pgoutput-client 0.0.0 → 0.2.0

data/lib/pgoutput/client/configuration.rb

@@ -0,0 +1,187 @@
+# frozen_string_literal: true
+
+module Pgoutput
+  module Client
+    # Immutable configuration for a PostgreSQL logical replication stream.
+    #
+    # A configuration describes how `pgoutput-client` should connect to
+    # PostgreSQL and how it should request logical replication from the server.
+    # It deliberately contains transport-level settings only; parsing pgoutput
+    # records and decoding PostgreSQL values belong to downstream layers.
+    #
+    # The object freezes itself and its string/array attributes during
+    # initialization so it can be safely shared by transport components without
+    # defensive copying.
+    #
+    # @example Minimal configuration
+    #   config = Pgoutput::Client::Configuration.new(
+    #     database_url: "postgres://localhost/app",
+    #     slot_name: "cdc_slot",
+    #     publication_names: "app_publication"
+    #   )
+    #
+    # @example Start from a known LSN and request binary values from pgoutput
+    #   config = Pgoutput::Client::Configuration.new(
+    #     database_url: ENV.fetch("DATABASE_URL"),
+    #     slot_name: "cdc_slot",
+    #     publication_names: %w[app_publication],
+    #     start_lsn: "0/16B6C50",
+    #     binary: true
+    #   )
+    #
+    # @api public
+    class Configuration
+      # Fixed logical decoding output plugin.
+      #
+      # @return [String]
+      DEFAULT_PLUGIN = "pgoutput"
+
+      # Default pgoutput protocol version.
+      #
+      # @return [Integer]
+      DEFAULT_PROTO_VERSION = 1
+
+      # Default interval, in seconds, between standby status feedback messages.
+      #
+      # @return [Float]
+      DEFAULT_FEEDBACK_INTERVAL = 10.0
+
+      # @!attribute [r] database_url
+      #   PostgreSQL connection URL.
+      #   @return [String]
+      # @!attribute [r] slot_name
+      #   Logical replication slot name.
+      #   @return [String]
+      # @!attribute [r] publication_names
+      #   Publication names requested from pgoutput.
+      #   @return [Array<String>]
+      # @!attribute [r] start_lsn
+      #   Optional normalized starting LSN.
+      #   @return [String, nil]
+      # @!attribute [r] proto_version
+      #   pgoutput protocol version.
+      #   @return [Integer]
+      # @!attribute [r] binary
+      #   Whether to request binary column values from pgoutput.
+      #   @return [Boolean]
+      # @!attribute [r] messages
+      #   Whether to request logical decoding messages from pgoutput.
+      #   @return [Boolean]
+      # @!attribute [r] auto_create_slot
+      #   Whether the client should create the slot before streaming.
+      #   @return [Boolean]
+      # @!attribute [r] temporary_slot
+      #   Whether a newly created slot should be temporary.
+      #   @return [Boolean]
+      # @!attribute [r] feedback_interval
+      #   Standby feedback interval in seconds.
+      #   @return [Float]
+      attr_reader :database_url,
+                  :slot_name,
+                  :publication_names,
+                  :start_lsn,
+                  :proto_version,
+                  :binary,
+                  :messages,
+                  :auto_create_slot,
+                  :temporary_slot,
+                  :feedback_interval
+
+      # Build and validate a logical replication stream configuration.
+      #
+      # `slot_name` and every publication name are intentionally limited to
+      # simple PostgreSQL identifier-like strings. This keeps command rendering
+      # small and predictable while avoiding quoting rules in this transport
+      # layer.
+      #
+      # Boolean options are normalized with Ruby truthiness. `nil` and `false`
+      # become `false`; all other values become `true`.
+      #
+      # @param database_url [#to_s] PostgreSQL connection URL
+      # @param slot_name [#to_s] logical replication slot name
+      # @param publication_names [Array<#to_s>, #to_s] one or more publication
+      #   names to pass to pgoutput
+      # @param start_lsn [String, Integer, nil] starting LSN as a PostgreSQL LSN
+      #   string, an integer WAL position, or `nil` for `0/0`
+      # @param proto_version [#to_int, #to_s] pgoutput protocol version
+      # @param binary [Object] truthy to request binary column values
+      # @param messages [Object] truthy to request logical decoding messages
+      # @param auto_create_slot [Object] truthy to create the slot before
+      #   starting replication
+      # @param temporary_slot [Object] truthy to create a temporary replication
+      #   slot when `auto_create_slot` is enabled
+      # @param feedback_interval [#to_f, #to_s] seconds between periodic standby
+      #   feedback messages
+      # @return [void]
+      # @raise [ConfigurationError] if publication names are empty or numeric
+      #   settings are invalid
+      # @raise [ArgumentError] if `start_lsn`, `proto_version`, or
+      #   `feedback_interval` cannot be coerced
+      def initialize(database_url:,
+                     slot_name:,
+                     publication_names:,
+                     start_lsn: nil,
+                     proto_version: DEFAULT_PROTO_VERSION,
+                     binary: false,
+                     messages: false,
+                     auto_create_slot: false,
+                     temporary_slot: false,
+                     feedback_interval: DEFAULT_FEEDBACK_INTERVAL)
+        @database_url = String(database_url).freeze
+        @slot_name = validate_identifier(slot_name, "slot_name").freeze
+        @publication_names = Array(publication_names).map do |name|
+          validate_identifier(name, "publication_name").freeze
+        end.freeze
+        @start_lsn = normalize_lsn(start_lsn).freeze
+        @proto_version = Integer(proto_version)
+        @binary = boolean(binary, "binary")
+        @messages = boolean(messages, "messages")
+        @auto_create_slot = boolean(auto_create_slot, "auto_create_slot")
+        @temporary_slot = boolean(temporary_slot, "temporary_slot")
+        @feedback_interval = Float(feedback_interval)
+
+        validate!
+        freeze
+      end
+
+      # Starting LSN to render in `START_REPLICATION`.
+      #
+      # @return [String] normalized LSN string, defaulting to `"0/0"`
+      def start_lsn_string
+        start_lsn || "0/0"
+      end
+
+      private
+
+      def validate!
+        raise ConfigurationError, "publication_names must not be empty" if publication_names.empty?
+        raise ConfigurationError, "proto_version must be positive" unless proto_version.positive?
+        raise ConfigurationError, "feedback_interval must be positive" unless feedback_interval.positive?
+      end
+
+      def normalize_lsn(value)
+        return nil if value.nil?
+        return LSN.format(value) if value.is_a?(Integer)
+
+        LSN.format(LSN.parse(String(value)))
+      end
+
+      def validate_identifier(value, field)
+        string = String(value)
+        unless string.match?(/\A[a-zA-Z_][a-zA-Z0-9_]*\z/)
+          raise ConfigurationError, "#{field} must be a PostgreSQL identifier-like string"
+        end
+
+        string
+      end
+
+      # Boolean type checking helper
+      def boolean(value, name)
+        return true if value == true
+        return false if value == false
+
+        raise ArgumentError, "#{name} must be true or false"
+      end
+    end
+  end
+end

data/lib/pgoutput/client/connection.rb

@@ -0,0 +1,120 @@
+# frozen_string_literal: true
+
+module Pgoutput
+  module Client
+    # Thin wrapper around `PG::Connection` for logical replication operations.
+    #
+    # `Connection` hides the small amount of PostgreSQL driver plumbing needed by
+    # the rest of the transport layer. It opens the connection in replication
+    # mode, renders replication commands through {Commands}, and translates
+    # `PG::Error` exceptions into {ConnectionError}.
+    #
+    # @api private
+    class Connection
+      # Configuration associated with this connection.
+      #
+      # @return [Configuration]
+      attr_reader :configuration
+
+      # Open a PostgreSQL connection in database replication mode.
+      #
+      # @param configuration [Configuration] replication configuration
+      # @return [Connection] wrapper around an open `PG::Connection`
+      # @raise [ConnectionError] if the `pg` gem cannot connect
+      def self.open(configuration)
+        require "pg"
+        connection = PG.connect(configuration.database_url, replication: "database")
+        new(configuration:, pg_connection: connection)
+      rescue PG::Error => e
+        raise ConnectionError, e.message
+      end
+
+      # Build a connection wrapper.
+      #
+      # This constructor is public primarily for tests and alternative connection
+      # factories. Normal callers should use {.open}.
+      #
+      # @param configuration [Configuration] replication configuration
+      # @param pg_connection [PG::Connection] connected PostgreSQL driver object
+      # @return [void]
+      def initialize(configuration:, pg_connection:)
+        @configuration = configuration
+        @pg_connection = pg_connection
+      end
+
+      # Execute PostgreSQL's `IDENTIFY_SYSTEM` replication command.
+      #
+      # @return [PG::Result] server identity result
+      # @raise [ConnectionError] if PostgreSQL rejects the command
+      def identify_system
+        exec("IDENTIFY_SYSTEM")
+      end
+
+      # Create the configured logical replication slot.
+      #
+      # @return [PG::Result] command result
+      # @raise [ConnectionError] if PostgreSQL rejects the command
+      def create_replication_slot
+        exec(Commands.create_replication_slot(configuration))
+      end
+
+      # Drop the configured logical replication slot.
+      #
+      # @return [PG::Result] command result
+      # @raise [ConnectionError] if PostgreSQL rejects the command
+      def drop_replication_slot
+        exec(Commands.drop_replication_slot(configuration))
+      end
+
+      # Start streaming logical replication from the configured slot and LSN.
+      #
+      # @return [PG::Result] command result
+      # @raise [ConnectionError] if PostgreSQL rejects the command
+      def start_replication
+        exec(Commands.start_replication(configuration))
+      end
+
+      # Receive one CopyData payload from the server.
+      #
+      # The call is non-blocking because the underlying `pg` call receives
+      # `false` for its blocking argument. `nil` means no complete CopyData
+      # payload is currently available.
+      #
+      # @return [String, nil] raw CopyData payload or `nil`
+      # @raise [ConnectionError] if receiving fails
+      def get_copy_data # rubocop:disable Naming/AccessorMethodName
+        @pg_connection.get_copy_data(false)
+      rescue PG::Error => e
+        raise ConnectionError, e.message
+      end
+
+      # Send one CopyData payload to the server.
+      #
+      # Used for standby status feedback messages.
+      #
+      # @param payload [String] raw CopyData payload
+      # @return [void]
+      # @raise [ConnectionError] if sending fails
+      def put_copy_data(payload)
+        @pg_connection.put_copy_data(payload)
+      rescue PG::Error => e
+        raise ConnectionError, e.message
+      end
+
+      # Close the PostgreSQL connection if it is still open.
+      #
+      # @return [void]
+      def close
+        @pg_connection.close unless @pg_connection.finished?
+      end
+
+      private
+
+      def exec(sql)
+        @pg_connection.exec(sql)
+      rescue PG::Error => e
+        raise ConnectionError, e.message
+      end
+    end
+  end
+end

data/lib/pgoutput/client/errors.rb

@@ -0,0 +1,41 @@
+# frozen_string_literal: true
+
+module Pgoutput
+  module Client
+    # Base error class for all pgoutput-client failures.
+    #
+    # Rescue this class when callers want to handle any error raised by the
+    # transport layer without also rescuing unrelated Ruby or PostgreSQL driver
+    # exceptions.
+    #
+    # @api public
+    class Error < StandardError; end
+
+    # Raised when stream configuration is invalid.
+    #
+    # Examples include an empty publication list, invalid replication slot name,
+    # invalid publication name, non-positive protocol version, or non-positive
+    # feedback interval.
+    #
+    # @api public
+    class ConfigurationError < Error; end
+
+    # Raised when a replication protocol envelope cannot be parsed.
+    #
+    # This error represents malformed or unexpected CopyData payloads at the
+    # transport-envelope level. It does not describe pgoutput plugin payload
+    # parsing errors; those belong to the parser layer.
+    #
+    # @api public
+    class ProtocolError < Error; end
+
+    # Raised when a PostgreSQL replication connection operation fails.
+    #
+    # `Connection` converts `PG::Error` instances into this error so public
+    # callers do not need to depend on the PostgreSQL driver's exception classes
+    # for transport-level handling.
+    #
+    # @api public
+    class ConnectionError < Error; end
+  end
+end

data/lib/pgoutput/client/feedback.rb

@@ -0,0 +1,62 @@
+# frozen_string_literal: true
+
+module Pgoutput
+  module Client
+    FeedbackData = Data.define(:received_lsn, :flushed_lsn, :applied_lsn, :client_clock, :reply_requested)
+
+    # Standby status feedback message builder.
+    #
+    # Logical replication clients periodically send standby status updates to
+    # tell PostgreSQL which WAL location has been received, flushed, and applied.
+    # `Feedback` models that update and can serialize itself into the CopyData
+    # payload expected by the replication protocol.
+    #
+    # @attr_reader received_lsn [Integer] latest WAL location received by the client
+    # @attr_reader flushed_lsn [Integer] latest WAL location flushed by the client
+    # @attr_reader applied_lsn [Integer] latest WAL location applied by the client
+    # @attr_reader client_clock [Integer] PostgreSQL timestamp in microseconds since 2000-01-01 UTC
+    # @attr_reader reply_requested [Boolean] whether this feedback is responding to an immediate-reply request
+    class Feedback < FeedbackData
+      # Build feedback using the current wall-clock time.
+      #
+      # By default, flushed and applied LSNs follow the received LSN. Callers can
+      # pass lower values if they need to distinguish receipt from durable flush
+      # or application progress.
+      #
+      # @param received_lsn [Integer] latest WAL location received by the client
+      # @param flushed_lsn [Integer] latest WAL location flushed by the client
+      # @param applied_lsn [Integer] latest WAL location applied by the client
+      # @param reply_requested [Boolean] whether this is an immediate reply
+      # @return [Feedback] immutable feedback value
+      def self.now(received_lsn:, flushed_lsn: received_lsn, applied_lsn: flushed_lsn, reply_requested: false)
+        new(received_lsn, flushed_lsn, applied_lsn, current_pg_time, reply_requested)
+      end
+
+      # Build a protocol CopyData payload for standby status update.
+      #
+      # The payload begins with the standby status update tag `r`, followed by
+      # three unsigned 64-bit LSN fields, the PostgreSQL timestamp, and a
+      # one-byte reply-requested flag.
+      #
+      # @return [String] frozen binary CopyData payload
+      def to_copy_data
+        (
+          "r".b +
+          [received_lsn, flushed_lsn, applied_lsn, client_clock].pack("Q>Q>Q>Q>") +
+          [reply_requested ? 1 : 0].pack("C")
+        ).freeze
+      end
+
+      # Current PostgreSQL protocol timestamp.
+      #
+      # PostgreSQL timestamps in replication messages are expressed as
+      # microseconds since 2000-01-01 00:00:00 UTC, not Unix epoch
+      # microseconds.
+      #
+      # @return [Integer] microseconds since 2000-01-01 UTC
+      def self.current_pg_time
+        ((Time.now.utc - Time.utc(2000, 1, 1)) * 1_000_000).to_i
+      end
+    end
+  end
+end

data/lib/pgoutput/client/keepalive.rb

@@ -0,0 +1,64 @@
+# frozen_string_literal: true
+
+module Pgoutput
+  module Client
+    KeepaliveData = Data.define(:wal_end, :server_clock, :reply_requested)
+
+    # Immutable primary keepalive replication message.
+    #
+    # PostgreSQL sends keepalive CopyData payloads while a replication stream is
+    # active. The payload layout is:
+    #
+    # ```text
+    # Byte 0      : message tag `k`
+    # Bytes 1-8   : current server WAL end, unsigned 64-bit big-endian
+    # Bytes 9-16  : server clock, PostgreSQL timestamp format
+    # Byte 17     : reply-requested flag, 1 for immediate feedback
+    # ```
+    #
+    # The stream layer uses this message to advance its known WAL position and to
+    # decide whether to send standby status feedback immediately.
+    #
+    # @attr_reader wal_end [Integer] latest server WAL position
+    # @attr_reader server_clock [Integer] PostgreSQL server timestamp in
+    #   microseconds since 2000-01-01 UTC
+    # @attr_reader reply_requested [Boolean] whether PostgreSQL requested
+    #   immediate feedback
+    class Keepalive < KeepaliveData
+      # Parse a keepalive CopyData payload.
+      #
+      # @param bytes [String] raw CopyData payload beginning with `k`
+      # @return [Keepalive] immutable parsed keepalive message
+      # @raise [ProtocolError] if the payload is empty, has the wrong message
+      #   tag, or is too short to contain the required fields
+      def self.parse(bytes)
+        binary = bytes.b
+        raise ProtocolError, "empty CopyData payload" if binary.empty?
+        raise ProtocolError, "expected keepalive message" unless binary.getbyte(0) == "k".ord
+        raise ProtocolError, "truncated keepalive message" if binary.bytesize < 18
+
+        wal_end = unpack_u64(binary, 1)
+        server_clock = unpack_u64(binary, 9)
+        reply_requested = binary.getbyte(17) == 1
+
+        Ractor.make_shareable(new(wal_end, server_clock, reply_requested))
+      end
+
+      # Latest server WAL position formatted as a PostgreSQL LSN string.
+      #
+      # @return [String]
+      def wal_end_lsn = LSN.format(wal_end)
+
+      # @param binary [String]
+      # @param offset [Integer]
+      # @return [Integer]
+      def self.unpack_u64(binary, offset)
+        value = binary.byteslice(offset, 8)&.unpack1("Q>")
+        raise ProtocolError, "failed to unpack uint64" unless value.is_a?(Integer)
+
+        value
+      end
+      private_class_method :unpack_u64
+    end
+  end
+end

data/lib/pgoutput/client/lsn.rb

@@ -0,0 +1,54 @@
+# frozen_string_literal: true
+
+module Pgoutput
+  module Client
+    # PostgreSQL Log Sequence Number conversion helpers.
+    #
+    # PostgreSQL represents LSNs as two hexadecimal 32-bit halves separated by a
+    # slash, such as `0/16B6C50`. The replication protocol transmits the same WAL
+    # position as an unsigned 64-bit integer. This module converts between those
+    # two representations.
+    #
+    # @example Parse a textual LSN
+    #   Pgoutput::Client::LSN.parse("0/10")
+    #   # => 16
+    #
+    # @example Format an integer WAL position
+    #   Pgoutput::Client::LSN.format(16)
+    #   # => "0/10"
+    #
+    # @api public
+    module LSN
+      module_function
+
+      # Parse a PostgreSQL LSN string into an integer WAL position.
+      #
+      # @param value [#to_s] LSN string in `HEX/HEX` form
+      # @return [Integer] unsigned 64-bit WAL position
+      # @raise [ArgumentError] if the value is not a valid LSN string
+      def parse(value)
+        high, low = String(value).split("/", 2)
+        raise ArgumentError, "invalid LSN: #{value.inspect}" if high.nil? || low.nil?
+
+        (Integer(high, 16) << 32) + Integer(low, 16)
+      rescue ArgumentError
+        raise ArgumentError, "invalid LSN: #{value.inspect}"
+      end
+
+      # Format an integer WAL position as a PostgreSQL LSN string.
+      #
+      # @param value [#to_int, #to_s] non-negative integer WAL position
+      # @return [String] LSN string in uppercase hexadecimal `HEX/HEX` form
+      # @raise [ArgumentError] if the value is negative or cannot be coerced to
+      #   an integer
+      def format(value)
+        integer = Integer(value)
+        raise ArgumentError, "LSN must be non-negative" if integer.negative?
+
+        high = integer >> 32
+        low = integer & 0xFFFF_FFFF
+        "#{high.to_s(16).upcase}/#{low.to_s(16).upcase}"
+      end
+    end
+  end
+end