x66ound\x18\x01 \x01(\x08\x12\x31\n\x06worker\x18\x02 \x01(\x0b\x32!.model_express.p2p.WorkerMetadata\x12\x14\n\x0cmx_source_id\x18\x03 \x01(\t\x12\x11\n\tworker_id\x18\x04 \x01(\t\"\x84\x01\n\x13UpdateStatusRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0bworker_rank\x18\x02 \x01(\r\x12/\n\x06status\x18\x03 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatus\x12\x11\n\tworker_id\x18\x04 \x01(\t\"8\n\x14UpdateStatusResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t*\x8a\x01\n\x10\x42\x61\x63kendFramework\x12\x1d\n\x19\x42\x41\x43KEND_FRAMEWORK_UNKNOWN\x10\x00\x12\x1a\n\x16\x42\x41\x43KEND_FRAMEWORK_VLLM\x10\x01\x12\x1c\n\x18\x42\x41\x43KEND_FRAMEWORK_SGLANG\x10\x02\x12\x1d\n\x19\x42\x41\x43KEND_FRAMEWORK_TRT_LLM\x10\x03*\xba\x02\n\x0cMxSourceType\x12\x1a\n\x16MX_SOURCE_TYPE_WEIGHTS\x10\x00\x12\x17\n\x13MX_SOURCE_TYPE_LORA\x10\x01\x12\x1d\n\x19MX_SOURCE_TYPE_CUDA_GRAPH\x10\x02\x12&\n\"MX_SOURCE_TYPE_TORCH_COMPILE_CACHE\x10\x03\x12\x1f\n\x1bMX_SOURCE_TYPE_TRITON_CACHE\x10\x04\x12\"\n\x1eMX_SOURCE_TYPE_DEEP_GEMM_CACHE\x10\x05\x12!\n\x1dMX_SOURCE_TYPE_TILELANG_CACHE\x10\x06\x12!\n\x1dMX_SOURCE_TYPE_CUTE_DSL_CACHE\x10\x07\x12#\n\x1fMX_SOURCE_TYPE_FLASHINFER_CACHE\x10\x08*{\n\x0cSourceStatus\x12\x19\n\x15SOURCE_STATUS_UNKNOWN\x10\x00\x12\x1e\n\x1aSOURCE_STATUS_INITIALIZING\x10\x01\x12\x17\n\x13SOURCE_STATUS_READY\x10\x02\x12\x17\n\x13SOURCE_STATUS_STALE\x10\x03\x32\x93\x03\n\nP2pService\x12h\n\x0fPublishMetadata\x12).model_express.p2p.PublishMetadataRequest\x1a*.model_express.p2p.PublishMetadataResponse\x12\\\n\x0bListSources\x12%.model_express.p2p.ListSourcesRequest\x1a&.model_express.p2p.ListSourcesResponse\x12\\\n\x0bGetMetadata\x12%.model_express.p2p.GetMetadataRequest\x1a&.model_express.p2p.GetMetadataResponse\x12_\n\x0cUpdateStatus\x12&.model_express.p2p.UpdateStatusRequest\x1a\'.model_express.p2p.UpdateStatusResponse2\x83\x05\n\rWorkerService\x12n\n\x11GetTensorManifest\x12+.model_express.p2p.GetTensorManifestRequest\x1a,.model_express.p2p.GetTensorManifestResponse\x12\x86\x01\n\x19GetArtifactManifestHeader\x12\x33.model_express.p2p.GetArtifactManifestHeaderRequest\x1a\x34.model_express.p2p.GetArtifactManifestHeaderResponse\x12\x86\x01\n\x19GetArtifactManifestChunks\x12\x33.model_express.p2p.GetArtifactManifestChunksRequest\x1a\x34.model_express.p2p.GetArtifactManifestChunksResponse\x12w\n\x14PrepareArtifactChunk\x12..model_express.p2p.PrepareArtifactChunkRequest\x1a/.model_express.p2p.PrepareArtifactChunkResponse\x12w\n\x14ReleaseArtifactChunk\x12..model_express.p2p.ReleaseArtifactChunkRequest\x1a/.model_express.p2p.ReleaseArtifactChunkResponseb\x06proto3')
 
 _globals = globals()
 _builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
@@ -38,12 +38,12 @@
   _globals['_SOURCEIDENTITY_EXTRAPARAMETERSENTRY']._serialized_options = b'8\001'
   _globals['_WORKERMETADATA'].fields_by_name['tensors']._loaded_options = None
   _globals['_WORKERMETADATA'].fields_by_name['tensors']._serialized_options = b'\030\001'
-  _globals['_BACKENDFRAMEWORK']._serialized_start=4635
-  _globals['_BACKENDFRAMEWORK']._serialized_end=4773
-  _globals['_MXSOURCETYPE']._serialized_start=4776
-  _globals['_MXSOURCETYPE']._serialized_end=5090
-  _globals['_SOURCESTATUS']._serialized_start=5092
-  _globals['_SOURCESTATUS']._serialized_end=5215
+  _globals['_BACKENDFRAMEWORK']._serialized_start=4670
+  _globals['_BACKENDFRAMEWORK']._serialized_end=4808
+  _globals['_MXSOURCETYPE']._serialized_start=4811
+  _globals['_MXSOURCETYPE']._serialized_end=5125
+  _globals['_SOURCESTATUS']._serialized_start=5127
+  _globals['_SOURCESTATUS']._serialized_end=5250
   _globals['_SOURCEIDENTITY']._serialized_start=33
   _globals['_SOURCEIDENTITY']._serialized_end=648
   _globals['_SOURCEIDENTITY_EXTRAPARAMETERSENTRY']._serialized_start=594
@@ -85,25 +85,25 @@
   _globals['_RELEASEARTIFACTCHUNKRESPONSE']._serialized_start=3504
   _globals['_RELEASEARTIFACTCHUNKRESPONSE']._serialized_end=3634
   _globals['_PUBLISHMETADATAREQUEST']._serialized_start=3637
-  _globals['_PUBLISHMETADATAREQUEST']._serialized_end=3784
-  _globals['_PUBLISHMETADATARESPONSE']._serialized_start=3786
-  _globals['_PUBLISHMETADATARESPONSE']._serialized_end=3886
-  _globals['_SOURCEINSTANCEREF']._serialized_start=3888
-  _globals['_SOURCEINSTANCEREF']._serialized_end=4010
-  _globals['_LISTSOURCESREQUEST']._serialized_start=4013
-  _globals['_LISTSOURCESREQUEST']._serialized_end=4165
-  _globals['_LISTSOURCESRESPONSE']._serialized_start=4167
-  _globals['_LISTSOURCESRESPONSE']._serialized_end=4245
-  _globals['_GETMETADATAREQUEST']._serialized_start=4247
-  _globals['_GETMETADATAREQUEST']._serialized_end=4308
-  _globals['_GETMETADATARESPONSE']._serialized_start=4311
-  _globals['_GETMETADATARESPONSE']._serialized_end=4439
-  _globals['_UPDATESTATUSREQUEST']._serialized_start=4442
-  _globals['_UPDATESTATUSREQUEST']._serialized_end=4574
-  _globals['_UPDATESTATUSRESPONSE']._serialized_start=4576
-  _globals['_UPDATESTATUSRESPONSE']._serialized_end=4632
-  _globals['_P2PSERVICE']._serialized_start=5218
-  _globals['_P2PSERVICE']._serialized_end=5621
-  _globals['_WORKERSERVICE']._serialized_start=5624
-  _globals['_WORKERSERVICE']._serialized_end=6267
+  _globals['_PUBLISHMETADATAREQUEST']._serialized_end=3819
+  _globals['_PUBLISHMETADATARESPONSE']._serialized_start=3821
+  _globals['_PUBLISHMETADATARESPONSE']._serialized_end=3921
+  _globals['_SOURCEINSTANCEREF']._serialized_start=3923
+  _globals['_SOURCEINSTANCEREF']._serialized_end=4045
+  _globals['_LISTSOURCESREQUEST']._serialized_start=4048
+  _globals['_LISTSOURCESREQUEST']._serialized_end=4200
+  _globals['_LISTSOURCESRESPONSE']._serialized_start=4202
+  _globals['_LISTSOURCESRESPONSE']._serialized_end=4280
+  _globals['_GETMETADATAREQUEST']._serialized_start=4282
+  _globals['_GETMETADATAREQUEST']._serialized_end=4343
+  _globals['_GETMETADATARESPONSE']._serialized_start=4346
+  _globals['_GETMETADATARESPONSE']._serialized_end=4474
+  _globals['_UPDATESTATUSREQUEST']._serialized_start=4477
+  _globals['_UPDATESTATUSREQUEST']._serialized_end=4609
+  _globals['_UPDATESTATUSRESPONSE']._serialized_start=4611
+  _globals['_UPDATESTATUSRESPONSE']._serialized_end=4667
+  _globals['_P2PSERVICE']._serialized_start=5253
+  _globals['_P2PSERVICE']._serialized_end=5656
+  _globals['_WORKERSERVICE']._serialized_start=5659
+  _globals['_WORKERSERVICE']._serialized_end=6302
 # @@protoc_insertion_point(module_scope)

From 95a21c64bc534e6b95e8511e0d9043c6023f445a Mon Sep 17 00:00:00 2001
From: Omer Aplatony <omerap12@gmail.com>
Date: Tue, 14 Jul 2026 23:13:34 +0300
Subject: [PATCH 3/6] fixed ut

Signed-off-by: Omer Aplatony <omerap12@gmail.com>
---
 modelexpress_client/python/tests/test_artifact_transfer.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/modelexpress_client/python/tests/test_artifact_transfer.py b/modelexpress_client/python/tests/test_artifact_transfer.py
index 8400b582..1cd15236 100644
--- a/modelexpress_client/python/tests/test_artifact_transfer.py
+++ b/modelexpress_client/python/tests/test_artifact_transfer.py
@@ -1368,7 +1368,7 @@ def __init__(self, mx_source_id: str | None = None):
         self.published_worker_id = ""
         self.status_updates = []
 
-    def publish_metadata(self, identity, worker, worker_id):
+    def publish_metadata(self, identity, worker, worker_id, pod_name="", pod_uid=""):
         self.published_identity = p2p_pb2.SourceIdentity()
         self.published_identity.CopyFrom(identity)
         self.published_worker = p2p_pb2.WorkerMetadata()

From 7755c00aa3fa2dd53530e69b22633ada5d1b2edc Mon Sep 17 00:00:00 2001
From: Omer Aplatony <omerap12@gmail.com>
Date: Tue, 14 Jul 2026 23:17:54 +0300
Subject: [PATCH 4/6] fmt

Signed-off-by: Omer Aplatony <omerap12@gmail.com>
---
 modelexpress_client/python/modelexpress/envs.py | 4 ++--
 modelexpress_server/src/p2p/state.rs            | 8 +++++++-
 2 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/modelexpress_client/python/modelexpress/envs.py b/modelexpress_client/python/modelexpress/envs.py
index 8e3f242c..001d4bb3 100644
--- a/modelexpress_client/python/modelexpress/envs.py
+++ b/modelexpress_client/python/modelexpress/envs.py
@@ -215,8 +215,8 @@ def _env_float(name: str, default: float) -> float:
     "VLLM_ATTENTION_BACKEND": lambda: os.environ.get("VLLM_ATTENTION_BACKEND", "auto"),
     "HOSTNAME": lambda: os.environ.get("HOSTNAME", ""),
     "POD_NAMESPACE": lambda: os.environ.get("POD_NAMESPACE", ""),
-    "POD_NAME": lambda: os.environ.get("POD_NAME",""),
-    "POD_UID": lambda: os.environ.get("POD_UID","")
+    "POD_NAME": lambda: os.environ.get("POD_NAME", ""),
+    "POD_UID": lambda: os.environ.get("POD_UID", "")
 }
 
 
diff --git a/modelexpress_server/src/p2p/state.rs b/modelexpress_server/src/p2p/state.rs
index 5fddc950..9b186280 100644
--- a/modelexpress_server/src/p2p/state.rs
+++ b/modelexpress_server/src/p2p/state.rs
@@ -497,7 +497,13 @@ mod tests {
         let manager = P2pStateManager::with_backend(Arc::new(mock));
         assert!(
             manager
-                .publish_metadata(&test_identity(), "a1b2c3d4", WorkerMetadata::default(), "", "")
+                .publish_metadata(
+                    &test_identity(),
+                    "a1b2c3d4",
+                    WorkerMetadata::default(),
+                    "",
+                    ""
+                )
                 .await
                 .is_err()
         );

From ebdb78ab3485f52786c7600c3c0d470168ce7305 Mon Sep 17 00:00:00 2001
From: Zheng Luo <zheluo@nvidia.com>
Date: Wed, 15 Jul 2026 10:39:35 -0700
Subject: [PATCH 5/6] fix: update pod ownership metadata

Signed-off-by: Zheng Luo <zheluo@nvidia.com>
---
 .../go/gen/modelexpress/p2p/p2p.pb.go         | 29 +++++++-
 .../python/tests/test_artifact_transfer.py    |  4 ++
 .../src/p2p/backend/kubernetes.rs             | 67 ++++++++++++++-----
 modelexpress_server/src/p2p/service.rs        |  6 +-
 modelexpress_server/src/p2p/state.rs          |  8 ++-
 5 files changed, 91 insertions(+), 23 deletions(-)

diff --git a/modelexpress_client/go/gen/modelexpress/p2p/p2p.pb.go b/modelexpress_client/go/gen/modelexpress/p2p/p2p.pb.go
index 116d89cf..4d71e109 100644
--- a/modelexpress_client/go/gen/modelexpress/p2p/p2p.pb.go
+++ b/modelexpress_client/go/gen/modelexpress/p2p/p2p.pb.go
@@ -1937,7 +1937,14 @@ type PublishMetadataRequest struct {
 	// Unique identifier for this running instance (UUID).
 	// Differentiates multiple replicas with identical SourceIdentity.
 	// If empty, the server rejects the request.
-	WorkerId      string `protobuf:"bytes,3,opt,name=worker_id,json=workerId,proto3" json:"worker_id,omitempty"`
+	WorkerId string `protobuf:"bytes,3,opt,name=worker_id,json=workerId,proto3" json:"worker_id,omitempty"`
+	// Kubernetes pod name of the calling worker, set via the downward API.
+	// Used with pod_uid to set an ownerReference on the ModelMetadata CR
+	// so K8s garbage-collects it when the pod is deleted.
+	// When empty, no ownerReference is set.
+	PodName string `protobuf:"bytes,4,opt,name=pod_name,json=podName,proto3" json:"pod_name,omitempty"`
+	// Kubernetes pod UID of the calling worker, set via the downward API.
+	PodUid        string `protobuf:"bytes,5,opt,name=pod_uid,json=podUid,proto3" json:"pod_uid,omitempty"`
 	unknownFields protoimpl.UnknownFields
 	sizeCache     protoimpl.SizeCache
 }
@@ -1993,6 +2000,20 @@ func (x *PublishMetadataRequest) GetWorkerId() string {
 	return ""
 }
 
+func (x *PublishMetadataRequest) GetPodName() string {
+	if x != nil {
+		return x.PodName
+	}
+	return ""
+}
+
+func (x *PublishMetadataRequest) GetPodUid() string {
+	if x != nil {
+		return x.PodUid
+	}
+	return ""
+}
+
 type PublishMetadataResponse struct {
 	state   protoimpl.MessageState `protogen:"open.v1"`
 	Success bool                   `protobuf:"varint,1,opt,name=success,proto3" json:"success,omitempty"`
@@ -2680,11 +2701,13 @@ const file_p2p_proto_rawDesc = "" +
 	"mxSourceId\x12\x1f\n" +
 	"\vartifact_id\x18\x02 \x01(\tR\n" +
 	"artifactId\x12>\n" +
-	"\x05chunk\x18\x03 \x01(\v2(.model_express.p2p.ArtifactManifestChunkR\x05chunk\"\xaf\x01\n" +
+	"\x05chunk\x18\x03 \x01(\v2(.model_express.p2p.ArtifactManifestChunkR\x05chunk\"\xe3\x01\n" +
 	"\x16PublishMetadataRequest\x12=\n" +
 	"\bidentity\x18\x01 \x01(\v2!.model_express.p2p.SourceIdentityR\bidentity\x129\n" +
 	"\x06worker\x18\x02 \x01(\v2!.model_express.p2p.WorkerMetadataR\x06worker\x12\x1b\n" +
-	"\tworker_id\x18\x03 \x01(\tR\bworkerId\"\x8c\x01\n" +
+	"\tworker_id\x18\x03 \x01(\tR\bworkerId\x12\x19\n" +
+	"\bpod_name\x18\x04 \x01(\tR\apodName\x12\x17\n" +
+	"\apod_uid\x18\x05 \x01(\tR\x06podUid\"\x8c\x01\n" +
 	"\x17PublishMetadataResponse\x12\x18\n" +
 	"\asuccess\x18\x01 \x01(\bR\asuccess\x12\x18\n" +
 	"\amessage\x18\x02 \x01(\tR\amessage\x12 \n" +
diff --git a/modelexpress_client/python/tests/test_artifact_transfer.py b/modelexpress_client/python/tests/test_artifact_transfer.py
index 1cd15236..0742143a 100644
--- a/modelexpress_client/python/tests/test_artifact_transfer.py
+++ b/modelexpress_client/python/tests/test_artifact_transfer.py
@@ -1366,6 +1366,8 @@ def __init__(self, mx_source_id: str | None = None):
         self.published_identity = None
         self.published_worker = None
         self.published_worker_id = ""
+        self.published_pod_name = ""
+        self.published_pod_uid = ""
         self.status_updates = []
 
     def publish_metadata(self, identity, worker, worker_id, pod_name="", pod_uid=""):
@@ -1374,6 +1376,8 @@ def publish_metadata(self, identity, worker, worker_id, pod_name="", pod_uid="")
         self.published_worker = p2p_pb2.WorkerMetadata()
         self.published_worker.CopyFrom(worker)
         self.published_worker_id = worker_id
+        self.published_pod_name = pod_name
+        self.published_pod_uid = pod_uid
         return self.mx_source_id or compute_mx_source_id(identity)
 
     def list_sources(self, identity=None, status_filter=None):
diff --git a/modelexpress_server/src/p2p/backend/kubernetes.rs b/modelexpress_server/src/p2p/backend/kubernetes.rs
index 340cb6e0..2cbb7b0c 100644
--- a/modelexpress_server/src/p2p/backend/kubernetes.rs
+++ b/modelexpress_server/src/p2p/backend/kubernetes.rs
@@ -14,7 +14,7 @@ use crate::p2p::k8s_types::{
 };
 use async_trait::async_trait;
 use base64::{Engine, engine::general_purpose::STANDARD as BASE64};
-use k8s_openapi::api::core::v1::ConfigMap;
+use k8s_openapi::{api::core::v1::ConfigMap, apimachinery::pkg::apis::meta::v1::OwnerReference};
 use kube::{
     Client,
     api::{Api, ListParams, Patch, PatchParams, PostParams},
@@ -50,6 +50,21 @@ impl KubernetesBackend {
         Api::namespaced(self.client.clone(), &self.namespace)
     }
 
+    fn pod_owner_references(pod_name: &str, pod_uid: &str) -> Option<Vec<OwnerReference>> {
+        if pod_name.is_empty() || pod_uid.is_empty() {
+            return None;
+        }
+
+        Some(vec![OwnerReference {
+            api_version: "v1".to_string(),
+            kind: "Pod".to_string(),
+            name: pod_name.to_string(),
+            uid: pod_uid.to_string(),
+            controller: Some(false),
+            block_owner_deletion: Some(false),
+        }])
+    }
+
     /// Create or update a ConfigMap with tensor descriptors for a worker.
     /// If `owner_uid` and `owner_name` are provided, sets ownerReferences
     /// so K8s garbage-collects ConfigMaps when the parent CR is deleted.
@@ -204,7 +219,7 @@ impl MetadataBackend for KubernetesBackend {
 
         // First, ensure the CR exists
         let existing = api.get_opt(&cr_name).await?;
-        let has_pod_name_and_uid = !pod_name.is_empty() && !pod_uid.is_empty();
+        let pod_owner_references = Self::pod_owner_references(pod_name, pod_uid);
 
         if existing.is_none() {
             let new_cr = ModelMetadata {
@@ -223,20 +238,7 @@ impl MetadataBackend for KubernetesBackend {
                         );
                         labels
                     }),
-                    owner_references: if has_pod_name_and_uid {
-                        Some(vec![
-                            k8s_openapi::apimachinery::pkg::apis::meta::v1::OwnerReference {
-                                api_version: "v1".to_string(),
-                                kind: "Pod".to_string(),
-                                name: pod_name.to_string(),
-                                uid: pod_uid.to_string(),
-                                controller: Some(false),
-                                block_owner_deletion: Some(false),
-                            },
-                        ])
-                    } else {
-                        None
-                    },
+                    owner_references: pod_owner_references.clone(),
                     ..Default::default()
                 },
                 spec: ModelMetadataSpec {
@@ -264,6 +266,19 @@ impl MetadataBackend for KubernetesBackend {
 
         // Get CR UID for ownerReferences on ConfigMaps
         let cr = api.get(&cr_name).await?;
+        if let Some(owner_references) = &pod_owner_references
+            && cr.metadata.owner_references.as_ref() != Some(owner_references)
+        {
+            let owner_patch = json!({
+                "metadata": { "ownerReferences": owner_references }
+            });
+            api.patch(
+                &cr_name,
+                &PatchParams::default(),
+                &Patch::Merge(&owner_patch),
+            )
+            .await?;
+        }
         let owner_uid = cr.metadata.uid.as_deref();
         let owner_name = cr.metadata.name.as_deref();
 
@@ -828,4 +843,24 @@ mod tests {
 
         assert!(ArtifactSourceMetadataRecord::try_from(status).is_err());
     }
+
+    #[test]
+    fn pod_owner_references_require_name_and_uid() {
+        assert!(KubernetesBackend::pod_owner_references("pod-0", "").is_none());
+        assert!(KubernetesBackend::pod_owner_references("", "pod-uid").is_none());
+    }
+
+    #[test]
+    fn pod_owner_references_identify_the_publishing_pod() {
+        let Some(owner_references) = KubernetesBackend::pod_owner_references("pod-0", "pod-uid")
+        else {
+            panic!("pod name and UID should produce an owner reference");
+        };
+
+        assert_eq!(owner_references.len(), 1);
+        assert_eq!(owner_references[0].api_version, "v1");
+        assert_eq!(owner_references[0].kind, "Pod");
+        assert_eq!(owner_references[0].name, "pod-0");
+        assert_eq!(owner_references[0].uid, "pod-uid");
+    }
 }
diff --git a/modelexpress_server/src/p2p/service.rs b/modelexpress_server/src/p2p/service.rs
index 32d4eb24..f7c64e7b 100644
--- a/modelexpress_server/src/p2p/service.rs
+++ b/modelexpress_server/src/p2p/service.rs
@@ -433,6 +433,9 @@ mod tests {
     async fn test_publish_metadata_success() {
         let mut mock = MockMetadataBackend::new();
         mock.expect_publish_metadata()
+            .withf(|_, _, _, pod_name, pod_uid| {
+                pod_name == "vllm-worker-0" && pod_uid == "pod-uid-1"
+            })
             .once()
             .returning(|_, _, _, _, _| Ok(()));
 
@@ -451,7 +454,8 @@ mod tests {
                     ..Default::default()
                 }),
                 worker_id: "worker-uuid-1".to_string(),
-                ..Default::default()
+                pod_name: "vllm-worker-0".to_string(),
+                pod_uid: "pod-uid-1".to_string(),
             }))
             .await
             .expect("rpc")
diff --git a/modelexpress_server/src/p2p/state.rs b/modelexpress_server/src/p2p/state.rs
index 9b186280..a2880b9a 100644
--- a/modelexpress_server/src/p2p/state.rs
+++ b/modelexpress_server/src/p2p/state.rs
@@ -459,11 +459,13 @@ mod tests {
     async fn test_publish_metadata_calls_backend() {
         let mut mock = MockMetadataBackend::new();
         mock.expect_publish_metadata()
-            .withf(|identity, worker_id, worker, _, _| {
+            .withf(|identity, worker_id, worker, pod_name, pod_uid| {
                 identity.model_name == "my-model"
                     && identity.tensor_parallel_size == 8
                     && worker_id == "a1b2c3d4"
                     && worker.worker_rank == 3
+                    && pod_name == "vllm-worker-0"
+                    && pod_uid == "pod-uid-1"
             })
             .once()
             .returning(|_, _, _, _, _| Ok(()));
@@ -480,8 +482,8 @@ mod tests {
                     updated_at: 0,
                     ..Default::default()
                 },
-                "",
-                "",
+                "vllm-worker-0",
+                "pod-uid-1",
             )
             .await
             .expect("publish_metadata failed");

From 3a376cc9daf5a996b543023094b31063919f4a19 Mon Sep 17 00:00:00 2001
From: Zheng Luo <zheluo@nvidia.com>
Date: Wed, 15 Jul 2026 13:32:19 -0700
Subject: [PATCH 6/6] fix: make metadata pod ownership compatible

Signed-off-by: Zheng Luo <zheluo@nvidia.com>
---
 .../client/sgl/manifest-azure-mooncake.yaml   | 12 ++++
 ci/k8s/client/sgl/manifest-azure.yaml         | 12 ++++
 .../dynamo/manifest-azure-aggregated.yaml     | 12 ++++
 .../dynamo/manifest-azure-disaggregated.yaml  | 24 ++++++++
 ci/k8s/client/vllm/manifest-azure.yaml        | 12 ++++
 docs/DEPLOYMENT.md                            | 27 +++++++++
 docs/metadata.md                              | 27 ++++++++-
 examples/p2p_transfer_k8s/client/README.md    | 10 ++++
 .../client/sglang/sglang-single-node-p2p.yaml | 12 ++++
 .../sglang-single-node-transfer-engine.yaml   | 12 ++++
 .../client/vllm/vllm-multi-node.yaml          |  8 +++
 .../client/vllm/vllm-single-node.yaml         |  8 +++
 .../go/gen/modelexpress/p2p/p2p.pb.go         | 23 ++++++--
 .../python/modelexpress/client.py             |  9 +--
 .../metadata/k8s_service_client.py            |  2 -
 .../python/modelexpress/metadata/publish.py   | 14 +----
 .../python/modelexpress/p2p_pb2.py            | 56 +++++++++----------
 .../python/tests/test_artifact_transfer.py    |  6 +-
 .../python/tests/test_client.py               | 33 +++++++++++
 .../python/tests/test_types.py                | 49 ++++++++++++++++
 modelexpress_common/proto/p2p.proto           | 10 +++-
 modelexpress_server/src/p2p/backend.rs        |  1 +
 .../src/p2p/backend/kubernetes.rs             | 32 ++++++++---
 modelexpress_server/src/p2p/backend/memory.rs | 19 ++++---
 modelexpress_server/src/p2p/backend/redis.rs  |  1 +
 modelexpress_server/src/p2p/service.rs        | 18 ++++--
 modelexpress_server/src/p2p/state.rs          | 37 ++++++++----
 27 files changed, 388 insertions(+), 98 deletions(-)
 create mode 100644 modelexpress_client/python/tests/test_client.py

diff --git a/ci/k8s/client/sgl/manifest-azure-mooncake.yaml b/ci/k8s/client/sgl/manifest-azure-mooncake.yaml
index e17f5e08..07daff2e 100644
--- a/ci/k8s/client/sgl/manifest-azure-mooncake.yaml
+++ b/ci/k8s/client/sgl/manifest-azure-mooncake.yaml
@@ -98,6 +98,18 @@ spec:
             - --modelexpress-config
             - '{"transport": "transfer_engine"}'
           env:
+            - name: POD_NAME
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.name
+            - name: POD_UID
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.uid
+            - name: POD_NAMESPACE
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.namespace
             - name: MX_SERVER_ADDRESS
               value: "mx-server:8000"
             - name: MX_RDMA_NIC_PIN
diff --git a/ci/k8s/client/sgl/manifest-azure.yaml b/ci/k8s/client/sgl/manifest-azure.yaml
index f2862abd..19c686fa 100644
--- a/ci/k8s/client/sgl/manifest-azure.yaml
+++ b/ci/k8s/client/sgl/manifest-azure.yaml
@@ -105,6 +105,18 @@ spec:
             - --modelexpress-config
             - '{"transport": "nixl"}'
           env:
+            - name: POD_NAME
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.name
+            - name: POD_UID
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.uid
+            - name: POD_NAMESPACE
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.namespace
             - name: MX_SERVER_ADDRESS
               value: "mx-server:8000"
             - name: MX_RDMA_NIC_PIN
diff --git a/ci/k8s/client/vllm/dynamo/manifest-azure-aggregated.yaml b/ci/k8s/client/vllm/dynamo/manifest-azure-aggregated.yaml
index 7a0581ef..57f19192 100644
--- a/ci/k8s/client/vllm/dynamo/manifest-azure-aggregated.yaml
+++ b/ci/k8s/client/vllm/dynamo/manifest-azure-aggregated.yaml
@@ -153,6 +153,18 @@ spec:
           custom:
             rdma/ib: "1"
       envs:
+        - name: POD_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.name
+        - name: POD_UID
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.uid
+        - name: POD_NAMESPACE
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.namespace
         - name: VLLM_PLUGINS
           value: "modelexpress"
         - name: MX_SERVER_ADDRESS
diff --git a/ci/k8s/client/vllm/dynamo/manifest-azure-disaggregated.yaml b/ci/k8s/client/vllm/dynamo/manifest-azure-disaggregated.yaml
index 086617dc..348f9a7d 100644
--- a/ci/k8s/client/vllm/dynamo/manifest-azure-disaggregated.yaml
+++ b/ci/k8s/client/vllm/dynamo/manifest-azure-disaggregated.yaml
@@ -126,6 +126,18 @@ spec:
           custom:
             rdma/ib: "1"
       envs:
+        - name: POD_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.name
+        - name: POD_UID
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.uid
+        - name: POD_NAMESPACE
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.namespace
         - name: VLLM_PLUGINS
           value: "modelexpress"
         - name: MX_SERVER_ADDRESS
@@ -206,6 +218,18 @@ spec:
           custom:
             rdma/ib: "1"
       envs:
+        - name: POD_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.name
+        - name: POD_UID
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.uid
+        - name: POD_NAMESPACE
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.namespace
         - name: VLLM_PLUGINS
           value: "modelexpress"
         - name: MX_SERVER_ADDRESS
diff --git a/ci/k8s/client/vllm/manifest-azure.yaml b/ci/k8s/client/vllm/manifest-azure.yaml
index ea1e990e..6bbb6c9a 100644
--- a/ci/k8s/client/vllm/manifest-azure.yaml
+++ b/ci/k8s/client/vllm/manifest-azure.yaml
@@ -107,6 +107,18 @@ spec:
             - --port
             - "${WORKER_PORT}"
           env:
+            - name: POD_NAME
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.name
+            - name: POD_UID
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.uid
+            - name: POD_NAMESPACE
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.namespace
             - name: VLLM_PLUGINS
               value: "modelexpress"
             - name: MX_SERVER_ADDRESS
diff --git a/docs/DEPLOYMENT.md b/docs/DEPLOYMENT.md
index 8178f5dd..69e89701 100644
--- a/docs/DEPLOYMENT.md
+++ b/docs/DEPLOYMENT.md
@@ -105,6 +105,33 @@ reachable.
 
 To use the Kubernetes backend, apply `examples/crds.yaml` at cluster install time (installs both the `ModelMetadata` P2P CRD and the `ModelCacheEntry` registry CRD), then either enable `serviceAccount.rbac.enabled=true` on the Helm chart or apply `examples/p2p_transfer_k8s/server/kubernetes_backend/rbac-modelmetadata.yaml`.
 
+For automatic cleanup of P2P metadata, expose the client Pod identity through
+the Kubernetes Downward API. The checked-in vLLM, SGLang, and Dynamo manifests
+already include these fields:
+
+```yaml
+env:
+  - name: POD_NAME
+    valueFrom:
+      fieldRef:
+        fieldPath: metadata.name
+  - name: POD_UID
+    valueFrom:
+      fieldRef:
+        fieldPath: metadata.uid
+  - name: POD_NAMESPACE
+    valueFrom:
+      fieldRef:
+        fieldPath: metadata.namespace
+```
+
+With the Kubernetes metadata backend, a complete same-namespace identity makes
+the Pod the owner of every weight or artifact `ModelMetadata` CR it publishes.
+Kubernetes then garbage-collects those records when the Pod is deleted. Missing,
+partial, or cross-namespace identity does not block publication; ModelExpress
+omits the owner reference and retains the previous reaper-based lifecycle. Redis
+and non-Kubernetes deployments ignore these fields.
+
 #### Storage access modes
 
 MX has one configurable filesystem path, the model weights cache (`MODEL_EXPRESS_CACHE_DIRECTORY`, default `./cache`). Its access-mode requirement depends on deployment topology, not on MX itself:
diff --git a/docs/metadata.md b/docs/metadata.md
index bb6a3411..f28ef4bf 100644
--- a/docs/metadata.md
+++ b/docs/metadata.md
@@ -335,7 +335,18 @@ Uses `ModelMetadata` CRDs for P2P source metadata, `ConfigMap`s for tensor descr
 
 **ConfigMap name format**: `mx-source-{source_id}-{worker_id}-tensors-worker-{rank}`
 
-ConfigMaps use `ownerReferences` pointing to the parent CRD so they are garbage-collected automatically.
+When a client publishes a complete Kubernetes Pod identity (`POD_NAME`,
+`POD_UID`, and `POD_NAMESPACE`) and the Pod is in the metadata namespace, the
+`ModelMetadata` CR uses that Pod as an owner. This applies to both weight and
+artifact metadata because both use the same `PublishMetadata` RPC. Deleting the
+Pod therefore garbage-collects its `ModelMetadata` CR. Tensor `ConfigMap`s use a
+second owner reference pointing to the parent CR, so they are collected with it.
+
+Kubernetes owner references cannot cross namespaces. If the identity is
+missing, partial, or names a different namespace, publication still succeeds
+without a Pod owner reference. This preserves behavior for older clients and
+non-Kubernetes environments; the server-side stale metadata reaper remains the
+cleanup path in those cases.
 
 **Model lifecycle CRD name format**: `mx-cache-{sanitized-model-name}-{hash}`
 
@@ -353,6 +364,13 @@ apiVersion: modelexpress.nvidia.com/v1alpha1
 kind: ModelMetadata
 metadata:
   name: mx-source-a1b2c3d4e5f67890-f3a2b1c4
+  ownerReferences:
+    - apiVersion: v1
+      kind: Pod
+      name: mx-vllm-7d9f8f6c8b-k2m4p
+      uid: 8c69d55f-3e40-4b6e-a16b-6f0c1168e171
+      controller: false
+      blockOwnerDeletion: false
   labels:
     modelexpress.nvidia.com/mx-source-id: a1b2c3d4e5f67890
     modelexpress.nvidia.com/mx-worker-id: f3a2b1c4
@@ -385,6 +403,13 @@ apiVersion: modelexpress.nvidia.com/v1alpha1
 kind: ModelMetadata
 metadata:
   name: mx-source-b2c3d4e5f67890a1-f3a2b1c4
+  ownerReferences:
+    - apiVersion: v1
+      kind: Pod
+      name: mx-vllm-7d9f8f6c8b-k2m4p
+      uid: 8c69d55f-3e40-4b6e-a16b-6f0c1168e171
+      controller: false
+      blockOwnerDeletion: false
 spec:
   modelName: artifact-transfer-e2e
   sourceType: deep_gemm_cache
diff --git a/examples/p2p_transfer_k8s/client/README.md b/examples/p2p_transfer_k8s/client/README.md
index 3f975ed8..a008a990 100644
--- a/examples/p2p_transfer_k8s/client/README.md
+++ b/examples/p2p_transfer_k8s/client/README.md
@@ -41,6 +41,16 @@ On startup, the engine-specific ModelExpress loader auto-detects the best loadin
 
 After loading, every worker publishes its metadata so future instances can discover it as an RDMA source.
 
+When the server uses the Kubernetes metadata backend, the example manifests
+also inject `POD_NAME`, `POD_UID`, and `POD_NAMESPACE` through the Downward API.
+ModelExpress uses a complete same-namespace identity to make the Pod the owner
+of each `ModelMetadata` CR it publishes, including artifact metadata. This
+ownership behavior is independent of the inference engine.
+Deleting the Pod then removes its metadata through Kubernetes garbage
+collection. If any identity field is unavailable, publication continues without
+an owner reference so older clients and non-Kubernetes deployments remain
+compatible.
+
 ## Prerequisites
 
 - ModelExpress server deployed (see [`../server/`](../server/))
diff --git a/examples/p2p_transfer_k8s/client/sglang/sglang-single-node-p2p.yaml b/examples/p2p_transfer_k8s/client/sglang/sglang-single-node-p2p.yaml
index 6a2c5d0c..b34c6ad0 100644
--- a/examples/p2p_transfer_k8s/client/sglang/sglang-single-node-p2p.yaml
+++ b/examples/p2p_transfer_k8s/client/sglang/sglang-single-node-p2p.yaml
@@ -85,6 +85,18 @@ spec:
               port: 8000
             timeoutSeconds: 10
           env:
+            - name: POD_NAME
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.name
+            - name: POD_UID
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.uid
+            - name: POD_NAMESPACE
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.namespace
             - name: HF_HOME
               value: "/models"
             - name: MODEL_NAME
diff --git a/examples/p2p_transfer_k8s/client/sglang/sglang-single-node-transfer-engine.yaml b/examples/p2p_transfer_k8s/client/sglang/sglang-single-node-transfer-engine.yaml
index 37af3d3c..3ef04079 100644
--- a/examples/p2p_transfer_k8s/client/sglang/sglang-single-node-transfer-engine.yaml
+++ b/examples/p2p_transfer_k8s/client/sglang/sglang-single-node-transfer-engine.yaml
@@ -88,6 +88,18 @@ spec:
               port: 8000
             timeoutSeconds: 10
           env:
+            - name: POD_NAME
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.name
+            - name: POD_UID
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.uid
+            - name: POD_NAMESPACE
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.namespace
             - name: HF_HOME
               value: "/models"
             - name: MODEL_NAME
diff --git a/examples/p2p_transfer_k8s/client/vllm/vllm-multi-node.yaml b/examples/p2p_transfer_k8s/client/vllm/vllm-multi-node.yaml
index 01d96e47..b3d52c97 100644
--- a/examples/p2p_transfer_k8s/client/vllm/vllm-multi-node.yaml
+++ b/examples/p2p_transfer_k8s/client/vllm/vllm-multi-node.yaml
@@ -121,6 +121,14 @@ spec:
               valueFrom:
                 fieldRef:
                   fieldPath: metadata.namespace
+            - name: POD_NAME
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.name
+            - name: POD_UID
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.uid
             - name: HF_TOKEN
               valueFrom:
                 secretKeyRef:
diff --git a/examples/p2p_transfer_k8s/client/vllm/vllm-single-node.yaml b/examples/p2p_transfer_k8s/client/vllm/vllm-single-node.yaml
index 2f8bdb56..c771973d 100644
--- a/examples/p2p_transfer_k8s/client/vllm/vllm-single-node.yaml
+++ b/examples/p2p_transfer_k8s/client/vllm/vllm-single-node.yaml
@@ -95,6 +95,14 @@ spec:
               valueFrom:
                 fieldRef:
                   fieldPath: metadata.namespace
+            - name: POD_NAME
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.name
+            - name: POD_UID
+              valueFrom:
+                fieldRef:
+                  fieldPath: metadata.uid
             - name: HF_TOKEN
               valueFrom:
                 secretKeyRef:
diff --git a/modelexpress_client/go/gen/modelexpress/p2p/p2p.pb.go b/modelexpress_client/go/gen/modelexpress/p2p/p2p.pb.go
index 4d71e109..b319d0b4 100644
--- a/modelexpress_client/go/gen/modelexpress/p2p/p2p.pb.go
+++ b/modelexpress_client/go/gen/modelexpress/p2p/p2p.pb.go
@@ -1939,12 +1939,15 @@ type PublishMetadataRequest struct {
 	// If empty, the server rejects the request.
 	WorkerId string `protobuf:"bytes,3,opt,name=worker_id,json=workerId,proto3" json:"worker_id,omitempty"`
 	// Kubernetes pod name of the calling worker, set via the downward API.
-	// Used with pod_uid to set an ownerReference on the ModelMetadata CR
-	// so K8s garbage-collects it when the pod is deleted.
-	// When empty, no ownerReference is set.
+	// Used with pod_uid and pod_namespace to set an ownerReference on the
+	// ModelMetadata CR so K8s garbage-collects it when the pod is deleted.
 	PodName string `protobuf:"bytes,4,opt,name=pod_name,json=podName,proto3" json:"pod_name,omitempty"`
 	// Kubernetes pod UID of the calling worker, set via the downward API.
-	PodUid        string `protobuf:"bytes,5,opt,name=pod_uid,json=podUid,proto3" json:"pod_uid,omitempty"`
+	PodUid string `protobuf:"bytes,5,opt,name=pod_uid,json=podUid,proto3" json:"pod_uid,omitempty"`
+	// Kubernetes namespace of the calling worker, set via the downward API.
+	// Owner references are only valid when this matches the ModelMetadata
+	// namespace. Empty means no ownerReference is set.
+	PodNamespace  string `protobuf:"bytes,6,opt,name=pod_namespace,json=podNamespace,proto3" json:"pod_namespace,omitempty"`
 	unknownFields protoimpl.UnknownFields
 	sizeCache     protoimpl.SizeCache
 }
@@ -2014,6 +2017,13 @@ func (x *PublishMetadataRequest) GetPodUid() string {
 	return ""
 }
 
+func (x *PublishMetadataRequest) GetPodNamespace() string {
+	if x != nil {
+		return x.PodNamespace
+	}
+	return ""
+}
+
 type PublishMetadataResponse struct {
 	state   protoimpl.MessageState `protogen:"open.v1"`
 	Success bool                   `protobuf:"varint,1,opt,name=success,proto3" json:"success,omitempty"`
@@ -2701,13 +2711,14 @@ const file_p2p_proto_rawDesc = "" +
 	"mxSourceId\x12\x1f\n" +
 	"\vartifact_id\x18\x02 \x01(\tR\n" +
 	"artifactId\x12>\n" +
-	"\x05chunk\x18\x03 \x01(\v2(.model_express.p2p.ArtifactManifestChunkR\x05chunk\"\xe3\x01\n" +
+	"\x05chunk\x18\x03 \x01(\v2(.model_express.p2p.ArtifactManifestChunkR\x05chunk\"\x88\x02\n" +
 	"\x16PublishMetadataRequest\x12=\n" +
 	"\bidentity\x18\x01 \x01(\v2!.model_express.p2p.SourceIdentityR\bidentity\x129\n" +
 	"\x06worker\x18\x02 \x01(\v2!.model_express.p2p.WorkerMetadataR\x06worker\x12\x1b\n" +
 	"\tworker_id\x18\x03 \x01(\tR\bworkerId\x12\x19\n" +
 	"\bpod_name\x18\x04 \x01(\tR\apodName\x12\x17\n" +
-	"\apod_uid\x18\x05 \x01(\tR\x06podUid\"\x8c\x01\n" +
+	"\apod_uid\x18\x05 \x01(\tR\x06podUid\x12#\n" +
+	"\rpod_namespace\x18\x06 \x01(\tR\fpodNamespace\"\x8c\x01\n" +
 	"\x17PublishMetadataResponse\x12\x18\n" +
 	"\asuccess\x18\x01 \x01(\bR\asuccess\x12\x18\n" +
 	"\amessage\x18\x02 \x01(\tR\amessage\x12 \n" +
diff --git a/modelexpress_client/python/modelexpress/client.py b/modelexpress_client/python/modelexpress/client.py
index 0cbb5c8b..77eb1ba3 100644
--- a/modelexpress_client/python/modelexpress/client.py
+++ b/modelexpress_client/python/modelexpress/client.py
@@ -46,8 +46,6 @@ def publish_metadata(
         identity: "p2p_pb2.SourceIdentity",
         worker: "p2p_pb2.WorkerMetadata",
         worker_id: str,
-        pod_name: str = "",
-        pod_uid: str = "",
     ) -> str:
         """Publish worker metadata and return the computed mx_source_id."""
 
@@ -169,8 +167,6 @@ def publish_metadata(
         identity: "p2p_pb2.SourceIdentity",
         worker: "p2p_pb2.WorkerMetadata",
         worker_id: str,
-        pod_name: str = "",
-        pod_uid: str = "",
     ) -> str:
         """Publish metadata for one worker so targets can discover this source.
 
@@ -180,8 +176,9 @@ def publish_metadata(
             identity=identity,
             worker=worker,
             worker_id=worker_id,
-            pod_name=pod_name,
-            pod_uid=pod_uid,
+            pod_name=envs.POD_NAME,
+            pod_uid=envs.POD_UID,
+            pod_namespace=envs.POD_NAMESPACE,
         )
         response = self.stub.PublishMetadata(request, timeout=30)
         if not response.success:
diff --git a/modelexpress_client/python/modelexpress/metadata/k8s_service_client.py b/modelexpress_client/python/modelexpress/metadata/k8s_service_client.py
index d4259997..886321f5 100644
--- a/modelexpress_client/python/modelexpress/metadata/k8s_service_client.py
+++ b/modelexpress_client/python/modelexpress/metadata/k8s_service_client.py
@@ -104,8 +104,6 @@ def publish_metadata(
         identity: "p2p_pb2.SourceIdentity",
         worker: "p2p_pb2.WorkerMetadata",
         worker_id: str,
-        pod_name: str = "",
-        pod_uid: str = "",
     ) -> str:
         """Compute mx_source_id locally - there is no central store to hit.
 
diff --git a/modelexpress_client/python/modelexpress/metadata/publish.py b/modelexpress_client/python/modelexpress/metadata/publish.py
index c999a024..52c7a476 100644
--- a/modelexpress_client/python/modelexpress/metadata/publish.py
+++ b/modelexpress_client/python/modelexpress/metadata/publish.py
@@ -127,9 +127,6 @@ def publish_metadata_and_ready(
         f"[Worker {worker_rank}] Preparing {len(tensors)} tensors for model '{identity.model_name}'"
     )
 
-    pod_name = envs.POD_NAME
-    pod_uid = envs.POD_UID
-
     tensor_protos = build_tensor_protos(tensors, device_id, worker_rank)
 
     if _is_p2p_metadata_enabled(mx_client):
@@ -174,8 +171,6 @@ def publish_fn() -> str:
                 worker=worker,
                 worker_id=worker_id,
                 worker_rank=worker_rank,
-                pod_name=pod_name,
-                pod_uid=pod_uid,
             )
             grpc_server.set_mx_source_id(mx_source_id)
             logger.info(
@@ -209,8 +204,6 @@ def publish_fn() -> str:
                 worker=worker,
                 worker_id=worker_id,
                 worker_rank=worker_rank,
-                pod_name=pod_name,
-                pod_uid=pod_uid,
             )
             logger.info(
                 f"[Worker {worker_rank}] Published metadata to MX server "
@@ -238,18 +231,13 @@ def _publish_metadata_to_server(
     worker: "p2p_pb2.WorkerMetadata",
     worker_id: str,
     worker_rank: int,
-    pod_name: str = "",
-    pod_uid: str = "",
 ) -> str:
     """Publish metadata with bounded retries and exponential backoff."""
     last_error: grpc.RpcError | None = None
 
     for attempt in range(1, PUBLISH_METADATA_MAX_ATTEMPTS + 1):
         try:
-            return mx_client.publish_metadata(
-                identity, worker, worker_id,
-                pod_name=pod_name, pod_uid=pod_uid,
-            )
+            return mx_client.publish_metadata(identity, worker, worker_id)
         except grpc.RpcError as exc:
             if exc.code() not in PUBLISH_METADATA_RETRYABLE_STATUS_CODES:
                 raise
diff --git a/modelexpress_client/python/modelexpress/p2p_pb2.py b/modelexpress_client/python/modelexpress/p2p_pb2.py
index 654da8cc..146098fe 100644
--- a/modelexpress_client/python/modelexpress/p2p_pb2.py
+++ b/modelexpress_client/python/modelexpress/p2p_pb2.py
@@ -27,7 +27,7 @@
 
 
 
-DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\tp2p.proto\x12\x11model_express.p2p\"\xe7\x04\n\x0eSourceIdentity\x12\x12\n\nmx_version\x18\x01 \x01(\t\x12\x37\n\x0emx_source_type\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\nmodel_name\x18\x03 \x01(\t\x12>\n\x11\x62\x61\x63kend_framework\x18\x04 \x01(\x0e\x32#.model_express.p2p.BackendFramework\x12\x1c\n\x14tensor_parallel_size\x18\x05 \x01(\r\x12\x1e\n\x16pipeline_parallel_size\x18\x06 \x01(\r\x12\x1c\n\x14\x65xpert_parallel_size\x18\x07 \x01(\r\x12\r\n\x05\x64type\x18\x08 \x01(\t\x12\x14\n\x0cquantization\x18\t \x01(\t\x12P\n\x10\x65xtra_parameters\x18\n \x03(\x0b\x32\x36.model_express.p2p.SourceIdentity.ExtraParametersEntry\x12\x10\n\x08revision\x18\x0b \x01(\t\x12!\n\x19\x62\x61\x63kend_framework_version\x18\x0c \x01(\t\x12\x15\n\rtorch_version\x18\r \x01(\t\x12\x14\n\x0c\x63uda_version\x18\x0e \x01(\t\x12\x16\n\x0etriton_version\x18\x0f \x01(\t\x12\x10\n\x08gpu_arch\x18\x10 \x01(\t\x12\x1d\n\x15\x63ompile_config_digest\x18\x11 \x01(\t\x1a\x36\n\x14\x45xtraParametersEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\t:\x02\x38\x01\"^\n\x10TensorDescriptor\x12\x0c\n\x04name\x18\x01 \x01(\t\x12\x0c\n\x04\x61\x64\x64r\x18\x02 \x01(\x04\x12\x0c\n\x04size\x18\x03 \x01(\x04\x12\x11\n\tdevice_id\x18\x04 \x01(\r\x12\r\n\x05\x64type\x18\x05 \x01(\t\"}\n\x16\x41rtifactSourceMetadata\x12\x13\n\x0b\x61rtifact_id\x18\x01 \x01(\t\x12\x12\n\ntotal_size\x18\x02 \x01(\x04\x12\x12\n\nfile_count\x18\x03 \x01(\r\x12\x13\n\x0b\x63hunk_count\x18\x04 \x01(\r\x12\x11\n\tnode_rank\x18\x05 \x01(\r\"L\n\x14TensorSourceMetadata\x12\x34\n\x07tensors\x18\x01 \x03(\x0b\x32#.model_express.p2p.TensorDescriptor\"\xeb\x01\n\x10\x41rtifactManifest\x12\x18\n\x10manifest_version\x18\x01 \x01(\r\x12\x37\n\x0emx_source_type\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\nchunk_size\x18\x03 \x01(\x04\x12\x36\n\x05\x66iles\x18\x04 \x03(\x0b\x32\'.model_express.p2p.ArtifactManifestFile\x12\x38\n\x06\x63hunks\x18\x05 \x03(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\"X\n\x14\x41rtifactManifestFile\x12\x12\n\nfile_index\x18\x01 \x01(\r\x12\x0c\n\x04path\x18\x02 \x01(\t\x12\x0c\n\x04size\x18\x03 \x01(\x04\x12\x10\n\x08\x63hecksum\x18\x04 \x01(\t\"w\n\x15\x41rtifactManifestChunk\x12\x13\n\x0b\x63hunk_index\x18\x01 \x01(\r\x12\x12\n\nfile_index\x18\x02 \x01(\r\x12\x13\n\x0b\x66ile_offset\x18\x03 \x01(\x04\x12\x0e\n\x06length\x18\x04 \x01(\x04\x12\x10\n\x08\x63hecksum\x18\x05 \x01(\t\"\xf3\x03\n\x0eWorkerMetadata\x12\x13\n\x0bworker_rank\x18\x01 \x01(\r\x12\x17\n\rnixl_metadata\x18\x02 \x01(\x0cH\x00\x12$\n\x1atransfer_engine_session_id\x18\n \x01(\tH\x00\x12\x38\n\x07tensors\x18\x03 \x03(\x0b\x32#.model_express.p2p.TensorDescriptorB\x02\x18\x01\x12/\n\x06status\x18\x04 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatus\x12\x12\n\nupdated_at\x18\x05 \x01(\x03\x12\x19\n\x11metadata_endpoint\x18\x06 \x01(\t\x12\x12\n\nagent_name\x18\x07 \x01(\t\x12\x1c\n\x14worker_grpc_endpoint\x18\x08 \x01(\t\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\t \x01(\t\x12@\n\rtensor_source\x18\x14 \x01(\x0b\x32\'.model_express.p2p.TensorSourceMetadataH\x01\x12\x44\n\x0f\x61rtifact_source\x18\x15 \x01(\x0b\x32).model_express.p2p.ArtifactSourceMetadataH\x01\x42\x12\n\x10\x62\x61\x63kend_metadataB\x10\n\x0esource_payload\"V\n\x18GetTensorManifestRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x16\n\tworker_id\x18\x02 \x01(\tH\x00\x88\x01\x01\x42\x0c\n\n_worker_id\"\xe6\x01\n\x19GetTensorManifestResponse\x12\x34\n\x07tensors\x18\x01 \x03(\x0b\x32#.model_express.p2p.TensorDescriptor\x12\x14\n\x0cmx_source_id\x18\x02 \x01(\t\x12\x19\n\x11metadata_endpoint\x18\x03 \x01(\t\x12\x12\n\nagent_name\x18\x04 \x01(\t\x12\x13\n\x0bworker_rank\x18\x05 \x01(\r\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\x06 \x01(\t\x12\x16\n\tworker_id\x18\x07 \x01(\tH\x00\x88\x01\x01\x42\x0c\n\n_worker_id\"M\n GetArtifactManifestHeaderRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\"\xee\x02\n!GetArtifactManifestHeaderResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x18\n\x10manifest_version\x18\x03 \x01(\r\x12\x37\n\x0emx_source_type\x18\x04 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\ntotal_size\x18\x05 \x01(\x04\x12\x12\n\nfile_count\x18\x06 \x01(\r\x12\x13\n\x0b\x63hunk_count\x18\x07 \x01(\r\x12\x12\n\nchunk_size\x18\x08 \x01(\x04\x12\x19\n\x11metadata_endpoint\x18\t \x01(\t\x12\x12\n\nagent_name\x18\n \x01(\t\x12\x13\n\x0bworker_rank\x18\x0b \x01(\r\x12\x36\n\x05\x66iles\x18\x0c \x03(\x0b\x32\'.model_express.p2p.ArtifactManifestFile\"|\n GetArtifactManifestChunksRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x19\n\x11start_chunk_index\x18\x03 \x01(\r\x12\x12\n\nmax_chunks\x18\x04 \x01(\r\"\xbc\x01\n!GetArtifactManifestChunksResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x19\n\x11start_chunk_index\x18\x03 \x01(\r\x12\x38\n\x06\x63hunks\x18\x04 \x03(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\x12\x17\n\x0fnext_page_token\x18\x05 \x01(\t\"R\n\x1f\x41rtifactChunkTransferDescriptor\x12\x0c\n\x04\x61\x64\x64r\x18\x01 \x01(\x04\x12\x0e\n\x06length\x18\x02 \x01(\x04\x12\x11\n\tdevice_id\x18\x03 \x01(\r\"]\n\x1bPrepareArtifactChunkRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x13\n\x0b\x63hunk_index\x18\x03 \x01(\r\"\xf1\x01\n\x1cPrepareArtifactChunkResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x10\n\x08lease_id\x18\x03 \x01(\t\x12\x37\n\x05\x63hunk\x18\x04 \x01(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\x12\x42\n\x06source\x18\x05 \x01(\x0b\x32\x32.model_express.p2p.ArtifactChunkTransferDescriptor\x12\x17\n\x0fsource_metadata\x18\x06 \x01(\x0c\"Z\n\x1bReleaseArtifactChunkRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x10\n\x08lease_id\x18\x03 \x01(\t\"\x82\x01\n\x1cReleaseArtifactChunkResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x37\n\x05\x63hunk\x18\x03 \x01(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\"\xb6\x01\n\x16PublishMetadataRequest\x12\x33\n\x08identity\x18\x01 \x01(\x0b\x32!.model_express.p2p.SourceIdentity\x12\x31\n\x06worker\x18\x02 \x01(\x0b\x32!.model_express.p2p.WorkerMetadata\x12\x11\n\tworker_id\x18\x03 \x01(\t\x12\x10\n\x08pod_name\x18\x04 \x01(\t\x12\x0f\n\x07pod_uid\x18\x05 \x01(\t\"d\n\x17PublishMetadataResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\x12\x14\n\x0cmx_source_id\x18\x03 \x01(\t\x12\x11\n\tworker_id\x18\x04 \x01(\t\"z\n\x11SourceInstanceRef\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x11\n\tworker_id\x18\x02 \x01(\t\x12\x12\n\nmodel_name\x18\x03 \x01(\t\x12\x13\n\x0bworker_rank\x18\x04 \x01(\r\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\x05 \x01(\t\"\x98\x01\n\x12ListSourcesRequest\x12\x33\n\x08identity\x18\x01 \x01(\x0b\x32!.model_express.p2p.SourceIdentity\x12;\n\rstatus_filter\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatusH\x00\x88\x01\x01\x42\x10\n\x0e_status_filter\"N\n\x13ListSourcesResponse\x12\x37\n\tinstances\x18\x01 \x03(\x0b\x32$.model_express.p2p.SourceInstanceRef\"=\n\x12GetMetadataRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x11\n\tworker_id\x18\x02 \x01(\t\"\x80\x01\n\x13GetMetadataResponse\x12\r\n\x05\x66ound\x18\x01 \x01(\x08\x12\x31\n\x06worker\x18\x02 \x01(\x0b\x32!.model_express.p2p.WorkerMetadata\x12\x14\n\x0cmx_source_id\x18\x03 \x01(\t\x12\x11\n\tworker_id\x18\x04 \x01(\t\"\x84\x01\n\x13UpdateStatusRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0bworker_rank\x18\x02 \x01(\r\x12/\n\x06status\x18\x03 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatus\x12\x11\n\tworker_id\x18\x04 \x01(\t\"8\n\x14UpdateStatusResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t*\x8a\x01\n\x10\x42\x61\x63kendFramework\x12\x1d\n\x19\x42\x41\x43KEND_FRAMEWORK_UNKNOWN\x10\x00\x12\x1a\n\x16\x42\x41\x43KEND_FRAMEWORK_VLLM\x10\x01\x12\x1c\n\x18\x42\x41\x43KEND_FRAMEWORK_SGLANG\x10\x02\x12\x1d\n\x19\x42\x41\x43KEND_FRAMEWORK_TRT_LLM\x10\x03*\xba\x02\n\x0cMxSourceType\x12\x1a\n\x16MX_SOURCE_TYPE_WEIGHTS\x10\x00\x12\x17\n\x13MX_SOURCE_TYPE_LORA\x10\x01\x12\x1d\n\x19MX_SOURCE_TYPE_CUDA_GRAPH\x10\x02\x12&\n\"MX_SOURCE_TYPE_TORCH_COMPILE_CACHE\x10\x03\x12\x1f\n\x1bMX_SOURCE_TYPE_TRITON_CACHE\x10\x04\x12\"\n\x1eMX_SOURCE_TYPE_DEEP_GEMM_CACHE\x10\x05\x12!\n\x1dMX_SOURCE_TYPE_TILELANG_CACHE\x10\x06\x12!\n\x1dMX_SOURCE_TYPE_CUTE_DSL_CACHE\x10\x07\x12#\n\x1fMX_SOURCE_TYPE_FLASHINFER_CACHE\x10\x08*{\n\x0cSourceStatus\x12\x19\n\x15SOURCE_STATUS_UNKNOWN\x10\x00\x12\x1e\n\x1aSOURCE_STATUS_INITIALIZING\x10\x01\x12\x17\n\x13SOURCE_STATUS_READY\x10\x02\x12\x17\n\x13SOURCE_STATUS_STALE\x10\x03\x32\x93\x03\n\nP2pService\x12h\n\x0fPublishMetadata\x12).model_express.p2p.PublishMetadataRequest\x1a*.model_express.p2p.PublishMetadataResponse\x12\\\n\x0bListSources\x12%.model_express.p2p.ListSourcesRequest\x1a&.model_express.p2p.ListSourcesResponse\x12\\\n\x0bGetMetadata\x12%.model_express.p2p.GetMetadataRequest\x1a&.model_express.p2p.GetMetadataResponse\x12_\n\x0cUpdateStatus\x12&.model_express.p2p.UpdateStatusRequest\x1a\'.model_express.p2p.UpdateStatusResponse2\x83\x05\n\rWorkerService\x12n\n\x11GetTensorManifest\x12+.model_express.p2p.GetTensorManifestRequest\x1a,.model_express.p2p.GetTensorManifestResponse\x12\x86\x01\n\x19GetArtifactManifestHeader\x12\x33.model_express.p2p.GetArtifactManifestHeaderRequest\x1a\x34.model_express.p2p.GetArtifactManifestHeaderResponse\x12\x86\x01\n\x19GetArtifactManifestChunks\x12\x33.model_express.p2p.GetArtifactManifestChunksRequest\x1a\x34.model_express.p2p.GetArtifactManifestChunksResponse\x12w\n\x14PrepareArtifactChunk\x12..model_express.p2p.PrepareArtifactChunkRequest\x1a/.model_express.p2p.PrepareArtifactChunkResponse\x12w\n\x14ReleaseArtifactChunk\x12..model_express.p2p.ReleaseArtifactChunkRequest\x1a/.model_express.p2p.ReleaseArtifactChunkResponseb\x06proto3')
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\tp2p.proto\x12\x11model_express.p2p\"\xe7\x04\n\x0eSourceIdentity\x12\x12\n\nmx_version\x18\x01 \x01(\t\x12\x37\n\x0emx_source_type\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\nmodel_name\x18\x03 \x01(\t\x12>\n\x11\x62\x61\x63kend_framework\x18\x04 \x01(\x0e\x32#.model_express.p2p.BackendFramework\x12\x1c\n\x14tensor_parallel_size\x18\x05 \x01(\r\x12\x1e\n\x16pipeline_parallel_size\x18\x06 \x01(\r\x12\x1c\n\x14\x65xpert_parallel_size\x18\x07 \x01(\r\x12\r\n\x05\x64type\x18\x08 \x01(\t\x12\x14\n\x0cquantization\x18\t \x01(\t\x12P\n\x10\x65xtra_parameters\x18\n \x03(\x0b\x32\x36.model_express.p2p.SourceIdentity.ExtraParametersEntry\x12\x10\n\x08revision\x18\x0b \x01(\t\x12!\n\x19\x62\x61\x63kend_framework_version\x18\x0c \x01(\t\x12\x15\n\rtorch_version\x18\r \x01(\t\x12\x14\n\x0c\x63uda_version\x18\x0e \x01(\t\x12\x16\n\x0etriton_version\x18\x0f \x01(\t\x12\x10\n\x08gpu_arch\x18\x10 \x01(\t\x12\x1d\n\x15\x63ompile_config_digest\x18\x11 \x01(\t\x1a\x36\n\x14\x45xtraParametersEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\t:\x02\x38\x01\"^\n\x10TensorDescriptor\x12\x0c\n\x04name\x18\x01 \x01(\t\x12\x0c\n\x04\x61\x64\x64r\x18\x02 \x01(\x04\x12\x0c\n\x04size\x18\x03 \x01(\x04\x12\x11\n\tdevice_id\x18\x04 \x01(\r\x12\r\n\x05\x64type\x18\x05 \x01(\t\"}\n\x16\x41rtifactSourceMetadata\x12\x13\n\x0b\x61rtifact_id\x18\x01 \x01(\t\x12\x12\n\ntotal_size\x18\x02 \x01(\x04\x12\x12\n\nfile_count\x18\x03 \x01(\r\x12\x13\n\x0b\x63hunk_count\x18\x04 \x01(\r\x12\x11\n\tnode_rank\x18\x05 \x01(\r\"L\n\x14TensorSourceMetadata\x12\x34\n\x07tensors\x18\x01 \x03(\x0b\x32#.model_express.p2p.TensorDescriptor\"\xeb\x01\n\x10\x41rtifactManifest\x12\x18\n\x10manifest_version\x18\x01 \x01(\r\x12\x37\n\x0emx_source_type\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\nchunk_size\x18\x03 \x01(\x04\x12\x36\n\x05\x66iles\x18\x04 \x03(\x0b\x32\'.model_express.p2p.ArtifactManifestFile\x12\x38\n\x06\x63hunks\x18\x05 \x03(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\"X\n\x14\x41rtifactManifestFile\x12\x12\n\nfile_index\x18\x01 \x01(\r\x12\x0c\n\x04path\x18\x02 \x01(\t\x12\x0c\n\x04size\x18\x03 \x01(\x04\x12\x10\n\x08\x63hecksum\x18\x04 \x01(\t\"w\n\x15\x41rtifactManifestChunk\x12\x13\n\x0b\x63hunk_index\x18\x01 \x01(\r\x12\x12\n\nfile_index\x18\x02 \x01(\r\x12\x13\n\x0b\x66ile_offset\x18\x03 \x01(\x04\x12\x0e\n\x06length\x18\x04 \x01(\x04\x12\x10\n\x08\x63hecksum\x18\x05 \x01(\t\"\xf3\x03\n\x0eWorkerMetadata\x12\x13\n\x0bworker_rank\x18\x01 \x01(\r\x12\x17\n\rnixl_metadata\x18\x02 \x01(\x0cH\x00\x12$\n\x1atransfer_engine_session_id\x18\n \x01(\tH\x00\x12\x38\n\x07tensors\x18\x03 \x03(\x0b\x32#.model_express.p2p.TensorDescriptorB\x02\x18\x01\x12/\n\x06status\x18\x04 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatus\x12\x12\n\nupdated_at\x18\x05 \x01(\x03\x12\x19\n\x11metadata_endpoint\x18\x06 \x01(\t\x12\x12\n\nagent_name\x18\x07 \x01(\t\x12\x1c\n\x14worker_grpc_endpoint\x18\x08 \x01(\t\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\t \x01(\t\x12@\n\rtensor_source\x18\x14 \x01(\x0b\x32\'.model_express.p2p.TensorSourceMetadataH\x01\x12\x44\n\x0f\x61rtifact_source\x18\x15 \x01(\x0b\x32).model_express.p2p.ArtifactSourceMetadataH\x01\x42\x12\n\x10\x62\x61\x63kend_metadataB\x10\n\x0esource_payload\"V\n\x18GetTensorManifestRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x16\n\tworker_id\x18\x02 \x01(\tH\x00\x88\x01\x01\x42\x0c\n\n_worker_id\"\xe6\x01\n\x19GetTensorManifestResponse\x12\x34\n\x07tensors\x18\x01 \x03(\x0b\x32#.model_express.p2p.TensorDescriptor\x12\x14\n\x0cmx_source_id\x18\x02 \x01(\t\x12\x19\n\x11metadata_endpoint\x18\x03 \x01(\t\x12\x12\n\nagent_name\x18\x04 \x01(\t\x12\x13\n\x0bworker_rank\x18\x05 \x01(\r\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\x06 \x01(\t\x12\x16\n\tworker_id\x18\x07 \x01(\tH\x00\x88\x01\x01\x42\x0c\n\n_worker_id\"M\n GetArtifactManifestHeaderRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\"\xee\x02\n!GetArtifactManifestHeaderResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x18\n\x10manifest_version\x18\x03 \x01(\r\x12\x37\n\x0emx_source_type\x18\x04 \x01(\x0e\x32\x1f.model_express.p2p.MxSourceType\x12\x12\n\ntotal_size\x18\x05 \x01(\x04\x12\x12\n\nfile_count\x18\x06 \x01(\r\x12\x13\n\x0b\x63hunk_count\x18\x07 \x01(\r\x12\x12\n\nchunk_size\x18\x08 \x01(\x04\x12\x19\n\x11metadata_endpoint\x18\t \x01(\t\x12\x12\n\nagent_name\x18\n \x01(\t\x12\x13\n\x0bworker_rank\x18\x0b \x01(\r\x12\x36\n\x05\x66iles\x18\x0c \x03(\x0b\x32\'.model_express.p2p.ArtifactManifestFile\"|\n GetArtifactManifestChunksRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x19\n\x11start_chunk_index\x18\x03 \x01(\r\x12\x12\n\nmax_chunks\x18\x04 \x01(\r\"\xbc\x01\n!GetArtifactManifestChunksResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x19\n\x11start_chunk_index\x18\x03 \x01(\r\x12\x38\n\x06\x63hunks\x18\x04 \x03(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\x12\x17\n\x0fnext_page_token\x18\x05 \x01(\t\"R\n\x1f\x41rtifactChunkTransferDescriptor\x12\x0c\n\x04\x61\x64\x64r\x18\x01 \x01(\x04\x12\x0e\n\x06length\x18\x02 \x01(\x04\x12\x11\n\tdevice_id\x18\x03 \x01(\r\"]\n\x1bPrepareArtifactChunkRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x13\n\x0b\x63hunk_index\x18\x03 \x01(\r\"\xf1\x01\n\x1cPrepareArtifactChunkResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x10\n\x08lease_id\x18\x03 \x01(\t\x12\x37\n\x05\x63hunk\x18\x04 \x01(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\x12\x42\n\x06source\x18\x05 \x01(\x0b\x32\x32.model_express.p2p.ArtifactChunkTransferDescriptor\x12\x17\n\x0fsource_metadata\x18\x06 \x01(\x0c\"Z\n\x1bReleaseArtifactChunkRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x10\n\x08lease_id\x18\x03 \x01(\t\"\x82\x01\n\x1cReleaseArtifactChunkResponse\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0b\x61rtifact_id\x18\x02 \x01(\t\x12\x37\n\x05\x63hunk\x18\x03 \x01(\x0b\x32(.model_express.p2p.ArtifactManifestChunk\"\xcd\x01\n\x16PublishMetadataRequest\x12\x33\n\x08identity\x18\x01 \x01(\x0b\x32!.model_express.p2p.SourceIdentity\x12\x31\n\x06worker\x18\x02 \x01(\x0b\x32!.model_express.p2p.WorkerMetadata\x12\x11\n\tworker_id\x18\x03 \x01(\t\x12\x10\n\x08pod_name\x18\x04 \x01(\t\x12\x0f\n\x07pod_uid\x18\x05 \x01(\t\x12\x15\n\rpod_namespace\x18\x06 \x01(\t\"d\n\x17PublishMetadataResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\x12\x14\n\x0cmx_source_id\x18\x03 \x01(\t\x12\x11\n\tworker_id\x18\x04 \x01(\t\"z\n\x11SourceInstanceRef\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x11\n\tworker_id\x18\x02 \x01(\t\x12\x12\n\nmodel_name\x18\x03 \x01(\t\x12\x13\n\x0bworker_rank\x18\x04 \x01(\r\x12\x13\n\x0b\x61\x63\x63\x65lerator\x18\x05 \x01(\t\"\x98\x01\n\x12ListSourcesRequest\x12\x33\n\x08identity\x18\x01 \x01(\x0b\x32!.model_express.p2p.SourceIdentity\x12;\n\rstatus_filter\x18\x02 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatusH\x00\x88\x01\x01\x42\x10\n\x0e_status_filter\"N\n\x13ListSourcesResponse\x12\x37\n\tinstances\x18\x01 \x03(\x0b\x32$.model_express.p2p.SourceInstanceRef\"=\n\x12GetMetadataRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x11\n\tworker_id\x18\x02 \x01(\t\"\x80\x01\n\x13GetMetadataResponse\x12\r\n\x05\x66ound\x18\x01 \x01(\x08\x12\x31\n\x06worker\x18\x02 \x01(\x0b\x32!.model_express.p2p.WorkerMetadata\x12\x14\n\x0cmx_source_id\x18\x03 \x01(\t\x12\x11\n\tworker_id\x18\x04 \x01(\t\"\x84\x01\n\x13UpdateStatusRequest\x12\x14\n\x0cmx_source_id\x18\x01 \x01(\t\x12\x13\n\x0bworker_rank\x18\x02 \x01(\r\x12/\n\x06status\x18\x03 \x01(\x0e\x32\x1f.model_express.p2p.SourceStatus\x12\x11\n\tworker_id\x18\x04 \x01(\t\"8\n\x14UpdateStatusResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t*\x8a\x01\n\x10\x42\x61\x63kendFramework\x12\x1d\n\x19\x42\x41\x43KEND_FRAMEWORK_UNKNOWN\x10\x00\x12\x1a\n\x16\x42\x41\x43KEND_FRAMEWORK_VLLM\x10\x01\x12\x1c\n\x18\x42\x41\x43KEND_FRAMEWORK_SGLANG\x10\x02\x12\x1d\n\x19\x42\x41\x43KEND_FRAMEWORK_TRT_LLM\x10\x03*\xba\x02\n\x0cMxSourceType\x12\x1a\n\x16MX_SOURCE_TYPE_WEIGHTS\x10\x00\x12\x17\n\x13MX_SOURCE_TYPE_LORA\x10\x01\x12\x1d\n\x19MX_SOURCE_TYPE_CUDA_GRAPH\x10\x02\x12&\n\"MX_SOURCE_TYPE_TORCH_COMPILE_CACHE\x10\x03\x12\x1f\n\x1bMX_SOURCE_TYPE_TRITON_CACHE\x10\x04\x12\"\n\x1eMX_SOURCE_TYPE_DEEP_GEMM_CACHE\x10\x05\x12!\n\x1dMX_SOURCE_TYPE_TILELANG_CACHE\x10\x06\x12!\n\x1dMX_SOURCE_TYPE_CUTE_DSL_CACHE\x10\x07\x12#\n\x1fMX_SOURCE_TYPE_FLASHINFER_CACHE\x10\x08*{\n\x0cSourceStatus\x12\x19\n\x15SOURCE_STATUS_UNKNOWN\x10\x00\x12\x1e\n\x1aSOURCE_STATUS_INITIALIZING\x10\x01\x12\x17\n\x13SOURCE_STATUS_READY\x10\x02\x12\x17\n\x13SOURCE_STATUS_STALE\x10\x03\x32\x93\x03\n\nP2pService\x12h\n\x0fPublishMetadata\x12).model_express.p2p.PublishMetadataRequest\x1a*.model_express.p2p.PublishMetadataResponse\x12\\\n\x0bListSources\x12%.model_express.p2p.ListSourcesRequest\x1a&.model_express.p2p.ListSourcesResponse\x12\\\n\x0bGetMetadata\x12%.model_express.p2p.GetMetadataRequest\x1a&.model_express.p2p.GetMetadataResponse\x12_\n\x0cUpdateStatus\x12&.model_express.p2p.UpdateStatusRequest\x1a\'.model_express.p2p.UpdateStatusResponse2\x83\x05\n\rWorkerService\x12n\n\x11GetTensorManifest\x12+.model_express.p2p.GetTensorManifestRequest\x1a,.model_express.p2p.GetTensorManifestResponse\x12\x86\x01\n\x19GetArtifactManifestHeader\x12\x33.model_express.p2p.GetArtifactManifestHeaderRequest\x1a\x34.model_express.p2p.GetArtifactManifestHeaderResponse\x12\x86\x01\n\x19GetArtifactManifestChunks\x12\x33.model_express.p2p.GetArtifactManifestChunksRequest\x1a\x34.model_express.p2p.GetArtifactManifestChunksResponse\x12w\n\x14PrepareArtifactChunk\x12..model_express.p2p.PrepareArtifactChunkRequest\x1a/.model_express.p2p.PrepareArtifactChunkResponse\x12w\n\x14ReleaseArtifactChunk\x12..model_express.p2p.ReleaseArtifactChunkRequest\x1a/.model_express.p2p.ReleaseArtifactChunkResponseb\x06proto3')
 
 _globals = globals()
 _builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
@@ -38,12 +38,12 @@
   _globals['_SOURCEIDENTITY_EXTRAPARAMETERSENTRY']._serialized_options = b'8\001'
   _globals['_WORKERMETADATA'].fields_by_name['tensors']._loaded_options = None
   _globals['_WORKERMETADATA'].fields_by_name['tensors']._serialized_options = b'\030\001'
-  _globals['_BACKENDFRAMEWORK']._serialized_start=4670
-  _globals['_BACKENDFRAMEWORK']._serialized_end=4808
-  _globals['_MXSOURCETYPE']._serialized_start=4811
-  _globals['_MXSOURCETYPE']._serialized_end=5125
-  _globals['_SOURCESTATUS']._serialized_start=5127
-  _globals['_SOURCESTATUS']._serialized_end=5250
+  _globals['_BACKENDFRAMEWORK']._serialized_start=4693
+  _globals['_BACKENDFRAMEWORK']._serialized_end=4831
+  _globals['_MXSOURCETYPE']._serialized_start=4834
+  _globals['_MXSOURCETYPE']._serialized_end=5148
+  _globals['_SOURCESTATUS']._serialized_start=5150
+  _globals['_SOURCESTATUS']._serialized_end=5273
   _globals['_SOURCEIDENTITY']._serialized_start=33
   _globals['_SOURCEIDENTITY']._serialized_end=648
   _globals['_SOURCEIDENTITY_EXTRAPARAMETERSENTRY']._serialized_start=594
@@ -85,25 +85,25 @@
   _globals['_RELEASEARTIFACTCHUNKRESPONSE']._serialized_start=3504
   _globals['_RELEASEARTIFACTCHUNKRESPONSE']._serialized_end=3634
   _globals['_PUBLISHMETADATAREQUEST']._serialized_start=3637
-  _globals['_PUBLISHMETADATAREQUEST']._serialized_end=3819
-  _globals['_PUBLISHMETADATARESPONSE']._serialized_start=3821
-  _globals['_PUBLISHMETADATARESPONSE']._serialized_end=3921
-  _globals['_SOURCEINSTANCEREF']._serialized_start=3923
-  _globals['_SOURCEINSTANCEREF']._serialized_end=4045
-  _globals['_LISTSOURCESREQUEST']._serialized_start=4048
-  _globals['_LISTSOURCESREQUEST']._serialized_end=4200
-  _globals['_LISTSOURCESRESPONSE']._serialized_start=4202
-  _globals['_LISTSOURCESRESPONSE']._serialized_end=4280
-  _globals['_GETMETADATAREQUEST']._serialized_start=4282
-  _globals['_GETMETADATAREQUEST']._serialized_end=4343
-  _globals['_GETMETADATARESPONSE']._serialized_start=4346
-  _globals['_GETMETADATARESPONSE']._serialized_end=4474
-  _globals['_UPDATESTATUSREQUEST']._serialized_start=4477
-  _globals['_UPDATESTATUSREQUEST']._serialized_end=4609
-  _globals['_UPDATESTATUSRESPONSE']._serialized_start=4611
-  _globals['_UPDATESTATUSRESPONSE']._serialized_end=4667
-  _globals['_P2PSERVICE']._serialized_start=5253
-  _globals['_P2PSERVICE']._serialized_end=5656
-  _globals['_WORKERSERVICE']._serialized_start=5659
-  _globals['_WORKERSERVICE']._serialized_end=6302
+  _globals['_PUBLISHMETADATAREQUEST']._serialized_end=3842
+  _globals['_PUBLISHMETADATARESPONSE']._serialized_start=3844
+  _globals['_PUBLISHMETADATARESPONSE']._serialized_end=3944
+  _globals['_SOURCEINSTANCEREF']._serialized_start=3946
+  _globals['_SOURCEINSTANCEREF']._serialized_end=4068
+  _globals['_LISTSOURCESREQUEST']._serialized_start=4071
+  _globals['_LISTSOURCESREQUEST']._serialized_end=4223
+  _globals['_LISTSOURCESRESPONSE']._serialized_start=4225
+  _globals['_LISTSOURCESRESPONSE']._serialized_end=4303
+  _globals['_GETMETADATAREQUEST']._serialized_start=4305
+  _globals['_GETMETADATAREQUEST']._serialized_end=4366
+  _globals['_GETMETADATARESPONSE']._serialized_start=4369
+  _globals['_GETMETADATARESPONSE']._serialized_end=4497
+  _globals['_UPDATESTATUSREQUEST']._serialized_start=4500
+  _globals['_UPDATESTATUSREQUEST']._serialized_end=4632
+  _globals['_UPDATESTATUSRESPONSE']._serialized_start=4634
+  _globals['_UPDATESTATUSRESPONSE']._serialized_end=4690
+  _globals['_P2PSERVICE']._serialized_start=5276
+  _globals['_P2PSERVICE']._serialized_end=5679
+  _globals['_WORKERSERVICE']._serialized_start=5682
+  _globals['_WORKERSERVICE']._serialized_end=6325
 # @@protoc_insertion_point(module_scope)
diff --git a/modelexpress_client/python/tests/test_artifact_transfer.py b/modelexpress_client/python/tests/test_artifact_transfer.py
index 0742143a..8400b582 100644
--- a/modelexpress_client/python/tests/test_artifact_transfer.py
+++ b/modelexpress_client/python/tests/test_artifact_transfer.py
@@ -1366,18 +1366,14 @@ def __init__(self, mx_source_id: str | None = None):
         self.published_identity = None
         self.published_worker = None
         self.published_worker_id = ""
-        self.published_pod_name = ""
-        self.published_pod_uid = ""
         self.status_updates = []
 
-    def publish_metadata(self, identity, worker, worker_id, pod_name="", pod_uid=""):
+    def publish_metadata(self, identity, worker, worker_id):
         self.published_identity = p2p_pb2.SourceIdentity()
         self.published_identity.CopyFrom(identity)
         self.published_worker = p2p_pb2.WorkerMetadata()
         self.published_worker.CopyFrom(worker)
         self.published_worker_id = worker_id
-        self.published_pod_name = pod_name
-        self.published_pod_uid = pod_uid
         return self.mx_source_id or compute_mx_source_id(identity)
 
     def list_sources(self, identity=None, status_filter=None):
diff --git a/modelexpress_client/python/tests/test_client.py b/modelexpress_client/python/tests/test_client.py
new file mode 100644
index 00000000..729a087b
--- /dev/null
+++ b/modelexpress_client/python/tests/test_client.py
@@ -0,0 +1,33 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+from unittest.mock import MagicMock
+
+from modelexpress import p2p_pb2
+from modelexpress.client import MxClient
+
+
+def test_publish_metadata_adds_pod_identity_without_changing_public_api(monkeypatch):
+    monkeypatch.setenv("POD_NAME", "source-0")
+    monkeypatch.setenv("POD_UID", "pod-uid")
+    monkeypatch.setenv("POD_NAMESPACE", "default")
+    stub = MagicMock()
+    stub.PublishMetadata.return_value = p2p_pb2.PublishMetadataResponse(
+        success=True,
+        mx_source_id="source-id",
+    )
+    client = MxClient("localhost:8001")
+    client._channel = MagicMock()
+    client._stub = stub
+
+    source_id = client.publish_metadata(
+        p2p_pb2.SourceIdentity(model_name="model"),
+        p2p_pb2.WorkerMetadata(worker_rank=0),
+        "worker-id",
+    )
+
+    assert source_id == "source-id"
+    request = stub.PublishMetadata.call_args.args[0]
+    assert request.pod_name == "source-0"
+    assert request.pod_uid == "pod-uid"
+    assert request.pod_namespace == "default"
diff --git a/modelexpress_client/python/tests/test_types.py b/modelexpress_client/python/tests/test_types.py
index 4328afa3..a629937e 100644
--- a/modelexpress_client/python/tests/test_types.py
+++ b/modelexpress_client/python/tests/test_types.py
@@ -6,6 +6,7 @@
 import re
 
 from google.protobuf import __version__ as _pb_version
+from google.protobuf import descriptor_pb2, descriptor_pool, message_factory
 
 from modelexpress import p2p_pb2
 from modelexpress.metadata.payload import (
@@ -36,6 +37,54 @@ def test_p2p_pb2_gencode_matches_runtime_major_version(self):
             f"../../modelexpress_common/proto/p2p.proto"
         )
 
+    def test_publish_request_is_readable_by_0_4_schema(self):
+        legacy_file = descriptor_pb2.FileDescriptorProto(
+            name="modelexpress_0_4_publish_metadata.proto",
+            package="model_express.p2p.compat_0_4",
+            dependency=[p2p_pb2.DESCRIPTOR.name],
+            syntax="proto3",
+        )
+        request = legacy_file.message_type.add(name="PublishMetadataRequest")
+        for name, number, type_name in (
+            ("identity", 1, ".model_express.p2p.SourceIdentity"),
+            ("worker", 2, ".model_express.p2p.WorkerMetadata"),
+        ):
+            request.field.add(
+                name=name,
+                number=number,
+                label=descriptor_pb2.FieldDescriptorProto.LABEL_OPTIONAL,
+                type=descriptor_pb2.FieldDescriptorProto.TYPE_MESSAGE,
+                type_name=type_name,
+            )
+        request.field.add(
+            name="worker_id",
+            number=3,
+            label=descriptor_pb2.FieldDescriptorProto.LABEL_OPTIONAL,
+            type=descriptor_pb2.FieldDescriptorProto.TYPE_STRING,
+        )
+
+        pool = descriptor_pool.Default()
+        pool.Add(legacy_file)
+        legacy_request = message_factory.GetMessageClass(
+            pool.FindMessageTypeByName(
+                "model_express.p2p.compat_0_4.PublishMetadataRequest"
+            )
+        )()
+        current_request = p2p_pb2.PublishMetadataRequest(
+            identity=p2p_pb2.SourceIdentity(model_name="model"),
+            worker=p2p_pb2.WorkerMetadata(worker_rank=1),
+            worker_id="worker",
+            pod_name="source-0",
+            pod_uid="pod-uid",
+            pod_namespace="default",
+        )
+
+        legacy_request.ParseFromString(current_request.SerializeToString())
+
+        assert legacy_request.identity.model_name == "model"
+        assert legacy_request.worker.worker_rank == 1
+        assert legacy_request.worker_id == "worker"
+
 
 class TestTensorDescriptor:
     """Tests for TensorDescriptor dataclass."""
diff --git a/modelexpress_common/proto/p2p.proto b/modelexpress_common/proto/p2p.proto
index 435e84db..253197e1 100644
--- a/modelexpress_common/proto/p2p.proto
+++ b/modelexpress_common/proto/p2p.proto
@@ -501,13 +501,17 @@ message PublishMetadataRequest {
   string worker_id = 3;
 
   // Kubernetes pod name of the calling worker, set via the downward API.
-  // Used with pod_uid to set an ownerReference on the ModelMetadata CR
-  // so K8s garbage-collects it when the pod is deleted.
-  // When empty, no ownerReference is set.
+  // Used with pod_uid and pod_namespace to set an ownerReference on the
+  // ModelMetadata CR so K8s garbage-collects it when the pod is deleted.
   string pod_name = 4;
 
   // Kubernetes pod UID of the calling worker, set via the downward API.
   string pod_uid = 5;
+
+  // Kubernetes namespace of the calling worker, set via the downward API.
+  // Owner references are only valid when this matches the ModelMetadata
+  // namespace. Empty means no ownerReference is set.
+  string pod_namespace = 6;
 }
 
 message PublishMetadataResponse {
diff --git a/modelexpress_server/src/p2p/backend.rs b/modelexpress_server/src/p2p/backend.rs
index c00f6408..b050c14b 100644
--- a/modelexpress_server/src/p2p/backend.rs
+++ b/modelexpress_server/src/p2p/backend.rs
@@ -298,6 +298,7 @@ pub trait MetadataBackend: Send + Sync {
         worker: WorkerMetadata,
         pod_name: &str,
         pod_uid: &str,
+        pod_namespace: &str,
     ) -> MetadataResult<()>;
 
     /// Get full tensor metadata for one specific worker.
diff --git a/modelexpress_server/src/p2p/backend/kubernetes.rs b/modelexpress_server/src/p2p/backend/kubernetes.rs
index 2cbb7b0c..56f708a7 100644
--- a/modelexpress_server/src/p2p/backend/kubernetes.rs
+++ b/modelexpress_server/src/p2p/backend/kubernetes.rs
@@ -50,8 +50,17 @@ impl KubernetesBackend {
         Api::namespaced(self.client.clone(), &self.namespace)
     }
 
-    fn pod_owner_references(pod_name: &str, pod_uid: &str) -> Option<Vec<OwnerReference>> {
-        if pod_name.is_empty() || pod_uid.is_empty() {
+    fn pod_owner_references(
+        pod_name: &str,
+        pod_uid: &str,
+        pod_namespace: &str,
+        metadata_namespace: &str,
+    ) -> Option<Vec<OwnerReference>> {
+        if pod_name.is_empty()
+            || pod_uid.is_empty()
+            || pod_namespace.is_empty()
+            || pod_namespace != metadata_namespace
+        {
             return None;
         }
 
@@ -207,6 +216,7 @@ impl MetadataBackend for KubernetesBackend {
         worker: WorkerMetadata,
         pod_name: &str,
         pod_uid: &str,
+        pod_namespace: &str,
     ) -> MetadataResult<()> {
         let source_id = crate::p2p::source_identity::compute_mx_source_id(identity);
         let source_id = source_id.as_str();
@@ -219,7 +229,8 @@ impl MetadataBackend for KubernetesBackend {
 
         // First, ensure the CR exists
         let existing = api.get_opt(&cr_name).await?;
-        let pod_owner_references = Self::pod_owner_references(pod_name, pod_uid);
+        let pod_owner_references =
+            Self::pod_owner_references(pod_name, pod_uid, pod_namespace, &self.namespace);
 
         if existing.is_none() {
             let new_cr = ModelMetadata {
@@ -845,16 +856,21 @@ mod tests {
     }
 
     #[test]
-    fn pod_owner_references_require_name_and_uid() {
-        assert!(KubernetesBackend::pod_owner_references("pod-0", "").is_none());
-        assert!(KubernetesBackend::pod_owner_references("", "pod-uid").is_none());
+    fn pod_owner_references_require_complete_same_namespace_identity() {
+        assert!(KubernetesBackend::pod_owner_references("pod-0", "", "ns", "ns").is_none());
+        assert!(KubernetesBackend::pod_owner_references("", "pod-uid", "ns", "ns").is_none());
+        assert!(KubernetesBackend::pod_owner_references("pod-0", "pod-uid", "", "ns").is_none());
+        assert!(
+            KubernetesBackend::pod_owner_references("pod-0", "pod-uid", "other", "ns").is_none()
+        );
     }
 
     #[test]
     fn pod_owner_references_identify_the_publishing_pod() {
-        let Some(owner_references) = KubernetesBackend::pod_owner_references("pod-0", "pod-uid")
+        let Some(owner_references) =
+            KubernetesBackend::pod_owner_references("pod-0", "pod-uid", "ns", "ns")
         else {
-            panic!("pod name and UID should produce an owner reference");
+            panic!("a complete same-namespace pod identity should produce an owner reference");
         };
 
         assert_eq!(owner_references.len(), 1);
diff --git a/modelexpress_server/src/p2p/backend/memory.rs b/modelexpress_server/src/p2p/backend/memory.rs
index a32f9370..7e964025 100644
--- a/modelexpress_server/src/p2p/backend/memory.rs
+++ b/modelexpress_server/src/p2p/backend/memory.rs
@@ -61,6 +61,7 @@ impl MetadataBackend for InMemoryMetadataBackend {
         worker: WorkerMetadata,
         _pod_name: &str,
         _pod_uid: &str,
+        _pod_namespace: &str,
     ) -> MetadataResult<()> {
         let source_id = crate::p2p::source_identity::compute_mx_source_id(identity);
         let record = WorkerRecord::from(worker);
@@ -235,7 +236,7 @@ mod tests {
         let source_id = compute_mx_source_id(&id);
 
         backend
-            .publish_metadata(&id, "w1", worker(0, SourceStatus::Ready), "", "")
+            .publish_metadata(&id, "w1", worker(0, SourceStatus::Ready), "", "", "")
             .await
             .expect("publish");
         let record = backend
@@ -269,7 +270,7 @@ mod tests {
 
         for rank in [2, 0, 1] {
             backend
-                .publish_metadata(&id, "w1", worker(rank, SourceStatus::Ready), "", "")
+                .publish_metadata(&id, "w1", worker(rank, SourceStatus::Ready), "", "", "")
                 .await
                 .expect("publish");
         }
@@ -290,11 +291,11 @@ mod tests {
         let id = identity("m");
 
         backend
-            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing), "", "")
+            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing), "", "", "")
             .await
             .expect("publish r0");
         backend
-            .publish_metadata(&id, "w1", worker(1, SourceStatus::Ready), "", "")
+            .publish_metadata(&id, "w1", worker(1, SourceStatus::Ready), "", "", "")
             .await
             .expect("publish r1");
 
@@ -334,11 +335,11 @@ mod tests {
         let source_id = compute_mx_source_id(&id);
 
         backend
-            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing), "", "")
+            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing), "", "", "")
             .await
             .expect("publish r0");
         backend
-            .publish_metadata(&id, "w1", worker(3, SourceStatus::Ready), "", "")
+            .publish_metadata(&id, "w1", worker(3, SourceStatus::Ready), "", "", "")
             .await
             .expect("publish r3");
 
@@ -363,7 +364,7 @@ mod tests {
         let source_id = compute_mx_source_id(&id);
 
         backend
-            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing), "", "")
+            .publish_metadata(&id, "w1", worker(0, SourceStatus::Initializing), "", "", "")
             .await
             .expect("publish");
         backend
@@ -403,11 +404,11 @@ mod tests {
         let source_id = compute_mx_source_id(&id);
 
         backend
-            .publish_metadata(&id, "w1", worker(0, SourceStatus::Ready), "", "")
+            .publish_metadata(&id, "w1", worker(0, SourceStatus::Ready), "", "", "")
             .await
             .expect("publish w1");
         backend
-            .publish_metadata(&id, "w2", worker(0, SourceStatus::Ready), "", "")
+            .publish_metadata(&id, "w2", worker(0, SourceStatus::Ready), "", "", "")
             .await
             .expect("publish w2");
 
diff --git a/modelexpress_server/src/p2p/backend/redis.rs b/modelexpress_server/src/p2p/backend/redis.rs
index c9aee250..27cdd973 100644
--- a/modelexpress_server/src/p2p/backend/redis.rs
+++ b/modelexpress_server/src/p2p/backend/redis.rs
@@ -409,6 +409,7 @@ impl MetadataBackend for RedisBackend {
         worker: WorkerMetadata,
         _pod_name: &str,
         _pod_uid: &str,
+        _pod_namespace: &str,
     ) -> MetadataResult<()> {
         let source_id = crate::p2p::source_identity::compute_mx_source_id(identity);
         let mut conn = self.get_conn().await?;
diff --git a/modelexpress_server/src/p2p/service.rs b/modelexpress_server/src/p2p/service.rs
index f7c64e7b..7e5ee501 100644
--- a/modelexpress_server/src/p2p/service.rs
+++ b/modelexpress_server/src/p2p/service.rs
@@ -116,7 +116,14 @@ impl P2pService for P2pServiceImpl {
 
         match self
             .state
-            .publish_metadata(&identity, &worker_id, worker, &req.pod_name, &req.pod_uid)
+            .publish_metadata(
+                &identity,
+                &worker_id,
+                worker,
+                &req.pod_name,
+                &req.pod_uid,
+                &req.pod_namespace,
+            )
             .await
         {
             Ok(()) => {
@@ -433,11 +440,11 @@ mod tests {
     async fn test_publish_metadata_success() {
         let mut mock = MockMetadataBackend::new();
         mock.expect_publish_metadata()
-            .withf(|_, _, _, pod_name, pod_uid| {
-                pod_name == "vllm-worker-0" && pod_uid == "pod-uid-1"
+            .withf(|_, _, _, pod_name, pod_uid, pod_namespace| {
+                pod_name == "vllm-worker-0" && pod_uid == "pod-uid-1" && pod_namespace == "default"
             })
             .once()
-            .returning(|_, _, _, _, _| Ok(()));
+            .returning(|_, _, _, _, _, _| Ok(()));
 
         let svc = make_service(mock);
         let resp = svc
@@ -456,6 +463,7 @@ mod tests {
                 worker_id: "worker-uuid-1".to_string(),
                 pod_name: "vllm-worker-0".to_string(),
                 pod_uid: "pod-uid-1".to_string(),
+                pod_namespace: "default".to_string(),
             }))
             .await
             .expect("rpc")
@@ -471,7 +479,7 @@ mod tests {
         let mut mock = MockMetadataBackend::new();
         mock.expect_publish_metadata()
             .once()
-            .returning(|_, _, _, _, _| Err("storage unavailable".into()));
+            .returning(|_, _, _, _, _, _| Err("storage unavailable".into()));
 
         let svc = make_service(mock);
         let resp = svc
diff --git a/modelexpress_server/src/p2p/state.rs b/modelexpress_server/src/p2p/state.rs
index a2880b9a..24e87d1f 100644
--- a/modelexpress_server/src/p2p/state.rs
+++ b/modelexpress_server/src/p2p/state.rs
@@ -97,10 +97,18 @@ impl P2pStateManager {
         worker: WorkerMetadata,
         pod_name: &str,
         pod_uid: &str,
+        pod_namespace: &str,
     ) -> MetadataResult<()> {
         self.get_backend()
             .await?
-            .publish_metadata(identity, worker_id, worker, pod_name, pod_uid)
+            .publish_metadata(
+                identity,
+                worker_id,
+                worker,
+                pod_name,
+                pod_uid,
+                pod_namespace,
+            )
             .await
     }
 
@@ -459,16 +467,19 @@ mod tests {
     async fn test_publish_metadata_calls_backend() {
         let mut mock = MockMetadataBackend::new();
         mock.expect_publish_metadata()
-            .withf(|identity, worker_id, worker, pod_name, pod_uid| {
-                identity.model_name == "my-model"
-                    && identity.tensor_parallel_size == 8
-                    && worker_id == "a1b2c3d4"
-                    && worker.worker_rank == 3
-                    && pod_name == "vllm-worker-0"
-                    && pod_uid == "pod-uid-1"
-            })
+            .withf(
+                |identity, worker_id, worker, pod_name, pod_uid, pod_namespace| {
+                    identity.model_name == "my-model"
+                        && identity.tensor_parallel_size == 8
+                        && worker_id == "a1b2c3d4"
+                        && worker.worker_rank == 3
+                        && pod_name == "vllm-worker-0"
+                        && pod_uid == "pod-uid-1"
+                        && pod_namespace == "default"
+                },
+            )
             .once()
-            .returning(|_, _, _, _, _| Ok(()));
+            .returning(|_, _, _, _, _, _| Ok(()));
 
         let manager = P2pStateManager::with_backend(Arc::new(mock));
         manager
@@ -484,6 +495,7 @@ mod tests {
                 },
                 "vllm-worker-0",
                 "pod-uid-1",
+                "default",
             )
             .await
             .expect("publish_metadata failed");
@@ -494,7 +506,7 @@ mod tests {
         let mut mock = MockMetadataBackend::new();
         mock.expect_publish_metadata()
             .once()
-            .returning(|_, _, _, _, _| Err("storage unavailable".into()));
+            .returning(|_, _, _, _, _, _| Err("storage unavailable".into()));
 
         let manager = P2pStateManager::with_backend(Arc::new(mock));
         assert!(
@@ -504,7 +516,8 @@ mod tests {
                     "a1b2c3d4",
                     WorkerMetadata::default(),
                     "",
-                    ""
+                    "",
+                    "",
                 )
                 .await
                 .is_err()