@cofy-x/axern-sdk 0.4.1 → 0.5.0

package/README.md

@@ -101,10 +101,11 @@ Common options:
 - `client`: explicit `AxernClient` shared by the sandbox
 - `argv`, `env`, `cwd`: initial sandbox process configuration
 - `runtimeClass`: runtime selector, for example `runsc` or `runc`
-- `requestCpu`, `requestMemory`: scheduler resource requests such as `500m`
-  and `512MiB`; numeric CPU values are cores and numeric memory values are bytes
-- `limitCpu`, `limitMemory`: runtime cgroup resource limits such as `1` and
-  `1GiB`; numeric CPU values are cores and numeric memory values are bytes
+- `requestCpu`, `requestMemory`, `requestEphemeralStorage`: scheduler resource
+  requests such as `500m`, `512MiB`, and `1GiB`; numeric CPU values are cores
+  and numeric memory/storage values are bytes
+- `limitCpu`, `limitMemory`, `limitEphemeralStorage`: runtime hard limits;
+  numeric CPU values are cores and numeric memory/storage values are bytes
 - `readyTimeoutMs`: service replica readiness timeout
 
 Tunnel options:

package/dist/client/index.d.ts

@@ -33,13 +33,20 @@ export interface CreateServiceOptions {
     env?: Record<string, string>;
     cwd?: string;
     runtimeClass?: string;
+    extensionCapabilities?: readonly ExtensionCapability[];
     volumes?: readonly VolumeMount[];
     requestCpu?: ResourceQuantity;
     requestMemory?: ResourceQuantity;
+    requestEphemeralStorage?: ResourceQuantity;
     limitCpu?: ResourceQuantity;
     limitMemory?: ResourceQuantity;
+    limitEphemeralStorage?: ResourceQuantity;
     labels?: Record<string, string>;
 }
+export interface ExtensionCapability {
+    name: string;
+    value?: string;
+}
 export interface ReadRunOutputOptions {
     cursor?: string;
     follow?: boolean;

package/dist/client/index.js

@@ -207,6 +207,9 @@ export class AxernClient {
                     env: options.env ?? {},
                     cwd: options.cwd ?? "",
                     runtime_class: options.runtimeClass ?? "",
+                    extension_capability_requirements: (options.extensionCapabilities ?? []).map((capability) => ({
+                        capability: { name: capability.name, value: capability.value ?? "" },
+                    })),
                     volume_mounts: serviceVolumeMounts(options.volumes),
                     resources,
                 },

package/dist/proto/axern/control/admin/v1/node.proto

@@ -5,6 +5,8 @@ package axern.control.admin.v1;
 option go_package = "github.com/cofy-x/axern/sdk/go/gen/axern/control/admin/v1;adminv1";
 
 import "google/protobuf/timestamp.proto";
+import "axern/control/capability/v1/capability.proto";
+import "axern/control/node/v1/node_control.proto";
 
 enum AdminNodeLifecycleStatus {
   ADMIN_NODE_LIFECYCLE_STATUS_UNSPECIFIED = 0;
@@ -43,7 +45,93 @@ message RetireAdminNodeResponse {
   AdminNode node = 1;
 }
 
+message GetNodeCapabilitySnapshotRequest {
+  string node_id = 1;
+}
+
+message GetNodeCapabilitySnapshotResponse {
+  axern.control.capability.v1.CapabilitySnapshot snapshot = 1;
+}
+
+message AdminCapabilityTransition {
+  string transition_id = 1;
+  string node_id = 2;
+  string snapshot_id = 3;
+  int64 snapshot_sequence = 4;
+  axern.control.capability.v1.CapabilityKey key = 5;
+  axern.control.capability.v1.CapabilityState old_state = 6;
+  axern.control.capability.v1.CapabilityState new_state = 7;
+  axern.control.capability.v1.CapabilityEvidence old_evidence = 8;
+  axern.control.capability.v1.CapabilityEvidence new_evidence = 9;
+  axern.control.capability.v1.CapabilityReasonCode old_reason_code = 10;
+  axern.control.capability.v1.CapabilityReasonCode new_reason_code = 11;
+  string reason = 12;
+  google.protobuf.Timestamp observed_at = 13;
+  google.protobuf.Timestamp reported_at = 14;
+}
+
+message ListNodeCapabilityTransitionsRequest {
+  string node_id = 1;
+  int32 limit = 2;
+}
+
+message ListNodeCapabilityTransitionsResponse {
+  repeated AdminCapabilityTransition transitions = 1;
+}
+
+message AdminCapabilityReconcileItem {
+  string allocation_id = 1;
+  string node_id = 2;
+  repeated axern.control.capability.v1.CapabilityDependency pending_dependencies = 3;
+  int32 attempts = 4;
+  google.protobuf.Timestamp next_run_at = 5;
+  google.protobuf.Timestamp lease_expires_at = 6;
+  string last_error = 7;
+  google.protobuf.Timestamp updated_at = 8;
+}
+
+message ListCapabilityReconcileQueueRequest {
+  string node_id = 1;
+  int32 limit = 2;
+}
+
+message ListCapabilityReconcileQueueResponse {
+  repeated AdminCapabilityReconcileItem items = 1;
+}
+
+message GetAllocationCapabilityDiagnosticsRequest {
+  string allocation_id = 1;
+}
+
+message GetAllocationCapabilityDiagnosticsResponse {
+  string allocation_id = 1;
+  string node_id = 2;
+  repeated axern.control.capability.v1.CapabilityDependency required_dependencies = 3;
+  repeated axern.control.capability.v1.CapabilityDependency admitted_dependencies = 4;
+  axern.control.capability.v1.CapabilityConditionSet condition_set = 5;
+  AdminCapabilityReconcileItem reconcile = 6;
+  int64 allocation_attempt = 7;
+  bool create_admission_recorded = 8;
+  string create_dependency_set_digest = 9;
+  google.protobuf.Timestamp create_admitted_at = 10;
+  AllocationMemoryAdmissionEvidence memory_admission = 11;
+  axern.control.node.v1.AllocationMemoryObservation latest_memory_observation = 12;
+}
+
+message AllocationMemoryAdmissionEvidence {
+  int64 sandbox_memory_request_bytes = 1;
+  int64 sandbox_memory_limit_bytes = 2;
+  axern.control.node.v1.NodeMemoryBudget node_memory_budget = 3;
+  google.protobuf.Timestamp summary_collected_at = 4;
+  int64 node_local_commitment_bytes = 5;
+  google.protobuf.Timestamp admitted_at = 6;
+}
+
 service NodeAdmin {
   rpc ListAdminNodes(ListAdminNodesRequest) returns (ListAdminNodesResponse) {}
   rpc RetireAdminNode(RetireAdminNodeRequest) returns (RetireAdminNodeResponse) {}
+  rpc GetNodeCapabilitySnapshot(GetNodeCapabilitySnapshotRequest) returns (GetNodeCapabilitySnapshotResponse) {}
+  rpc ListNodeCapabilityTransitions(ListNodeCapabilityTransitionsRequest) returns (ListNodeCapabilityTransitionsResponse) {}
+  rpc ListCapabilityReconcileQueue(ListCapabilityReconcileQueueRequest) returns (ListCapabilityReconcileQueueResponse) {}
+  rpc GetAllocationCapabilityDiagnostics(GetAllocationCapabilityDiagnosticsRequest) returns (GetAllocationCapabilityDiagnosticsResponse) {}
 }

package/dist/proto/axern/control/capability/v1/capability.proto

@@ -0,0 +1,209 @@
+syntax = "proto3";
+
+package axern.control.capability.v1;
+
+option go_package = "github.com/cofy-x/axern/sdk/go/gen/axern/control/capability/v1;capabilityv1";
+
+import "google/protobuf/timestamp.proto";
+
+// PlatformCapability is the closed set of Axern-owned placement and
+// enforcement claims. User and node configuration cannot manufacture these
+// claims; only the owning observed-capability provider may publish them.
+enum PlatformCapability {
+  PLATFORM_CAPABILITY_UNSPECIFIED = 0;
+  PLATFORM_CAPABILITY_PORT_FORWARDING = 1;
+  PLATFORM_CAPABILITY_NETWORK_BRIDGE = 2;
+  PLATFORM_CAPABILITY_NETWORK_BPFNET = 3;
+  PLATFORM_CAPABILITY_CGROUP_V2_MEMORY_CONTROLLER = 4;
+  PLATFORM_CAPABILITY_RUNC_MEMORY_HARD_LIMIT = 5;
+  PLATFORM_CAPABILITY_RUNSC_MEMORY_HARD_LIMIT = 6;
+  PLATFORM_CAPABILITY_FILESTORE_OVERLAYFS_UPPER = 7;
+  PLATFORM_CAPABILITY_XFS_PROJECT_QUOTA = 8;
+  PLATFORM_CAPABILITY_RUNC_EPHEMERAL_STORAGE_HARD_LIMIT = 9;
+  PLATFORM_CAPABILITY_RUNSC_EPHEMERAL_STORAGE_HARD_LIMIT = 10;
+  PLATFORM_CAPABILITY_ROOTFS_LOWER_EROFS = 11;
+  // Provider facts consumed by the catalog's derived workload capabilities.
+  // They are not valid workload requirements on their own.
+  PLATFORM_CAPABILITY_RUNC_MEMORY_ENFORCEMENT_SELF_TEST = 12;
+  PLATFORM_CAPABILITY_RUNSC_MEMORY_ENFORCEMENT_SELF_TEST = 13;
+  PLATFORM_CAPABILITY_RUNC_EPHEMERAL_ENFORCEMENT_SELF_TEST = 14;
+  PLATFORM_CAPABILITY_RUNSC_EPHEMERAL_ENFORCEMENT_SELF_TEST = 15;
+}
+
+message ExtensionCapability {
+  // A DNS-qualified name such as example.com/accelerator. Axern-owned DNS
+  // suffixes are reserved and rejected by API and node configuration parsing.
+  string name = 1;
+  // Values use exact-match semantics. An empty value is a valid claim.
+  string value = 2;
+}
+
+message CapabilityKey {
+  oneof kind {
+    PlatformCapability platform = 1;
+    ExtensionCapability extension = 2;
+  }
+}
+
+enum CapabilityState {
+  CAPABILITY_STATE_UNSPECIFIED = 0;
+  CAPABILITY_STATE_AVAILABLE = 1;
+  CAPABILITY_STATE_DEGRADED = 2;
+  CAPABILITY_STATE_UNAVAILABLE = 3;
+  CAPABILITY_STATE_UNKNOWN = 4;
+}
+
+enum CapabilityProvider {
+  CAPABILITY_PROVIDER_UNSPECIFIED = 0;
+  CAPABILITY_PROVIDER_CONFIG = 1;
+  CAPABILITY_PROVIDER_HOST_CGROUP = 2;
+  CAPABILITY_PROVIDER_FILESTORE = 3;
+  CAPABILITY_PROVIDER_RUNC_SELF_TEST = 4;
+  CAPABILITY_PROVIDER_RUNSC_SELF_TEST = 5;
+  CAPABILITY_PROVIDER_NETWORK_HEALTH = 6;
+  CAPABILITY_PROVIDER_DERIVED = 7;
+}
+
+enum CapabilityReasonCode {
+  CAPABILITY_REASON_CODE_UNSPECIFIED = 0;
+  CAPABILITY_REASON_CODE_AVAILABLE = 1;
+  CAPABILITY_REASON_CODE_DISABLED = 2;
+  CAPABILITY_REASON_CODE_PROBE_FAILED = 3;
+  CAPABILITY_REASON_CODE_PROBE_ERROR = 4;
+  CAPABILITY_REASON_CODE_EXPIRED = 5;
+  CAPABILITY_REASON_CODE_IDENTITY_CHANGED = 6;
+  CAPABILITY_REASON_CODE_DEPENDENCY_UNAVAILABLE = 7;
+  CAPABILITY_REASON_CODE_RECOVERY_PENDING = 8;
+  CAPABILITY_REASON_CODE_ENFORCEMENT_LOST = 9;
+}
+
+message ConfigEvidenceIdentity {
+  string config_digest = 1;
+}
+
+message BootEvidenceIdentity {
+  string boot_id = 1;
+}
+
+message MountEvidenceIdentity {
+  string boot_id = 1;
+  string mount_identity = 2;
+}
+
+message RuntimeEvidenceIdentity {
+  string boot_id = 1;
+  string runtime_name = 2;
+  string runtime_binary_digest = 3;
+  string runtime_config_digest = 4;
+}
+
+message DerivedEvidenceIdentity {
+  string catalog_digest = 1;
+  // Canonical digest of the typed evidence identities for every catalog
+  // dependency. This makes the derived subject change when its mount,
+  // runtime, boot, or config subjects change without treating a normal TTL
+  // refresh as a new identity.
+  string dependency_evidence_digest = 2;
+}
+
+// CapabilityEvidence identifies the concrete subject that was probed. Its
+// identity is orthogonal to observation freshness: a mount or runtime proof
+// may still require valid_until even though its subject identity is stable.
+message CapabilityEvidence {
+  string evidence_id = 1;
+  oneof identity {
+    ConfigEvidenceIdentity config = 2;
+    BootEvidenceIdentity boot = 3;
+    MountEvidenceIdentity mount = 4;
+    RuntimeEvidenceIdentity runtime = 5;
+    DerivedEvidenceIdentity derived = 6;
+  }
+}
+
+// CapabilityObservationProof is the durable, non-recursive proof selected
+// from a snapshot. Dependencies are represented as one bounded flat proof set
+// so persistence and validation remain bounded.
+message CapabilityObservationProof {
+  CapabilityKey key = 1;
+  string observation_id = 2;
+  CapabilityProvider provider = 3;
+  google.protobuf.Timestamp observed_at = 4;
+  google.protobuf.Timestamp valid_until = 5;
+  CapabilityEvidence evidence = 6;
+}
+
+message CapabilityObservation {
+  CapabilityKey key = 1;
+  CapabilityState state = 2;
+  CapabilityProvider provider = 3;
+  string observation_id = 4;
+  google.protobuf.Timestamp observed_at = 5;
+  google.protobuf.Timestamp valid_until = 6;
+  CapabilityEvidence evidence = 7;
+  repeated CapabilityObservationProof dependencies = 8;
+  CapabilityReasonCode reason_code = 9;
+  string reason = 10;
+}
+
+message CapabilitySnapshot {
+  string node_instance_id = 1;
+  int64 sequence = 2;
+  string snapshot_id = 3;
+  google.protobuf.Timestamp collected_at = 4;
+  repeated CapabilityObservation observations = 5;
+}
+
+message CapabilitySnapshotReference {
+  string node_instance_id = 1;
+  int64 sequence = 2;
+  string snapshot_id = 3;
+  google.protobuf.Timestamp collected_at = 4;
+}
+
+message ExtensionCapabilityRequirement {
+  ExtensionCapability capability = 1;
+}
+
+enum CapabilityLossPolicy {
+  CAPABILITY_LOSS_POLICY_UNSPECIFIED = 0;
+  CAPABILITY_LOSS_POLICY_ADMISSION_ONLY = 1;
+  CAPABILITY_LOSS_POLICY_DEGRADE = 2;
+  CAPABILITY_LOSS_POLICY_FAIL_STOP = 3;
+}
+
+message CapabilityDependency {
+  CapabilityKey key = 1;
+  CapabilityLossPolicy loss_policy = 2;
+  CapabilitySnapshotReference selected_snapshot = 3;
+  CapabilityObservationProof selected_observation = 4;
+  repeated CapabilityObservationProof dependency_observations = 5;
+}
+
+// CapabilityDependencySet is the durable representation stored with an
+// allocation. A wrapper keeps JSON persistence on the canonical proto shape.
+message CapabilityDependencySet {
+  repeated CapabilityDependency dependencies = 1;
+}
+
+enum CapabilityConditionState {
+  CAPABILITY_CONDITION_STATE_UNSPECIFIED = 0;
+  CAPABILITY_CONDITION_STATE_HEALTHY = 1;
+  CAPABILITY_CONDITION_STATE_DEGRADED = 2;
+  CAPABILITY_CONDITION_STATE_FAILED = 3;
+  CAPABILITY_CONDITION_STATE_UNKNOWN = 4;
+}
+
+message CapabilityCondition {
+  CapabilityKey key = 1;
+  CapabilityConditionState state = 2;
+  CapabilityReasonCode reason_code = 3;
+  string message = 4;
+  google.protobuf.Timestamp observed_at = 5;
+  CapabilityObservationProof proof = 6;
+}
+
+message CapabilityConditionSet {
+  int64 revision = 1;
+  google.protobuf.Timestamp observed_at = 2;
+  repeated CapabilityCondition conditions = 3;
+}

package/dist/proto/axern/control/common/v1/common.proto

@@ -5,11 +5,21 @@ package axern.control.common.v1;
 option go_package = "github.com/cofy-x/axern/sdk/go/gen/axern/control/common/v1;commonv1";
 
 import "google/protobuf/timestamp.proto";
+import "axern/control/capability/v1/capability.proto";
 import "axern/control/storage/v1/storage_types.proto";
 
 message ResourceQuantity {
   int64 cpu_milli = 1;
+  // Total sandbox memory budget. requests.memory_bytes is the placement,
+  // reservation, and namespace-quota commitment. limits.memory_bytes is the
+  // host cgroup v2 hard boundary and includes runtime processes, guest
+  // accounting, kernel memory, shmem, and charged file/page cache.
   int64 memory_bytes = 2;
+  // Sandbox-lifetime node-local disposable storage managed by Axern. The
+  // current accounting scope is the runtime writable root overlay and its
+  // metadata/copy-up/whiteouts; persistent volumes and immutable image data
+  // are excluded.
+  int64 ephemeral_storage_bytes = 3;
 }
 
 message ResourceSpec {
@@ -41,11 +51,6 @@ message NetworkSpec {
   NetworkMode mode = 1;
 }
 
-message CapabilityRequirement {
-  string name = 1;
-  string value = 2;
-}
-
 message PlacementConstraints {
   map<string, string> node_selector = 1;
 }
@@ -110,7 +115,7 @@ message ExecutionConfig {
   ResourceSpec resources = 4;
   repeated PortSpec ports = 5;
   NetworkSpec network = 6;
-  repeated CapabilityRequirement capability_requirements = 7;
+  repeated axern.control.capability.v1.ExtensionCapabilityRequirement extension_capability_requirements = 7;
   PlacementConstraints placement = 8;
   repeated SecretEnvVar secret_env = 9;
   repeated SecretFile secret_files = 10;
@@ -147,6 +152,8 @@ enum WorkloadDiagnosticCode {
   WORKLOAD_DIAGNOSTIC_CODE_VOLUME_PUBLISH_ERROR = 11;
   WORKLOAD_DIAGNOSTIC_CODE_VOLUME_RELEASE_ERROR = 12;
   WORKLOAD_DIAGNOSTIC_CODE_VOLUME_SPEC_CONFLICT = 13;
+  WORKLOAD_DIAGNOSTIC_CODE_CAPABILITY_ENFORCEMENT_LOST = 14;
+  WORKLOAD_DIAGNOSTIC_CODE_MEMORY_LIMIT_EXCEEDED = 15;
 }
 
 enum LeaseType {

package/dist/proto/axern/control/node/v1/node_control.proto

@@ -5,6 +5,7 @@ package axern.control.node.v1;
 option go_package = "github.com/cofy-x/axern/sdk/go/gen/axern/control/node/v1;nodev1";
 
 import "axern/control/common/v1/common.proto";
+import "axern/control/capability/v1/capability.proto";
 import "axern/control/tunnel/v1/tunnel.proto";
 import "google/protobuf/timestamp.proto";
 
@@ -21,6 +22,7 @@ enum MountType {
   MOUNT_TYPE_OCI = 2;
   MOUNT_TYPE_NYDUS = 3;
   MOUNT_TYPE_OSS = 4;
+  MOUNT_TYPE_EROFS = 5;
 }
 
 enum ComponentState {
@@ -54,6 +56,48 @@ message ResourcesSummary {
   int64 axnoded_committed_bytes = 4;
   int64 axnoded_used_bytes = 5;
   int64 axnoded_memory_unbounded_count = 6;
+  int64 axnoded_ephemeral_storage_committed_bytes = 7;
+  int64 axnoded_ephemeral_storage_used_bytes = 8;
+  int64 axnoded_ephemeral_storage_unbounded_count = 9;
+}
+
+// NodeMemoryBudget is the node-observed scheduling boundary. The system
+// reserve covers node-local control processes and is never charged to a
+// namespace or allocation.
+enum NodeMemoryBudgetMode {
+  NODE_MEMORY_BUDGET_MODE_UNSPECIFIED = 0;
+  // Capacity and internal usage are observed from the delegated cgroup v2
+  // domain. Runtime-specific hard-limit capabilities remain separate facts.
+  NODE_MEMORY_BUDGET_MODE_CGROUP_V2 = 1;
+  // Local-development capacity is observed from the configured node resource
+  // source without claiming a cgroup hard boundary.
+  NODE_MEMORY_BUDGET_MODE_DISABLED_DEV = 2;
+}
+
+message NodeMemoryBudget {
+  // Physical capacity reported by the selected node resource source. This is
+  // a consistency/diagnostic fact and is never added to allocatable memory.
+  int64 physical_capacity_bytes = 1;
+  // The resource source's schedulable value before the Axern system reserve.
+  int64 source_allocatable_bytes = 2;
+  // A finite value further caps source_allocatable_bytes; zero with
+  // delegated_root_limit_finite=false represents kernel `max`.
+  int64 delegated_root_limit_bytes = 3;
+  bool delegated_root_limit_finite = 4;
+  int64 system_reserve_bytes = 5;
+  int64 effective_allocatable_bytes = 6;
+  int64 local_commitment_bytes = 7;
+  int64 cleanup_debt_bytes = 8;
+  int64 internal_current_bytes = 9;
+  // Identity of the capacity domain. Production uses the boot-scoped cgroup
+  // mount/root identity; disabled_dev uses a boot-scoped resource-source
+  // identity and cannot satisfy runtime hard-limit requirements.
+  string capacity_identity = 10;
+  google.protobuf.Timestamp sampled_at = 11;
+  int32 retiring_cgroup_count = 12;
+  int64 oldest_retiring_age_seconds = 13;
+  bool system_reserve_exhausted = 14;
+  NodeMemoryBudgetMode mode = 15;
 }
 
 message PoolsSummary {
@@ -121,6 +165,14 @@ message NodeStorageSummary {
   int64 inodes_available = 7;
   bool collected = 8;
   string error = 9;
+  int64 system_reserve_bytes = 10;
+  int64 reserved_bytes = 11;
+  int64 allocatable_bytes = 12;
+  int64 active_reservations = 13;
+  string filesystem_type = 14;
+  string mount_identity = 15;
+  int64 allocation_used_bytes = 16;
+  bool unlinked_backing_usage_unknown = 17;
 }
 
 message ComponentsSummary {
@@ -157,10 +209,11 @@ message NodeSummary {
   repeated LocalitySummary locality = 5;
   NodeState node_state = 6;
   map<string, string> labels = 7;
-  repeated string capabilities = 8;
+  axern.control.capability.v1.CapabilitySnapshot capability_snapshot = 8;
   axern.control.common.v1.ResourceQuantity capacity = 9;
   axern.control.common.v1.ResourceQuantity allocatable = 10;
   repeated NodeStorageSummary storage = 11;
+  NodeMemoryBudget memory_budget = 12;
 }
 
 message RegisterNodeRequest {
@@ -192,6 +245,10 @@ message AllocationStatusObservation {
   google.protobuf.Timestamp observed_at = 7;
   bool ready = 8;
   string readiness_message = 9;
+  // diagnostic_code is the bounded, machine-readable terminal or degraded
+  // cause observed by the node. The message remains operator context and is
+  // never parsed to recover this code.
+  axern.control.common.v1.WorkloadDiagnosticCode diagnostic_code = 10;
 }
 
 message BatchReportAllocationStatusRequest {
@@ -202,6 +259,85 @@ message BatchReportAllocationStatusRequest {
 
 message BatchReportAllocationStatusResponse {}
 
+// AllocationCapabilityConditionReport is deliberately independent from
+// lifecycle status. Updating capability conditions cannot make an allocation
+// RUNNING, READY, terminal, or alter its exit status/message.
+message AllocationCapabilityConditionReport {
+  string allocation_id = 1;
+  int64 attempt = 2;
+  axern.control.capability.v1.CapabilityConditionSet condition_set = 3;
+}
+
+message BatchReportAllocationCapabilityConditionsRequest {
+  string node_id = 1;
+  string node_auth_token = 2;
+  repeated AllocationCapabilityConditionReport reports = 3;
+}
+
+message BatchReportAllocationCapabilityConditionsResponse {}
+
+// AllocationMemoryObservation is a bounded latest-value report from the host
+// cgroup. Runtime-specific diagnostics such as `runsc usage --full` are not
+// included because they cannot be added to these kernel counters.
+enum AllocationMemoryCleanupState {
+  ALLOCATION_MEMORY_CLEANUP_STATE_UNSPECIFIED = 0;
+  ALLOCATION_MEMORY_CLEANUP_STATE_ASSIGNED = 1;
+  ALLOCATION_MEMORY_CLEANUP_STATE_RETIRING = 2;
+}
+
+message AllocationMemoryObservation {
+  string allocation_id = 1;
+  int64 attempt = 2;
+  int64 revision = 3;
+  google.protobuf.Timestamp observed_at = 4;
+  int64 request_bytes = 5;
+  // Zero means the sandbox has no per-allocation hard limit. Usage and cgroup
+  // identity remain observable, while the control-verification fields must be
+  // false and swap_current_bytes is informational rather than constrained.
+  int64 limit_bytes = 6;
+  int64 current_bytes = 7;
+  int64 peak_bytes = 8;
+  int64 swap_current_bytes = 9;
+  int64 anon_bytes = 10;
+  int64 file_bytes = 11;
+  int64 shmem_bytes = 12;
+  int64 kernel_bytes = 13;
+  int64 dirty_bytes = 14;
+  int64 writeback_bytes = 15;
+  uint64 event_high = 16;
+  uint64 event_max = 17;
+  uint64 event_oom = 18;
+  uint64 event_oom_kill = 19;
+  uint64 event_oom_group_kill = 20;
+  double psi_some_avg10 = 21;
+  double psi_full_avg10 = 22;
+  uint64 psi_some_total_usec = 23;
+  uint64 psi_full_total_usec = 24;
+  string cgroup_identity = 25;
+  string runtime = 26;
+  bool parent_controls_verified = 27;
+  bool leaf_controls_verified = 28;
+  bool pid_roles_verified = 29;
+  AllocationMemoryCleanupState cleanup_state = 30;
+  // psi_available distinguishes an unsupported per-cgroup PSI interface from
+  // a supported interface that currently reports zero pressure. When false,
+  // all PSI values above must be zero.
+  bool psi_available = 31;
+  // peak_available reports whether peak_bytes came from the kernel
+  // memory.peak interface. Older cgroup v2 kernels may omit that optional
+  // file; in that case peak_bytes equals current_bytes and is only a sampled
+  // lower bound. Hard-limit evidence never depends on this observability bit.
+  bool peak_available = 32;
+}
+
+message BatchReportAllocationMemoryObservationsRequest {
+  string node_id = 1;
+  string node_auth_token = 2;
+  repeated AllocationMemoryObservation observations = 3;
+}
+
+message BatchReportAllocationMemoryObservationsResponse {}
+
 message WatchExecutionLeasesRequest {
   string node_id = 1;
   int64 after_revision = 2;
@@ -263,6 +399,9 @@ enum PlacementRejectionReason {
   PLACEMENT_REJECTION_REASON_NETWORK_UNSUPPORTED = 13;
   PLACEMENT_REJECTION_REASON_CAPABILITY_UNSUPPORTED = 14;
   PLACEMENT_REJECTION_REASON_NODE_RETIRED = 15;
+  PLACEMENT_REJECTION_REASON_INSUFFICIENT_EPHEMERAL_STORAGE = 16;
+  PLACEMENT_REJECTION_REASON_NODE_MEMORY_SYSTEM_RESERVE_EXHAUSTED = 17;
+  PLACEMENT_REJECTION_REASON_NODE_MEMORY_BUDGET_UNAVAILABLE = 18;
 }
 
 message PlacementRank {
@@ -296,6 +435,8 @@ service NodeControl {
   rpc RegisterNode(RegisterNodeRequest) returns (RegisterNodeResponse) {}
   rpc ReportNode(ReportNodeRequest) returns (ReportNodeResponse) {}
   rpc BatchReportAllocationStatus(BatchReportAllocationStatusRequest) returns (BatchReportAllocationStatusResponse) {}
+  rpc BatchReportAllocationCapabilityConditions(BatchReportAllocationCapabilityConditionsRequest) returns (BatchReportAllocationCapabilityConditionsResponse) {}
+  rpc BatchReportAllocationMemoryObservations(BatchReportAllocationMemoryObservationsRequest) returns (BatchReportAllocationMemoryObservationsResponse) {}
   rpc WatchExecutionLeases(WatchExecutionLeasesRequest) returns (stream WatchExecutionLeasesResponse) {}
   rpc WatchTunnelSessions(WatchTunnelSessionsRequest) returns (stream WatchTunnelSessionsResponse) {}
   rpc ReportTunnelSessionStatus(ReportTunnelSessionStatusRequest) returns (ReportTunnelSessionStatusResponse) {}

package/dist/proto/axern/control/quota/v1/quota.proto

@@ -18,6 +18,9 @@ message NamespaceQuota {
   int64 version = 8;
   google.protobuf.Timestamp created_at = 9;
   google.protobuf.Timestamp updated_at = 10;
+  google.protobuf.Int64Value ephemeral_storage_bytes_limit = 11;
+  int64 reserved_ephemeral_storage_bytes = 12;
+  google.protobuf.Int64Value available_ephemeral_storage_bytes = 13;
 }
 
 enum NamespaceQuotaEventType {
@@ -36,6 +39,7 @@ enum NamespaceQuotaEventReason {
   NAMESPACE_QUOTA_EVENT_REASON_INSUFFICIENT_CPU = 1;
   NAMESPACE_QUOTA_EVENT_REASON_INSUFFICIENT_MEMORY = 2;
   NAMESPACE_QUOTA_EVENT_REASON_INSUFFICIENT_CPU_MEMORY = 3;
+  NAMESPACE_QUOTA_EVENT_REASON_INSUFFICIENT_EPHEMERAL_STORAGE = 4;
 }
 
 message NamespaceQuotaEvent {
@@ -56,6 +60,10 @@ message NamespaceQuotaEvent {
   google.protobuf.Int64Value available_memory_bytes = 15;
   string message = 16;
   google.protobuf.Timestamp created_at = 17;
+  int64 requested_ephemeral_storage_bytes = 18;
+  int64 reserved_ephemeral_storage_bytes = 19;
+  google.protobuf.Int64Value ephemeral_storage_bytes_limit = 20;
+  google.protobuf.Int64Value available_ephemeral_storage_bytes = 21;
 }
 
 message GetNamespaceQuotaRequest {
@@ -76,6 +84,7 @@ message ListNamespaceQuotasResponse {
 message NamespaceQuotaLimits {
   google.protobuf.Int64Value cpu_milli = 1;
   google.protobuf.Int64Value memory_bytes = 2;
+  google.protobuf.Int64Value ephemeral_storage_bytes = 3;
 }
 
 message SetNamespaceQuotaRequest {

package/dist/proto/axern/control/run/v1/run.proto

@@ -5,6 +5,7 @@ package axern.control.run.v1;
 option go_package = "github.com/cofy-x/axern/sdk/go/gen/axern/control/run/v1;runv1";
 
 import "axern/control/common/v1/common.proto";
+import "axern/control/capability/v1/capability.proto";
 import "google/protobuf/timestamp.proto";
 
 enum RunStatus {
@@ -34,9 +35,10 @@ message Run {
   bool exit_code_known = 13;
   string message = 14;
   axern.control.common.v1.WorkloadDiagnosticCode diagnostic_code = 15;
-  // Reserved for durable output metadata once output retention is owned by
-  // the control plane rather than inferred from a node-local allocation.
-  reserved 16 to 18;
+	// Full, monotonically revised allocation capability condition projection.
+	// It is independent from run lifecycle status and message.
+	axern.control.capability.v1.CapabilityConditionSet capability_conditions = 16;
+  reserved 17 to 18;
 }
 
 message RunListFilter {

package/dist/proto/axern/control/service/v1/service_replica.proto

@@ -5,6 +5,7 @@ package axern.control.service.v1;
 option go_package = "github.com/cofy-x/axern/sdk/go/gen/axern/control/service/v1;servicev1";
 
 import "axern/control/common/v1/common.proto";
+import "axern/control/capability/v1/capability.proto";
 import "google/protobuf/timestamp.proto";
 
 enum ServiceReplicaView {
@@ -35,6 +36,9 @@ message ServiceReplica {
   string readiness_message = 15;
   ServiceReplicaLifecycleRetry lifecycle_retry = 16;
   axern.control.common.v1.WorkspacePreparationFacts workspace_preparation = 17;
+	// Full, monotonically revised allocation capability condition projection.
+	// It is independent from replica lifecycle status and message.
+	axern.control.capability.v1.CapabilityConditionSet capability_conditions = 18;
 }
 
 enum ServiceReplicaLifecycleRetryReason {

package/dist/resources.d.ts

@@ -7,7 +7,9 @@ export type ResourceQuantity = string | number;
 export interface ResourceOptions {
     requestCpu?: ResourceQuantity;
     requestMemory?: ResourceQuantity;
+    requestEphemeralStorage?: ResourceQuantity;
     limitCpu?: ResourceQuantity;
     limitMemory?: ResourceQuantity;
+    limitEphemeralStorage?: ResourceQuantity;
 }
 export declare function buildResourceSpec(options: ResourceOptions): Record<string, unknown> | undefined;

package/dist/resources.js

@@ -5,22 +5,24 @@
  */
 import { SandboxValidationError } from "./errors/index.js";
 export function buildResourceSpec(options) {
-    const requests = resourceQuantity("requestCpu", options.requestCpu, "requestMemory", options.requestMemory);
-    const limits = resourceQuantity("limitCpu", options.limitCpu, "limitMemory", options.limitMemory);
+    const requests = resourceQuantity("requestCpu", options.requestCpu, "requestMemory", options.requestMemory, "requestEphemeralStorage", options.requestEphemeralStorage);
+    const limits = resourceQuantity("limitCpu", options.limitCpu, "limitMemory", options.limitMemory, "limitEphemeralStorage", options.limitEphemeralStorage);
     if (requests === undefined && limits === undefined) {
         return undefined;
     }
     return { requests, limits };
 }
-function resourceQuantity(cpuName, cpuValue, memoryName, memoryValue) {
+function resourceQuantity(cpuName, cpuValue, memoryName, memoryValue, ephemeralStorageName, ephemeralStorageValue) {
     const cpu = parseCpuQuantity(cpuName, cpuValue);
     const memory = parseMemoryQuantity(memoryName, memoryValue);
-    if (cpu <= 0 && memory <= 0) {
+    const ephemeralStorage = parseMemoryQuantity(ephemeralStorageName, ephemeralStorageValue);
+    if (cpu <= 0 && memory <= 0 && ephemeralStorage <= 0) {
         return undefined;
     }
     return {
         cpu_milli: cpu,
         memory_bytes: memory,
+        ephemeral_storage_bytes: ephemeralStorage,
     };
 }
 function parseCpuQuantity(name, value) {

package/dist/sandbox/index.d.ts

@@ -4,6 +4,7 @@
  * SPDX-License-Identifier: Apache-2.0
  */
 import { AxernClient } from "../client/index.js";
+import type { ExtensionCapability } from "../client/index.js";
 import type { SandboxProcess } from "../node/process.js";
 import type { ResourceQuantity } from "../resources.js";
 import type { ChmodOptions, CapabilityStatus, Command, ComputerUseDisplay, ComputerUseKeyboardOptions, ComputerUseMouseOptions, ComputerUseScreenshot, ComputerUseScreenshotOptions, ComputerUseStatus, CopyOptions, DownloadDirOptions, ExecOptions, ExecResult, ImageExecOptions, ImageProcessOptions, MkdirOptions, MoveOptions, NodeCallOptions, ProcessOptions, RemoveOptions, SandboxFileInfo, TunnelMetadata, TunnelOptions, TouchOptions, UploadDirOptions, VolumeMount, WriteFileOptions } from "../types.js";
@@ -17,11 +18,14 @@ export interface SandboxOptions {
     env?: Record<string, string>;
     cwd?: string;
     runtimeClass?: string;
+    extensionCapabilities?: readonly ExtensionCapability[];
     volumes?: readonly VolumeMount[];
     requestCpu?: ResourceQuantity;
     requestMemory?: ResourceQuantity;
+    requestEphemeralStorage?: ResourceQuantity;
     limitCpu?: ResourceQuantity;
     limitMemory?: ResourceQuantity;
+    limitEphemeralStorage?: ResourceQuantity;
     readyTimeoutMs?: number;
     labels?: Record<string, string>;
     registryCredentialId?: string;

package/dist/sandbox/index.js

@@ -59,11 +59,14 @@ export class Sandbox {
                 env: this.options.env,
                 cwd: this.options.cwd,
                 runtimeClass: this.options.runtimeClass,
+                extensionCapabilities: this.options.extensionCapabilities,
                 volumes: this.options.volumes,
                 requestCpu: this.options.requestCpu,
                 requestMemory: this.options.requestMemory,
+                requestEphemeralStorage: this.options.requestEphemeralStorage,
                 limitCpu: this.options.limitCpu,
                 limitMemory: this.options.limitMemory,
+                limitEphemeralStorage: this.options.limitEphemeralStorage,
                 labels: sandboxLabels(this.options.labels),
             });
             this.serviceId = String(service.id ?? "");

package/dist/version.d.ts

@@ -3,5 +3,5 @@
  * Copyright 2026 cofy-x
  * SPDX-License-Identifier: Apache-2.0
  */
-export declare const AXERN_VERSION = "0.4.1";
+export declare const AXERN_VERSION = "0.5.0";
 export declare function platformName(): string;

package/dist/version.js

@@ -3,7 +3,7 @@
  * Copyright 2026 cofy-x
  * SPDX-License-Identifier: Apache-2.0
  */
-export const AXERN_VERSION = "0.4.1";
+export const AXERN_VERSION = "0.5.0";
 export function platformName() {
     return "axern";
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cofy-x/axern-sdk",
-  "version": "0.4.1",
+  "version": "0.5.0",
   "description": "TypeScript SDK for Axern agentic infrastructure and isolated sandboxes",
   "license": "Apache-2.0",
   "repository": {