mirror of
https://github.com/wassname/ray.git
synced 2026-07-25 13:30:52 +08:00
[GCS]Use new getting all available resources interface instead of pub-sub … (#10914)
* Use new all available resources getting interface instead of pub-sub in state.py * add missing server handler and test cases, fix comments * add fine grained test assert * per comments * involve new added function _available_resources_per_node * change ClientID to NodeID * fix compile * fix client id and lint * robust tests check * robust tests
This commit is contained in:
@@ -538,6 +538,13 @@ class NodeInfoAccessor {
|
||||
virtual Status AsyncGetResources(const NodeID &node_id,
|
||||
const OptionalItemCallback<ResourceMap> &callback) = 0;
|
||||
|
||||
/// Get available resources of all nodes from GCS asynchronously.
|
||||
///
|
||||
/// \param callback Callback that will be called after lookup finishes.
|
||||
/// \return Status
|
||||
virtual Status AsyncGetAllAvailableResources(
|
||||
const MultiItemCallback<rpc::AvailableResources> &callback) = 0;
|
||||
|
||||
/// Update resources of node in GCS asynchronously.
|
||||
///
|
||||
/// \param node_id The ID of node to update dynamic resources.
|
||||
|
||||
@@ -148,6 +148,16 @@ std::string GlobalStateAccessor::GetNodeResourceInfo(const NodeID &node_id) {
|
||||
return node_resource_map.SerializeAsString();
|
||||
}
|
||||
|
||||
std::vector<std::string> GlobalStateAccessor::GetAllAvailableResources() {
|
||||
std::vector<std::string> available_resources;
|
||||
std::promise<bool> promise;
|
||||
RAY_CHECK_OK(gcs_client_->Nodes().AsyncGetAllAvailableResources(
|
||||
TransformForMultiItemCallback<rpc::AvailableResources>(available_resources,
|
||||
promise)));
|
||||
promise.get_future().get();
|
||||
return available_resources;
|
||||
}
|
||||
|
||||
std::string GlobalStateAccessor::GetInternalConfig() {
|
||||
rpc::StoredConfig config_proto;
|
||||
std::promise<void> promise;
|
||||
|
||||
@@ -86,6 +86,13 @@ class GlobalStateAccessor {
|
||||
/// deserialized with protobuf function.
|
||||
std::string GetNodeResourceInfo(const NodeID &node_id);
|
||||
|
||||
/// Get available resources of all nodes.
|
||||
///
|
||||
/// \return available resources of all nodes. To support multi-language, we serialize
|
||||
/// each AvailableResources and return the serialized string. Where used, it needs to be
|
||||
/// deserialized with protobuf function.
|
||||
std::vector<std::string> GetAllAvailableResources();
|
||||
|
||||
/// Get internal config from GCS Service.
|
||||
///
|
||||
/// \return map of internal config keys and values. It is stored as a StoredConfig proto
|
||||
|
||||
@@ -638,6 +638,21 @@ Status ServiceBasedNodeInfoAccessor::AsyncGetResources(
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
Status ServiceBasedNodeInfoAccessor::AsyncGetAllAvailableResources(
|
||||
const MultiItemCallback<rpc::AvailableResources> &callback) {
|
||||
rpc::GetAllAvailableResourcesRequest request;
|
||||
client_impl_->GetGcsRpcClient().GetAllAvailableResources(
|
||||
request,
|
||||
[callback](const Status &status, const rpc::GetAllAvailableResourcesReply &reply) {
|
||||
std::vector<rpc::AvailableResources> result =
|
||||
VectorFromProtobuf(reply.resources_list());
|
||||
callback(status, result);
|
||||
RAY_LOG(DEBUG) << "Finished getting available resources of all nodes, status = "
|
||||
<< status;
|
||||
});
|
||||
return Status::OK();
|
||||
}
|
||||
|
||||
Status ServiceBasedNodeInfoAccessor::AsyncUpdateResources(
|
||||
const NodeID &node_id, const ResourceMap &resources, const StatusCallback &callback) {
|
||||
RAY_LOG(DEBUG) << "Updating node resources, node id = " << node_id;
|
||||
|
||||
@@ -182,6 +182,9 @@ class ServiceBasedNodeInfoAccessor : public NodeInfoAccessor {
|
||||
Status AsyncGetResources(const NodeID &node_id,
|
||||
const OptionalItemCallback<ResourceMap> &callback) override;
|
||||
|
||||
Status AsyncGetAllAvailableResources(
|
||||
const MultiItemCallback<rpc::AvailableResources> &callback) override;
|
||||
|
||||
Status AsyncUpdateResources(const NodeID &node_id, const ResourceMap &resources,
|
||||
const StatusCallback &callback) override;
|
||||
|
||||
|
||||
@@ -331,6 +331,20 @@ class ServiceBasedGcsClientTest : public ::testing::Test {
|
||||
return WaitReady(promise.get_future(), timeout_ms_);
|
||||
}
|
||||
|
||||
std::vector<rpc::AvailableResources> GetAllAvailableResources() {
|
||||
std::promise<bool> promise;
|
||||
std::vector<rpc::AvailableResources> resources;
|
||||
RAY_CHECK_OK(gcs_client_->Nodes().AsyncGetAllAvailableResources(
|
||||
[&resources, &promise](Status status,
|
||||
const std::vector<rpc::AvailableResources> &result) {
|
||||
EXPECT_TRUE(!result.empty());
|
||||
resources.assign(result.begin(), result.end());
|
||||
promise.set_value(status.ok());
|
||||
}));
|
||||
EXPECT_TRUE(WaitReady(promise.get_future(), timeout_ms_));
|
||||
return resources;
|
||||
}
|
||||
|
||||
bool SubscribeTask(
|
||||
const TaskID &task_id,
|
||||
const gcs::SubscribeCallback<TaskID, rpc::TaskTableData> &subscribe) {
|
||||
@@ -732,6 +746,38 @@ TEST_F(ServiceBasedGcsClientTest, TestNodeHeartbeat) {
|
||||
WaitForExpectedCount(heartbeat_batch_count, 1);
|
||||
}
|
||||
|
||||
TEST_F(ServiceBasedGcsClientTest, TestGetAllAvailableResources) {
|
||||
// Subscribe batched state of all nodes from GCS.
|
||||
std::atomic<int> heartbeat_batch_count(0);
|
||||
auto on_subscribe =
|
||||
[&heartbeat_batch_count](const gcs::HeartbeatBatchTableData &result) {
|
||||
++heartbeat_batch_count;
|
||||
};
|
||||
ASSERT_TRUE(SubscribeBatchHeartbeat(on_subscribe));
|
||||
|
||||
// Register node.
|
||||
auto node_info = Mocker::GenNodeInfo();
|
||||
RAY_CHECK(RegisterNode(*node_info));
|
||||
|
||||
// Report heartbeat of a node to GCS.
|
||||
NodeID node_id = NodeID::FromBinary(node_info->node_id());
|
||||
auto heartbeat = std::make_shared<rpc::HeartbeatTableData>();
|
||||
heartbeat->set_client_id(node_id.Binary());
|
||||
// Set this flag because GCS won't publish unchanged heartbeat.
|
||||
heartbeat->set_should_global_gc(true);
|
||||
(*heartbeat->mutable_resources_available())["CPU"] = 1.0;
|
||||
(*heartbeat->mutable_resources_available())["GPU"] = 10.0;
|
||||
ASSERT_TRUE(ReportHeartbeat(heartbeat));
|
||||
WaitForExpectedCount(heartbeat_batch_count, 1);
|
||||
|
||||
// Assert get all available resources right.
|
||||
std::vector<rpc::AvailableResources> resources = GetAllAvailableResources();
|
||||
EXPECT_EQ(resources.size(), 1);
|
||||
EXPECT_EQ(resources[0].resources_available_size(), 2);
|
||||
EXPECT_EQ((*resources[0].mutable_resources_available())["CPU"], 1.0);
|
||||
EXPECT_EQ((*resources[0].mutable_resources_available())["GPU"], 10.0);
|
||||
}
|
||||
|
||||
TEST_F(ServiceBasedGcsClientTest, TestTaskInfo) {
|
||||
JobID job_id = JobID::FromInt(1);
|
||||
TaskID task_id = TaskID::ForDriverTask(job_id);
|
||||
|
||||
@@ -354,6 +354,21 @@ void GcsNodeManager::HandleGetInternalConfig(const rpc::GetInternalConfigRequest
|
||||
gcs_table_storage_->InternalConfigTable().Get(UniqueID::Nil(), get_system_config));
|
||||
}
|
||||
|
||||
void GcsNodeManager::HandleGetAllAvailableResources(
|
||||
const rpc::GetAllAvailableResourcesRequest &request,
|
||||
rpc::GetAllAvailableResourcesReply *reply,
|
||||
rpc::SendReplyCallback send_reply_callback) {
|
||||
for (const auto &iter : GetClusterRealtimeResources()) {
|
||||
rpc::AvailableResources resource;
|
||||
resource.set_node_id(iter.first.Binary());
|
||||
for (auto res : iter.second->GetResourceAmountMap()) {
|
||||
(*resource.mutable_resources_available())[res.first] = res.second.ToDouble();
|
||||
}
|
||||
reply->add_resources_list()->CopyFrom(resource);
|
||||
}
|
||||
GCS_RPC_SEND_REPLY(send_reply_callback, reply, Status::OK());
|
||||
}
|
||||
|
||||
std::shared_ptr<rpc::GcsNodeInfo> GcsNodeManager::GetNode(
|
||||
const ray::NodeID &node_id) const {
|
||||
auto iter = alive_nodes_.find(node_id);
|
||||
|
||||
@@ -79,16 +79,22 @@ class GcsNodeManager : public rpc::NodeInfoHandler {
|
||||
rpc::DeleteResourcesReply *reply,
|
||||
rpc::SendReplyCallback send_reply_callback) override;
|
||||
|
||||
/// Handle setting internal config.
|
||||
/// Handle set internal config.
|
||||
void HandleSetInternalConfig(const rpc::SetInternalConfigRequest &request,
|
||||
rpc::SetInternalConfigReply *reply,
|
||||
rpc::SendReplyCallback send_reply_callback) override;
|
||||
|
||||
/// Handle getting internal config.
|
||||
/// Handle get internal config.
|
||||
void HandleGetInternalConfig(const rpc::GetInternalConfigRequest &request,
|
||||
rpc::GetInternalConfigReply *reply,
|
||||
rpc::SendReplyCallback send_reply_callback) override;
|
||||
|
||||
/// Handle get available resources of all nodes.
|
||||
void HandleGetAllAvailableResources(
|
||||
const rpc::GetAllAvailableResourcesRequest &request,
|
||||
rpc::GetAllAvailableResourcesReply *reply,
|
||||
rpc::SendReplyCallback send_reply_callback) override;
|
||||
|
||||
/// Add an alive node.
|
||||
///
|
||||
/// \param node The info of the node to be added.
|
||||
|
||||
@@ -356,6 +356,11 @@ class RedisNodeInfoAccessor : public NodeInfoAccessor {
|
||||
Status AsyncGetResources(const NodeID &node_id,
|
||||
const OptionalItemCallback<ResourceMap> &callback) override;
|
||||
|
||||
Status AsyncGetAllAvailableResources(
|
||||
const MultiItemCallback<rpc::AvailableResources> &callback) override {
|
||||
return Status::NotImplemented("AsyncGetAllAvailableResources not implemented");
|
||||
}
|
||||
|
||||
Status AsyncUpdateResources(const NodeID &node_id, const ResourceMap &resources,
|
||||
const StatusCallback &callback) override;
|
||||
|
||||
|
||||
@@ -222,6 +222,13 @@ message ResourceTableData {
|
||||
double resource_capacity = 1;
|
||||
}
|
||||
|
||||
message AvailableResources {
|
||||
// Node manager client id.
|
||||
bytes node_id = 1;
|
||||
// Resource capacity currently available on this node manager.
|
||||
map<string, double> resources_available = 2;
|
||||
}
|
||||
|
||||
message GcsNodeInfo {
|
||||
// State of a node.
|
||||
enum GcsNodeState {
|
||||
|
||||
@@ -263,6 +263,14 @@ message GetInternalConfigReply {
|
||||
StoredConfig config = 2;
|
||||
}
|
||||
|
||||
message GetAllAvailableResourcesRequest {
|
||||
}
|
||||
|
||||
message GetAllAvailableResourcesReply {
|
||||
GcsStatus status = 1;
|
||||
repeated AvailableResources resources_list = 2;
|
||||
}
|
||||
|
||||
// Service for node info access.
|
||||
service NodeInfoGcsService {
|
||||
// Register a node to GCS Service.
|
||||
@@ -283,6 +291,9 @@ service NodeInfoGcsService {
|
||||
rpc SetInternalConfig(SetInternalConfigRequest) returns (SetInternalConfigReply);
|
||||
// Get cluster internal config.
|
||||
rpc GetInternalConfig(GetInternalConfigRequest) returns (GetInternalConfigReply);
|
||||
// Get available resources of all nodes.
|
||||
rpc GetAllAvailableResources(GetAllAvailableResourcesRequest)
|
||||
returns (GetAllAvailableResourcesReply);
|
||||
}
|
||||
|
||||
message GetObjectLocationsRequest {
|
||||
|
||||
@@ -190,6 +190,10 @@ class GcsRpcClient {
|
||||
VOID_GCS_RPC_CLIENT_METHOD(NodeInfoGcsService, GetInternalConfig,
|
||||
node_info_grpc_client_, )
|
||||
|
||||
/// Get available resources of all nodes from the GCS Service.
|
||||
VOID_GCS_RPC_CLIENT_METHOD(NodeInfoGcsService, GetAllAvailableResources,
|
||||
node_info_grpc_client_, )
|
||||
|
||||
/// Get object's locations from GCS Service.
|
||||
VOID_GCS_RPC_CLIENT_METHOD(ObjectInfoGcsService, GetObjectLocations,
|
||||
object_info_grpc_client_, )
|
||||
|
||||
@@ -215,6 +215,11 @@ class NodeInfoGcsServiceHandler {
|
||||
virtual void HandleGetInternalConfig(const GetInternalConfigRequest &request,
|
||||
GetInternalConfigReply *reply,
|
||||
SendReplyCallback send_reply_callback) = 0;
|
||||
|
||||
virtual void HandleGetAllAvailableResources(
|
||||
const rpc::GetAllAvailableResourcesRequest &request,
|
||||
rpc::GetAllAvailableResourcesReply *reply,
|
||||
rpc::SendReplyCallback send_reply_callback) = 0;
|
||||
};
|
||||
|
||||
/// The `GrpcService` for `NodeInfoGcsService`.
|
||||
@@ -242,6 +247,7 @@ class NodeInfoGrpcService : public GrpcService {
|
||||
NODE_INFO_SERVICE_RPC_HANDLER(DeleteResources);
|
||||
NODE_INFO_SERVICE_RPC_HANDLER(SetInternalConfig);
|
||||
NODE_INFO_SERVICE_RPC_HANDLER(GetInternalConfig);
|
||||
NODE_INFO_SERVICE_RPC_HANDLER(GetAllAvailableResources);
|
||||
}
|
||||
|
||||
private:
|
||||
|
||||
Reference in New Issue
Block a user