feat: 资源分区管理优化

.changeset/big-plants-nail.md

@@ -0,0 +1,6 @@
+---
+"@scow/scow-scheduler-adapter-interface": minor
+---
+在 QueryAccountBlockStatus 的Response中变更 blocked 的注释，增加 account_blocked_details
+在 ListImplementedOptionalFeatures 中增加可选功能枚举值
+在 account.proto中增加可选功能rpc接口

.changeset/yellow-chefs-itch.md

@@ -0,0 +1,5 @@
+---
+"@scow/scow-scheduler-adapter-interface": patch
+---
+
+在 GetClusterInfoResponse 中增加集群节点信息，作业信息等内容优化查询效率

protos/account.proto

@@ -68,8 +68,26 @@ message QueryAccountBlockStatusRequest {
   string account_name = 1;
 }
 
+message AccountStatusInPartition {
+  string partition  = 1;
+  bool blocked = 2;
+}
+
 message QueryAccountBlockStatusResponse {
+  // In versions 1.7.0 and earlier:
+  // - The account's status is uniformly consistent across all partitions by default.
+  // - false: The queried account is unblocked in all partitions.
+  // - true: The queried account has no available partitions.
+
+  // In versions later than 1.7.0:
+  // - The field "blocked" represents whether the account is completely blocked across all partitions:
+  // - true: The account is fully blocked in all partitions.
+  // - false: The account has one or more partitions where it is not blocked.
   bool blocked = 1;
+
+  // the details of account blocked status in every partition
+  // only returns in version later than 1.7.0
+  repeated AccountStatusInPartition account_blocked_details = 2;
 }
 
 message DeleteAccountRequest {
@@ -79,6 +97,72 @@ message DeleteAccountRequest {
 message DeleteAccountResponse {
 }
 
+
+
+// ************ Below is the interfaces for optional feature: RESOURCE_MANAGEMENT ***************
+message BlockAccountWithPartitionsRequest {
+  string account_name = 1;
+  // when the value exists: block specified partition(s) of the account
+  // when the value is [] or undefined: block the account in all partitions
+  repeated string blocked_partitions = 2;
+}
+
+message BlockAccountWithPartitionsResponse {
+}
+
+message UnblockAccountWithPartitionsRequest {
+  string account_name = 1;
+  // specify the available partition(s) when executing unblock
+  // when the value is [] or undefined: use all partitions
+  repeated string unblocked_partitions = 2;
+}
+
+message UnblockAccountWithPartitionsResponse {
+}
+
+message QueryAccountBlockStatusWithPartitionsRequest {
+  string account_name = 1;
+  // query in specified partition(s)
+  // when the value is [] or undefined, return all partitions blocked details
+  repeated string queried_partitions = 2;
+}
+
+message QueryAccountBlockStatusWithPartitionsResponse {
+  // return the overall blocked_status in all partitions
+  // false: when the queried account has on or more available partitions
+  // true: when the queried account has no available partitions
+  bool blocked = 1;
+  // the details of account blocked status in every partition
+  repeated AccountStatusInPartition account_blocked_details = 2;
+}
+
+message ClusterAccountInfoWithBlockedDetails {
+  message UserInAccount {
+    string user_id = 1;
+    string user_name = 2;
+    bool blocked = 3;
+  }
+  string account_name = 1;
+  repeated UserInAccount users = 2;
+  optional string owner = 3;
+  // return the overall blocked_status in all partitions
+  // false: when the queried account has one or more available partitions
+  // true: when the queried account has no available partitions
+  bool blocked = 4;
+  // the details of account blocked status in every partition
+  repeated AccountStatusInPartition account_blocked_details = 5;
+}
+
+message GetAllAccountsWithUsersAndBlockedDetailsRequest {
+
+}
+
+message GetAllAccountsWithUsersAndBlockedDetailsResponse {
+  repeated ClusterAccountInfoWithBlockedDetails accounts = 1;
+}
+// ********** Above is the interfaces for optional feature: RESOURCE_MANAGEMENT ***************
+
+
 service AccountService {
 
   /**
@@ -128,17 +212,63 @@ service AccountService {
 
   /*
    * description: query if an account is blocked
+   * Version differences:
+   * - In versions 1.7.0 and earlier:
+   * - The response's 'blocked' field indicates a uniform status across all partitions.
+   * - The 'account_blocked_details' field is not returned.
+   * - In versions after 1.7.0:
+   * - The 'blocked' field indicates whether the account is completely blocked across all partitions.
+   * - The 'account_blocked_details' field provides detailed status information for each partition.
    * errors: 
    * - account not exist
    *   NOT_FOUND, ACCOUNT_NOT_FOUND, {}
    */
   rpc QueryAccountBlockStatus(QueryAccountBlockStatusRequest) returns (QueryAccountBlockStatusResponse);
 
-   /*
+  /*
    * description: delete account
    * errors:
    * - account not exist
    *   NOT_FOUND, ACCOUNT_NOT_FOUND, {}
    */
    rpc DeleteAccount(DeleteAccountRequest) returns (DeleteAccountResponse);
+
+
+   // ********** Below is the rpcs for optional feature: RESOURCE_MANAGEMENT ***************
+  /*  
+   * FOR OPTIOANL FEATURE: RESOURCE_MANAGEMENT
+   * description: block an account with specified partitions
+   * errors: 
+   * - account not exist
+   *   NOT_FOUND, ACCOUNT_NOT_FOUND, {}
+   * special case:
+   * - account already blocked, don't throw error
+   */
+   rpc BlockAccountWithPartitions(BlockAccountWithPartitionsRequest) returns (BlockAccountWithPartitionsResponse);
+  /*
+   * FOR OPTIOANL FEATURE: RESOURCE_MANAGEMENT
+   * description: unblock an account with specified partitions
+   * errors: 
+   * - account not exist
+   *   NOT_FOUND, ACCOUNT_NOT_FOUND, {}
+   * special case:
+   * - account already unblocked, don't throw error
+   */
+   rpc UnblockAccountWithPartitions(UnblockAccountWithPartitionsRequest) returns (UnblockAccountWithPartitionsResponse);
+  /*
+   * FOR OPTIOANL FEATURE: RESOURCE_MANAGEMENT
+   * description: query if an account is blocked with specified partitions
+   * errors: 
+   * - account not exist
+   *   NOT_FOUND, ACCOUNT_NOT_FOUND, {}
+   */
+   rpc QueryAccountBlockStatusWithPartitions(QueryAccountBlockStatusWithPartitionsRequest) returns (QueryAccountBlockStatusWithPartitionsResponse);
+  /*
+   * FOR OPTIOANL FEATURE: RESOURCE_MANAGEMENT
+   * description: get all accounts with blocked partitions' detail and all associated users 
+   * special case:
+   * - account no users, exclude this account
+   */
+   rpc GetAllAccountsWithUsersAndBlockedDetails(GetAllAccountsWithUsersAndBlockedDetailsRequest) returns (GetAllAccountsWithUsersAndBlockedDetailsResponse);
+   // ********** Above is the interfaces for optional feature: RESOURCE_MANAGEMENT ***************
 }

protos/config.proto

@@ -84,6 +84,23 @@ message GetClusterInfoRequest {
 message GetClusterInfoResponse {
   string cluster_name = 1;
   repeated PartitionInfo partitions = 2;
+  // 3-17 Newly added parameters
+  // only returns in version later than 1.7.0
+  uint32 node_count = 3;
+  uint32 running_node_count = 4;
+  uint32 idle_node_count = 5;
+  uint32 not_available_node_count = 6;
+  uint32 cpu_core_count =7;
+  uint32 running_cpu_count = 8;
+  uint32 idle_cpu_count = 9;
+  uint32 not_available_cpu_count = 10;
+  uint32 gpu_core_count = 11;
+  uint32 running_gpu_count = 12;
+  uint32 idle_gpu_count = 13;
+  uint32 not_available_gpu_count = 14;
+  uint32 job_count = 15;
+  uint32 running_job_count = 16;
+  uint32 pending_job_count = 17;
 }
 
 message NodeInfo {
@@ -122,6 +139,7 @@ message ListImplementedOptionalFeaturesRequest {}
 
 enum OptionalFeatures {
   UNKNOWN = 0;
+  RESOURCE_MANAGEMENT = 1;
 }
 
 message ListImplementedOptionalFeaturesResponse {

protos/job.proto

@@ -117,7 +117,6 @@ message GetJobsRequest {
 
   // returned jobs should be sorted if set
   optional SortInfo sort = 4;
-
 }
 
 message GetJobsResponse {