# frozen_string_literal: true # Copyright 2020 Google LLC # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # https://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # Auto-generated by gapic-generator-ruby. DO NOT EDIT! require "google/cloud/errors" require "google/cloud/bigquery/storage/v1/storage_pb" module Google module Cloud module Bigquery module Storage module V1 module BigQueryRead ## # Client for the BigQueryRead service. # # BigQuery Read API. # # The Read API can be used to read data from BigQuery. # class Client # @private DEFAULT_ENDPOINT_TEMPLATE = "bigquerystorage.$UNIVERSE_DOMAIN$" include Paths # @private attr_reader :big_query_read_stub ## # Configure the BigQueryRead Client class. # # See {::Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Client::Configuration} # for a description of the configuration fields. # # @example # # # Modify the configuration for all BigQueryRead clients # ::Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Client.configure do |config| # config.timeout = 10.0 # end # # @yield [config] Configure the Client client. # @yieldparam config [Client::Configuration] # # @return [Client::Configuration] # def self.configure @configure ||= begin namespace = ["Google", "Cloud", "Bigquery", "Storage", "V1"] parent_config = while namespace.any? parent_name = namespace.join "::" parent_const = const_get parent_name break parent_const.configure if parent_const.respond_to? :configure namespace.pop end default_config = Client::Configuration.new parent_config default_config.rpcs.create_read_session.timeout = 600.0 default_config.rpcs.create_read_session.retry_policy = { initial_delay: 0.1, max_delay: 60.0, multiplier: 1.3, retry_codes: [4, 14] } default_config.rpcs.read_rows.timeout = 86_400.0 default_config.rpcs.read_rows.retry_policy = { initial_delay: 0.1, max_delay: 60.0, multiplier: 1.3, retry_codes: [14] } default_config.rpcs.split_read_stream.timeout = 600.0 default_config.rpcs.split_read_stream.retry_policy = { initial_delay: 0.1, max_delay: 60.0, multiplier: 1.3, retry_codes: [4, 14] } default_config end yield @configure if block_given? @configure end ## # Configure the BigQueryRead Client instance. # # The configuration is set to the derived mode, meaning that values can be changed, # but structural changes (adding new fields, etc.) are not allowed. Structural changes # should be made on {Client.configure}. # # See {::Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Client::Configuration} # for a description of the configuration fields. # # @yield [config] Configure the Client client. # @yieldparam config [Client::Configuration] # # @return [Client::Configuration] # def configure yield @config if block_given? @config end ## # The effective universe domain # # @return [String] # def universe_domain @big_query_read_stub.universe_domain end ## # Create a new BigQueryRead client object. # # @example # # # Create a client using the default configuration # client = ::Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Client.new # # # Create a client using a custom configuration # client = ::Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Client.new do |config| # config.timeout = 10.0 # end # # @yield [config] Configure the BigQueryRead client. # @yieldparam config [Client::Configuration] # def initialize # These require statements are intentionally placed here to initialize # the gRPC module only when it's required. # See https://github.com/googleapis/toolkit/issues/446 require "gapic/grpc" require "google/cloud/bigquery/storage/v1/storage_services_pb" # Create the configuration object @config = Configuration.new Client.configure # Yield the configuration if needed yield @config if block_given? # Create credentials credentials = @config.credentials # Use self-signed JWT if the endpoint is unchanged from default, # but only if the default endpoint does not have a region prefix. enable_self_signed_jwt = @config.endpoint.nil? || (@config.endpoint == Configuration::DEFAULT_ENDPOINT && !@config.endpoint.split(".").first.include?("-")) credentials ||= Credentials.default scope: @config.scope, enable_self_signed_jwt: enable_self_signed_jwt if credentials.is_a?(::String) || credentials.is_a?(::Hash) credentials = Credentials.new credentials, scope: @config.scope end @quota_project_id = @config.quota_project @quota_project_id ||= credentials.quota_project_id if credentials.respond_to? :quota_project_id @big_query_read_stub = ::Gapic::ServiceStub.new( ::Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Stub, credentials: credentials, endpoint: @config.endpoint, endpoint_template: DEFAULT_ENDPOINT_TEMPLATE, universe_domain: @config.universe_domain, channel_args: @config.channel_args, interceptors: @config.interceptors, channel_pool_config: @config.channel_pool ) end # Service calls ## # Creates a new read session. A read session divides the contents of a # BigQuery table into one or more streams, which can then be used to read # data from the table. The read session also specifies properties of the # data to be read, such as a list of columns or a push-down filter describing # the rows to be returned. # # A particular row can be read by at most one stream. When the caller has # reached the end of each stream in the session, then all the data in the # table has been read. # # Data is assigned to each stream such that roughly the same number of # rows can be read from each stream. Because the server-side unit for # assigning data is collections of rows, the API does not guarantee that # each stream will return the same number or rows. Additionally, the # limits are enforced based on the number of pre-filtered rows, so some # filters can lead to lopsided assignments. # # Read sessions automatically expire 6 hours after they are created and do # not require manual clean-up by the caller. # # @overload create_read_session(request, options = nil) # Pass arguments to `create_read_session` via a request object, either of type # {::Google::Cloud::Bigquery::Storage::V1::CreateReadSessionRequest} or an equivalent Hash. # # @param request [::Google::Cloud::Bigquery::Storage::V1::CreateReadSessionRequest, ::Hash] # A request object representing the call parameters. Required. To specify no # parameters, or to keep all the default parameter values, pass an empty Hash. # @param options [::Gapic::CallOptions, ::Hash] # Overrides the default settings for this call, e.g, timeout, retries, etc. Optional. # # @overload create_read_session(parent: nil, read_session: nil, max_stream_count: nil, preferred_min_stream_count: nil) # Pass arguments to `create_read_session` via keyword arguments. Note that at # least one keyword argument is required. To specify no parameters, or to keep all # the default parameter values, pass an empty Hash as a request object (see above). # # @param parent [::String] # Required. The request project that owns the session, in the form of # `projects/{project_id}`. # @param read_session [::Google::Cloud::Bigquery::Storage::V1::ReadSession, ::Hash] # Required. Session to be created. # @param max_stream_count [::Integer] # Max initial number of streams. If unset or zero, the server will # provide a value of streams so as to produce reasonable throughput. Must be # non-negative. The number of streams may be lower than the requested number, # depending on the amount parallelism that is reasonable for the table. # There is a default system max limit of 1,000. # # This must be greater than or equal to preferred_min_stream_count. # Typically, clients should either leave this unset to let the system to # determine an upper bound OR set this a size for the maximum "units of work" # it can gracefully handle. # @param preferred_min_stream_count [::Integer] # The minimum preferred stream count. This parameter can be used to inform # the service that there is a desired lower bound on the number of streams. # This is typically a target parallelism of the client (e.g. a Spark # cluster with N-workers would set this to a low multiple of N to ensure # good cluster utilization). # # The system will make a best effort to provide at least this number of # streams, but in some cases might provide less. # # @yield [response, operation] Access the result along with the RPC operation # @yieldparam response [::Google::Cloud::Bigquery::Storage::V1::ReadSession] # @yieldparam operation [::GRPC::ActiveCall::Operation] # # @return [::Google::Cloud::Bigquery::Storage::V1::ReadSession] # # @raise [::Google::Cloud::Error] if the RPC is aborted. # # @example Basic example # require "google/cloud/bigquery/storage/v1" # # # Create a client object. The client can be reused for multiple calls. # client = Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Client.new # # # Create a request. To set request fields, pass in keyword arguments. # request = Google::Cloud::Bigquery::Storage::V1::CreateReadSessionRequest.new # # # Call the create_read_session method. # result = client.create_read_session request # # # The returned object is of type Google::Cloud::Bigquery::Storage::V1::ReadSession. # p result # def create_read_session request, options = nil raise ::ArgumentError, "request must be provided" if request.nil? request = ::Gapic::Protobuf.coerce request, to: ::Google::Cloud::Bigquery::Storage::V1::CreateReadSessionRequest # Converts hash and nil to an options object options = ::Gapic::CallOptions.new(**options.to_h) if options.respond_to? :to_h # Customize the options with defaults metadata = @config.rpcs.create_read_session.metadata.to_h # Set x-goog-api-client and x-goog-user-project headers metadata[:"x-goog-api-client"] ||= ::Gapic::Headers.x_goog_api_client \ lib_name: @config.lib_name, lib_version: @config.lib_version, gapic_version: ::Google::Cloud::Bigquery::Storage::V1::VERSION metadata[:"x-goog-user-project"] = @quota_project_id if @quota_project_id header_params = {} if request.read_session&.table header_params["read_session.table"] = request.read_session.table end request_params_header = header_params.map { |k, v| "#{k}=#{v}" }.join("&") metadata[:"x-goog-request-params"] ||= request_params_header options.apply_defaults timeout: @config.rpcs.create_read_session.timeout, metadata: metadata, retry_policy: @config.rpcs.create_read_session.retry_policy options.apply_defaults timeout: @config.timeout, metadata: @config.metadata, retry_policy: @config.retry_policy @big_query_read_stub.call_rpc :create_read_session, request, options: options do |response, operation| yield response, operation if block_given? return response end rescue ::GRPC::BadStatus => e raise ::Google::Cloud::Error.from_error(e) end ## # Reads rows from the stream in the format prescribed by the ReadSession. # Each response contains one or more table rows, up to a maximum of 100 MiB # per response; read requests which attempt to read individual rows larger # than 100 MiB will fail. # # Each request also returns a set of stream statistics reflecting the current # state of the stream. # # @overload read_rows(request, options = nil) # Pass arguments to `read_rows` via a request object, either of type # {::Google::Cloud::Bigquery::Storage::V1::ReadRowsRequest} or an equivalent Hash. # # @param request [::Google::Cloud::Bigquery::Storage::V1::ReadRowsRequest, ::Hash] # A request object representing the call parameters. Required. To specify no # parameters, or to keep all the default parameter values, pass an empty Hash. # @param options [::Gapic::CallOptions, ::Hash] # Overrides the default settings for this call, e.g, timeout, retries, etc. Optional. # # @overload read_rows(read_stream: nil, offset: nil) # Pass arguments to `read_rows` via keyword arguments. Note that at # least one keyword argument is required. To specify no parameters, or to keep all # the default parameter values, pass an empty Hash as a request object (see above). # # @param read_stream [::String] # Required. Stream to read rows from. # @param offset [::Integer] # The offset requested must be less than the last row read from Read. # Requesting a larger offset is undefined. If not specified, start reading # from offset zero. # # @yield [response, operation] Access the result along with the RPC operation # @yieldparam response [::Enumerable<::Google::Cloud::Bigquery::Storage::V1::ReadRowsResponse>] # @yieldparam operation [::GRPC::ActiveCall::Operation] # # @return [::Enumerable<::Google::Cloud::Bigquery::Storage::V1::ReadRowsResponse>] # # @raise [::Google::Cloud::Error] if the RPC is aborted. # # @example Basic example # require "google/cloud/bigquery/storage/v1" # # # Create a client object. The client can be reused for multiple calls. # client = Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Client.new # # # Create a request. To set request fields, pass in keyword arguments. # request = Google::Cloud::Bigquery::Storage::V1::ReadRowsRequest.new # # # Call the read_rows method to start streaming. # output = client.read_rows request # # # The returned object is a streamed enumerable yielding elements of type # # ::Google::Cloud::Bigquery::Storage::V1::ReadRowsResponse # output.each do |current_response| # p current_response # end # def read_rows request, options = nil raise ::ArgumentError, "request must be provided" if request.nil? request = ::Gapic::Protobuf.coerce request, to: ::Google::Cloud::Bigquery::Storage::V1::ReadRowsRequest # Converts hash and nil to an options object options = ::Gapic::CallOptions.new(**options.to_h) if options.respond_to? :to_h # Customize the options with defaults metadata = @config.rpcs.read_rows.metadata.to_h # Set x-goog-api-client and x-goog-user-project headers metadata[:"x-goog-api-client"] ||= ::Gapic::Headers.x_goog_api_client \ lib_name: @config.lib_name, lib_version: @config.lib_version, gapic_version: ::Google::Cloud::Bigquery::Storage::V1::VERSION metadata[:"x-goog-user-project"] = @quota_project_id if @quota_project_id header_params = {} if request.read_stream header_params["read_stream"] = request.read_stream end request_params_header = header_params.map { |k, v| "#{k}=#{v}" }.join("&") metadata[:"x-goog-request-params"] ||= request_params_header options.apply_defaults timeout: @config.rpcs.read_rows.timeout, metadata: metadata, retry_policy: @config.rpcs.read_rows.retry_policy options.apply_defaults timeout: @config.timeout, metadata: @config.metadata, retry_policy: @config.retry_policy @big_query_read_stub.call_rpc :read_rows, request, options: options do |response, operation| yield response, operation if block_given? return response end rescue ::GRPC::BadStatus => e raise ::Google::Cloud::Error.from_error(e) end ## # Splits a given `ReadStream` into two `ReadStream` objects. These # `ReadStream` objects are referred to as the primary and the residual # streams of the split. The original `ReadStream` can still be read from in # the same manner as before. Both of the returned `ReadStream` objects can # also be read from, and the rows returned by both child streams will be # the same as the rows read from the original stream. # # Moreover, the two child streams will be allocated back-to-back in the # original `ReadStream`. Concretely, it is guaranteed that for streams # original, primary, and residual, that original[0-j] = primary[0-j] and # original[j-n] = residual[0-m] once the streams have been read to # completion. # # @overload split_read_stream(request, options = nil) # Pass arguments to `split_read_stream` via a request object, either of type # {::Google::Cloud::Bigquery::Storage::V1::SplitReadStreamRequest} or an equivalent Hash. # # @param request [::Google::Cloud::Bigquery::Storage::V1::SplitReadStreamRequest, ::Hash] # A request object representing the call parameters. Required. To specify no # parameters, or to keep all the default parameter values, pass an empty Hash. # @param options [::Gapic::CallOptions, ::Hash] # Overrides the default settings for this call, e.g, timeout, retries, etc. Optional. # # @overload split_read_stream(name: nil, fraction: nil) # Pass arguments to `split_read_stream` via keyword arguments. Note that at # least one keyword argument is required. To specify no parameters, or to keep all # the default parameter values, pass an empty Hash as a request object (see above). # # @param name [::String] # Required. Name of the stream to split. # @param fraction [::Float] # A value in the range (0.0, 1.0) that specifies the fractional point at # which the original stream should be split. The actual split point is # evaluated on pre-filtered rows, so if a filter is provided, then there is # no guarantee that the division of the rows between the new child streams # will be proportional to this fractional value. Additionally, because the # server-side unit for assigning data is collections of rows, this fraction # will always map to a data storage boundary on the server side. # # @yield [response, operation] Access the result along with the RPC operation # @yieldparam response [::Google::Cloud::Bigquery::Storage::V1::SplitReadStreamResponse] # @yieldparam operation [::GRPC::ActiveCall::Operation] # # @return [::Google::Cloud::Bigquery::Storage::V1::SplitReadStreamResponse] # # @raise [::Google::Cloud::Error] if the RPC is aborted. # # @example Basic example # require "google/cloud/bigquery/storage/v1" # # # Create a client object. The client can be reused for multiple calls. # client = Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Client.new # # # Create a request. To set request fields, pass in keyword arguments. # request = Google::Cloud::Bigquery::Storage::V1::SplitReadStreamRequest.new # # # Call the split_read_stream method. # result = client.split_read_stream request # # # The returned object is of type Google::Cloud::Bigquery::Storage::V1::SplitReadStreamResponse. # p result # def split_read_stream request, options = nil raise ::ArgumentError, "request must be provided" if request.nil? request = ::Gapic::Protobuf.coerce request, to: ::Google::Cloud::Bigquery::Storage::V1::SplitReadStreamRequest # Converts hash and nil to an options object options = ::Gapic::CallOptions.new(**options.to_h) if options.respond_to? :to_h # Customize the options with defaults metadata = @config.rpcs.split_read_stream.metadata.to_h # Set x-goog-api-client and x-goog-user-project headers metadata[:"x-goog-api-client"] ||= ::Gapic::Headers.x_goog_api_client \ lib_name: @config.lib_name, lib_version: @config.lib_version, gapic_version: ::Google::Cloud::Bigquery::Storage::V1::VERSION metadata[:"x-goog-user-project"] = @quota_project_id if @quota_project_id header_params = {} if request.name header_params["name"] = request.name end request_params_header = header_params.map { |k, v| "#{k}=#{v}" }.join("&") metadata[:"x-goog-request-params"] ||= request_params_header options.apply_defaults timeout: @config.rpcs.split_read_stream.timeout, metadata: metadata, retry_policy: @config.rpcs.split_read_stream.retry_policy options.apply_defaults timeout: @config.timeout, metadata: @config.metadata, retry_policy: @config.retry_policy @big_query_read_stub.call_rpc :split_read_stream, request, options: options do |response, operation| yield response, operation if block_given? return response end rescue ::GRPC::BadStatus => e raise ::Google::Cloud::Error.from_error(e) end ## # Configuration class for the BigQueryRead API. # # This class represents the configuration for BigQueryRead, # providing control over timeouts, retry behavior, logging, transport # parameters, and other low-level controls. Certain parameters can also be # applied individually to specific RPCs. See # {::Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Client::Configuration::Rpcs} # for a list of RPCs that can be configured independently. # # Configuration can be applied globally to all clients, or to a single client # on construction. # # @example # # # Modify the global config, setting the timeout for # # create_read_session to 20 seconds, # # and all remaining timeouts to 10 seconds. # ::Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Client.configure do |config| # config.timeout = 10.0 # config.rpcs.create_read_session.timeout = 20.0 # end # # # Apply the above configuration only to a new client. # client = ::Google::Cloud::Bigquery::Storage::V1::BigQueryRead::Client.new do |config| # config.timeout = 10.0 # config.rpcs.create_read_session.timeout = 20.0 # end # # @!attribute [rw] endpoint # A custom service endpoint, as a hostname or hostname:port. The default is # nil, indicating to use the default endpoint in the current universe domain. # @return [::String,nil] # @!attribute [rw] credentials # Credentials to send with calls. You may provide any of the following types: # * (`String`) The path to a service account key file in JSON format # * (`Hash`) A service account key as a Hash # * (`Google::Auth::Credentials`) A googleauth credentials object # (see the [googleauth docs](https://rubydoc.info/gems/googleauth/Google/Auth/Credentials)) # * (`Signet::OAuth2::Client`) A signet oauth2 client object # (see the [signet docs](https://rubydoc.info/gems/signet/Signet/OAuth2/Client)) # * (`GRPC::Core::Channel`) a gRPC channel with included credentials # * (`GRPC::Core::ChannelCredentials`) a gRPC credentails object # * (`nil`) indicating no credentials # @return [::Object] # @!attribute [rw] scope # The OAuth scopes # @return [::Array<::String>] # @!attribute [rw] lib_name # The library name as recorded in instrumentation and logging # @return [::String] # @!attribute [rw] lib_version # The library version as recorded in instrumentation and logging # @return [::String] # @!attribute [rw] channel_args # Extra parameters passed to the gRPC channel. Note: this is ignored if a # `GRPC::Core::Channel` object is provided as the credential. # @return [::Hash] # @!attribute [rw] interceptors # An array of interceptors that are run before calls are executed. # @return [::Array<::GRPC::ClientInterceptor>] # @!attribute [rw] timeout # The call timeout in seconds. # @return [::Numeric] # @!attribute [rw] metadata # Additional gRPC headers to be sent with the call. # @return [::Hash{::Symbol=>::String}] # @!attribute [rw] retry_policy # The retry policy. The value is a hash with the following keys: # * `:initial_delay` (*type:* `Numeric`) - The initial delay in seconds. # * `:max_delay` (*type:* `Numeric`) - The max delay in seconds. # * `:multiplier` (*type:* `Numeric`) - The incremental backoff multiplier. # * `:retry_codes` (*type:* `Array`) - The error codes that should # trigger a retry. # @return [::Hash] # @!attribute [rw] quota_project # A separate project against which to charge quota. # @return [::String] # @!attribute [rw] universe_domain # The universe domain within which to make requests. This determines the # default endpoint URL. The default value of nil uses the environment # universe (usually the default "googleapis.com" universe). # @return [::String,nil] # class Configuration extend ::Gapic::Config # @private # The endpoint specific to the default "googleapis.com" universe. Deprecated. DEFAULT_ENDPOINT = "bigquerystorage.googleapis.com" config_attr :endpoint, nil, ::String, nil config_attr :credentials, nil do |value| allowed = [::String, ::Hash, ::Proc, ::Symbol, ::Google::Auth::Credentials, ::Signet::OAuth2::Client, nil] allowed += [::GRPC::Core::Channel, ::GRPC::Core::ChannelCredentials] if defined? ::GRPC allowed.any? { |klass| klass === value } end config_attr :scope, nil, ::String, ::Array, nil config_attr :lib_name, nil, ::String, nil config_attr :lib_version, nil, ::String, nil config_attr(:channel_args, { "grpc.service_config_disable_resolution" => 1 }, ::Hash, nil) config_attr :interceptors, nil, ::Array, nil config_attr :timeout, nil, ::Numeric, nil config_attr :metadata, nil, ::Hash, nil config_attr :retry_policy, nil, ::Hash, ::Proc, nil config_attr :quota_project, nil, ::String, nil config_attr :universe_domain, nil, ::String, nil # @private def initialize parent_config = nil @parent_config = parent_config unless parent_config.nil? yield self if block_given? end ## # Configurations for individual RPCs # @return [Rpcs] # def rpcs @rpcs ||= begin parent_rpcs = nil parent_rpcs = @parent_config.rpcs if defined?(@parent_config) && @parent_config.respond_to?(:rpcs) Rpcs.new parent_rpcs end end ## # Configuration for the channel pool # @return [::Gapic::ServiceStub::ChannelPool::Configuration] # def channel_pool @channel_pool ||= ::Gapic::ServiceStub::ChannelPool::Configuration.new end ## # Configuration RPC class for the BigQueryRead API. # # Includes fields providing the configuration for each RPC in this service. # Each configuration object is of type `Gapic::Config::Method` and includes # the following configuration fields: # # * `timeout` (*type:* `Numeric`) - The call timeout in seconds # * `metadata` (*type:* `Hash{Symbol=>String}`) - Additional gRPC headers # * `retry_policy (*type:* `Hash`) - The retry policy. The policy fields # include the following keys: # * `:initial_delay` (*type:* `Numeric`) - The initial delay in seconds. # * `:max_delay` (*type:* `Numeric`) - The max delay in seconds. # * `:multiplier` (*type:* `Numeric`) - The incremental backoff multiplier. # * `:retry_codes` (*type:* `Array`) - The error codes that should # trigger a retry. # class Rpcs ## # RPC-specific configuration for `create_read_session` # @return [::Gapic::Config::Method] # attr_reader :create_read_session ## # RPC-specific configuration for `read_rows` # @return [::Gapic::Config::Method] # attr_reader :read_rows ## # RPC-specific configuration for `split_read_stream` # @return [::Gapic::Config::Method] # attr_reader :split_read_stream # @private def initialize parent_rpcs = nil create_read_session_config = parent_rpcs.create_read_session if parent_rpcs.respond_to? :create_read_session @create_read_session = ::Gapic::Config::Method.new create_read_session_config read_rows_config = parent_rpcs.read_rows if parent_rpcs.respond_to? :read_rows @read_rows = ::Gapic::Config::Method.new read_rows_config split_read_stream_config = parent_rpcs.split_read_stream if parent_rpcs.respond_to? :split_read_stream @split_read_stream = ::Gapic::Config::Method.new split_read_stream_config yield self if block_given? end end end end end end end end end end