fluentd 1.4.2-x86-mingw32 → 1.5.0-x86-mingw32

checksums.yaml

@@ -1,7 +1,7 @@
 ---
-SHA1:
-  metadata.gz: 41eb4fb94871c06b447e09c7d7562b23b12ece33
-  data.tar.gz: 33372302feaa22d3882f79e6ba092ce346f23799
+SHA256:
+  metadata.gz: e0ce49f0618f315773fde35ca2cad351e924cd7adc53e1bc5010eb8e374ee68d
+  data.tar.gz: 32cdc32e738e95a599eea86f94e8ed5182dbea9dbfa3b5d601869bd79eb7e04b
 SHA512:
-  metadata.gz: f69e5bd92b695b0d41fb253969c29ff81d162a6ce813a6740dd50d9f5f8c777da290a8bc8134c1bbeaba50c58ab22eb4c804d7692235c42109f6730cf804c4dc
-  data.tar.gz: 8bf65594d4efff13e1af280fcd59cca3ff4532d5de854e09ddea2e1284a3553ffbb043b2c342de84e38837933e3abb4fb13cff62d730ca73e679b9cffd99889a
+  metadata.gz: 4b49d16b3ad19f09c0765aa264512b18956a878d4b9ce420f56473c38bbde6d518a3f61d818c797c88bc5543f4cdc2276066087ba3d8839c1fa21fde1150f5ae
+  data.tar.gz: 83a8047f7081a7d630f0fefe97f0aaed623dd1c40500035061e9928b6f5c9fc3efeb0133c46c73266df676b1dd73b75cd00bf927fad6b2efacf7d1ff575a1070

data/.travis.yml

@@ -11,14 +11,20 @@ matrix:
       os: linux
     - rvm: 2.2.10
       os: linux
-    - rvm: 2.4.5
+    - rvm: 2.2.10
+      os: linux-ppc64le
+    - rvm: 2.4.6
       os: linux
-    - rvm: 2.5.3
+    - rvm: 2.4.6
+      os: linux-ppc64le
+    - rvm: 2.5.5
       os: linux
-    - rvm: 2.6.0
+    - rvm: 2.6.3
       os: linux
     - rvm: ruby-head
       os: linux
+    - rvm: ruby-head
+      os: linux-ppc64le
     - rvm: 2.1.10
       os: osx
       osx_image: xcode8.3 # OSX 10.12
@@ -28,7 +34,7 @@ matrix:
     # - rvm: 2.3.3
     #   os: osx
     #   osx_image: xcode8.2 # OSX 10.12
-    - rvm: 2.4.1
+    - rvm: 2.4.6
       os: osx
       osx_image: xcode8.3 # OSX 10.12
     - rvm: ruby-head
@@ -40,7 +46,7 @@ matrix:
     - rvm: 2.1.10
       os: osx
       osx_image: xcode8.3
-    - rvm: 2.4.1
+    - rvm: 2.4.6
       os: osx
       osx_image: xcode8.3
     - rvm: ruby-head

data/CHANGELOG.md

@@ -1,3 +1,38 @@
+# v1.5
+
+## Release v1.5.0 - 2019/05/18
+
+### New feature
+
+* out_forward: Support keepalive feature
+  https://github.com/fluent/fluentd/pull/2393
+* in_http: Support TLS via server helper
+  https://github.com/fluent/fluentd/pull/2395
+* in_syslog: Support TLS via server helper
+  https://github.com/fluent/fluentd/pull/2399
+
+### Enhancement
+
+* in_syslog: Add delimiter parameter
+  https://github.com/fluent/fluentd/pull/2378
+* in_forward: Add tag/add_tag_prefix parameters
+  https://github.com/fluent/fluentd/pull/2396
+* parser_json: Add stream_buffer_size parameter for yajl
+  https://github.com/fluent/fluentd/pull/2381
+* command: Add deprecated message to show-plugin-config option
+  https://github.com/fluent/fluentd/pull/2401
+* storage_local: Ignore empty file. Call sync after write for XFS.
+  https://github.com/fluent/fluentd/pull/2409
+
+### Bug fixes
+
+* out_forward: Don't use SO_LINGER on SSL/TLS WinSock
+  https://github.com/fluent/fluentd/pull/2398
+* server helper: Fix recursive lock issue in TLSServer
+  https://github.com/fluent/fluentd/pull/2341
+* Fix typo
+  https://github.com/fluent/fluentd/pull/2369
+
 # v1.4
 
 ## Release v1.4.2 - 2019/04/02

data/README.md

@@ -30,6 +30,11 @@ Mobile/Web Application Logging | Fluentd can function as middleware to enable as
 
 ## Development
 
+### Branch
+
+- master: For v1 development.
+- v0.12: For v0.12. This is security maintenance mode. Only security fix is accepted.
+
 ### Prerequisites
 
 - Ruby 2.1 or later

data/example/in_http.conf

@@ -6,7 +6,9 @@
   keepalive_timeout 10
   # backlog 0
   add_http_headers false
-  format default
+  <parse>
+    @type json
+  </parse>
 </source>
 
 <match test>

data/lib/fluent/command/fluentd.rb

@@ -40,7 +40,7 @@ op.on('--dry-run', "Check fluentd setup is correct or not", TrueClass) {|b|
   opts[:dry_run] = b
 }
 
-op.on('--show-plugin-config=PLUGIN', "Show PLUGIN configuration and exit(ex: input:dummy)") {|plugin|
+op.on('--show-plugin-config=PLUGIN', "[DEPRECATED] Show PLUGIN configuration and exit(ex: input:dummy)") {|plugin|
   opts[:show_plugin_config] = plugin
 }
 

data/lib/fluent/compat/socket_util.rb

@@ -148,7 +148,7 @@ module Fluent
         def on_message(msg, addr)
           @parser.parse(msg) { |time, record|
             unless time && record
-              log.warn "pattern not match: #{msg.inspect}"
+              log.warn "pattern not matched: #{msg.inspect}"
               return
             end
 

data/lib/fluent/log.rb

@@ -148,10 +148,10 @@ module Fluent
     end
 
     attr_reader :format
+    attr_reader :time_format
     attr_accessor :log_event_enabled
     attr_accessor :out
     attr_accessor :level
-    attr_accessor :time_format
     attr_accessor :optional_header, :optional_attrs
 
     def logdev=(logdev)

data/lib/fluent/plugin/filter_parser.rb

@@ -82,7 +82,7 @@ module Fluent::Plugin
             return t, r
           else
             if @emit_invalid_record_to_error
-              router.emit_error_event(tag, time, record, Fluent::Plugin::Parser::ParserError.new("pattern not match with data '#{raw_value}'"))
+              router.emit_error_event(tag, time, record, Fluent::Plugin::Parser::ParserError.new("pattern not matched with data '#{raw_value}'"))
             end
             if @reserve_data
               t = time

data/lib/fluent/plugin/in_forward.rb

@@ -62,6 +62,11 @@ module Fluent::Plugin
     desc "The field name of the client's hostname."
     config_param :source_hostname_key, :string, default: nil
 
+    desc "New tag instead of incoming tag"
+    config_param :tag, :string, default: nil
+    desc "Add prefix to incoming tag"
+    config_param :add_tag_prefix, :string, default: nil
+
     config_section :security, required: false, multi: false do
       desc 'The hostname'
       config_param :self_hostname, :string
@@ -106,6 +111,9 @@ module Fluent::Plugin
       end
       @enable_field_injection = @source_address_key || @source_hostname_key
 
+      raise Fluent::ConfigError, "'tag' parameter must not be empty" if @tag && @tag.empty?
+      raise Fluent::ConfigError, "'add_tag_prefix' parameter must not be empty" if @add_tag_prefix && @add_tag_prefix.empty?
+
       if @security
         if @security.user_auth && @security.users.empty?
           raise Fluent::ConfigError, "<user> sections required if user_auth enabled"
@@ -293,6 +301,9 @@ module Fluent::Plugin
         log.warn "Input chunk size is larger than 'chunk_size_warn_limit':", tag: tag, host: conn.remote_host, limit: @chunk_size_warn_limit, size: chunk_size
       end
 
+      tag = @tag.dup if @tag
+      tag = "#{@add_tag_prefix}.#{tag}" if @add_tag_prefix
+
       case entries
       when String
         # PackedForward

data/lib/fluent/plugin/in_http.rb

@@ -36,9 +36,7 @@ module Fluent::Plugin
   class HttpInput < Input
     Fluent::Plugin.register_input('http', self)
 
-    # TODO: update this plugin implementation to use server plugin helper, after adding keepalive feature on it
-
-    helpers :parser, :compat_parameters, :event_loop
+    helpers :parser, :compat_parameters, :event_loop, :server
 
     EMPTY_GIF_IMAGE = "GIF89a\u0001\u0000\u0001\u0000\x80\xFF\u0000\xFF\xFF\xFF\u0000\u0000\u0000,\u0000\u0000\u0000\u0000\u0001\u0000\u0001\u0000\u0000\u0002\u0002D\u0001\u0000;".force_encoding("UTF-8")
 
@@ -130,28 +128,15 @@ module Fluent::Plugin
 
       log.debug "listening http", bind: @bind, port: @port
 
-      socket_manager_path = ENV['SERVERENGINE_SOCKETMANAGER_PATH']
-      if Fluent.windows?
-        socket_manager_path = socket_manager_path.to_i
-      end
-      client = ServerEngine::SocketManager::Client.new(socket_manager_path)
-      lsock = client.listen_tcp(@bind, @port)
-
       @km = KeepaliveManager.new(@keepalive_timeout)
-      @lsock = Coolio::TCPServer.new(
-        lsock, nil, Handler, @km, method(:on_request),
-        @body_size_limit, @format_name, log,
-        @cors_allow_origins
-      )
-      @lsock.listen(@backlog) unless @backlog.nil?
       event_loop_attach(@km)
-      event_loop_attach(@lsock)
 
+      server_create_connection(:in_http, @port, bind: @bind, backlog: @backlog, &method(:on_server_connect))
       @float_time_parser = Fluent::NumericTimeParser.new(:float)
     end
 
     def close
-      @lsock.close
+      server_wait_until_stop
       super
     end
 
@@ -240,6 +225,22 @@ module Fluent::Plugin
 
     private
 
+    def on_server_connect(conn)
+      handler = Handler.new(conn, @km, method(:on_request), @body_size_limit, @format_name, log, @cors_allow_origins)
+
+      conn.on(:data) do |data|
+        handler.on_read(data)
+      end
+
+      conn.on(:write_complete) do |_|
+        handler.on_write_complete
+      end
+
+      conn.on(:close) do |_|
+        handler.on_close
+      end
+    end
+
     def parse_params_default(params)
       if msgpack = params['msgpack']
         @parser_msgpack.parse(msgpack) do |_time, record|
@@ -265,11 +266,11 @@ module Fluent::Plugin
       end
     end
 
-    class Handler < Coolio::Socket
+    class Handler
       attr_reader :content_type
 
       def initialize(io, km, callback, body_size_limit, format_name, log, cors_allow_origins)
-        super(io)
+        @io = io
         @km = km
         @callback = callback
         @body_size_limit = body_size_limit
@@ -280,7 +281,8 @@ module Fluent::Plugin
         @idle = 0
         @km.add(self)
 
-        @remote_port, @remote_addr = *Socket.unpack_sockaddr_in(io.getpeername) rescue nil
+        @remote_port, @remote_addr = io.remote_port, io.remote_addr
+        @parser = Http::Parser.new(self)
       end
 
       def step_idle
@@ -291,17 +293,13 @@ module Fluent::Plugin
         @km.delete(self)
       end
 
-      def on_connect
-        @parser = Http::Parser.new(self)
-      end
-
       def on_read(data)
         @idle = 0
         @parser << data
       rescue
         @log.warn "unexpected error", error: $!.to_s
         @log.warn_backtrace
-        close
+        @io.close
       end
 
       def on_message_begin
@@ -477,8 +475,12 @@ module Fluent::Plugin
         end
       end
 
+      def close
+        @io.close
+      end
+
       def on_write_complete
-        close if @next_close
+        @io.close if @next_close
       end
 
       def send_response_and_close(code, header, body)
@@ -499,9 +501,9 @@ module Fluent::Plugin
           data << "#{k}: #{v}\r\n"
         }
         data << "\r\n"
-        write data
+        @io.write(data)
 
-        write body
+        @io.write(body)
       end
 
       def send_response_nobody(code, header)
@@ -510,7 +512,7 @@ module Fluent::Plugin
           data << "#{k}: #{v}\r\n"
         }
         data << "\r\n"
-        write data
+        @io.write(data)
       end
 
       def include_cors_allow_origin

data/lib/fluent/plugin/in_syslog.rb

@@ -74,7 +74,7 @@ module Fluent::Plugin
     desc 'The prefix of the tag. The tag itself is generated by the tag prefix, facility level, and priority.'
     config_param :tag, :string
     desc 'The transport protocol used to receive logs.(udp, tcp)'
-    config_param :protocol_type, :enum, list: [:tcp, :udp], default: :udp
+    config_param :protocol_type, :enum, list: [:tcp, :udp], default: nil, deprecated: "use transport directive"
     desc 'The message frame type.(traditional, octet_count)'
     config_param :frame_type, :enum, list: [:traditional, :octet_count], default: :traditional
 
@@ -99,11 +99,19 @@ module Fluent::Plugin
 
     config_param :blocking_timeout, :time, default: 0.5
 
+    desc 'The delimiter value "\n"'
+    config_param :delimiter, :string, default: "\n" # syslog family add "\n" to each message
+
     config_section :parse do
       config_set_default :@type, DEFAULT_PARSER
       config_param :with_priority, :bool, default: true
     end
 
+    # overwrite server plugin to change default to :udp
+    config_section :transport, required: false, multi: false, init: true, param_name: :transport_config do
+      config_argument :protocol, :enum, list: [:tcp, :udp, :tls], default: :udp
+    end
+
     def configure(conf)
       compat_parameters_convert(conf, :parser)
 
@@ -138,12 +146,13 @@ module Fluent::Plugin
     def start
       super
 
-      log.info "listening syslog socket on #{@bind}:#{@port} with #{@protocol_type}"
-      case @protocol_type
+      log.info "listening syslog socket on #{@bind}:#{@port} with #{@protocol_type || @transport_config.protocol}"
+      case @protocol_type || @transport_config.protocol
       when :udp then start_udp_server
       when :tcp then start_tcp_server
+      when :tls then start_tcp_server(tls: true)
       else
-        raise "BUG: invalid protocol_type value:#{@protocol_type}"
+        raise "BUG: invalid transport value: #{@protocol_type || @transport_config.protocol}"
       end
     end
 
@@ -153,13 +162,12 @@ module Fluent::Plugin
       end
     end
 
-    def start_tcp_server
+    def start_tcp_server(tls: false)
       octet_count_frame = @frame_type == :octet_count
 
-      # syslog family adds "\n" to each message when transport is TCP and traditional frame
-      delimiter = octet_count_frame ? " " : "\n"
+      delimiter = octet_count_frame ? " " : @delimiter
       delimiter_size = delimiter.size
-      server_create_connection(:in_syslog_tcp_server, @port, bind: @bind, resolve_name: @resolve_hostname) do |conn|
+      server_create_connection(tls ? :in_syslog_tls_server : :in_syslog_tcp_server, @port, bind: @bind, resolve_name: @resolve_hostname) do |conn|
         conn.data do |data|
           buffer = conn.buffer
           buffer << data

data/lib/fluent/plugin/in_tail.rb

@@ -427,7 +427,7 @@ module Fluent::Plugin
               record[@path_key] ||= tail_watcher.path unless @path_key.nil?
               es.add(Fluent::EventTime.now, record)
             end
-            log.warn "pattern not match: #{line.inspect}"
+            log.warn "pattern not matched: #{line.inspect}"
           end
         }
       rescue => e

data/lib/fluent/plugin/in_tcp.rb

@@ -71,7 +71,7 @@ module Fluent::Plugin
 
             @parser.parse(msg) do |time, record|
               unless time && record
-                log.warn "pattern not match", message: msg
+                log.warn "pattern not matched", message: msg
                 next
               end
 

data/lib/fluent/plugin/in_udp.rb

@@ -74,7 +74,7 @@ module Fluent::Plugin
         begin
           @parser.parse(data) do |time, record|
             unless time && record
-              log.warn "pattern not match", data: data
+              log.warn "pattern not matched", data: data
               next
             end
 

data/lib/fluent/plugin/out_exec_filter.rb

@@ -95,6 +95,7 @@ module Fluent::Plugin
     COMPAT_PARSE_PARAMS = {
       'out_format' => '@type',
       'out_keys' => 'keys',
+      'out_stream_buffer_size' => 'stream_buffer_size',
     }
     COMPAT_EXTRACT_PARAMS = {
       'out_tag_key' => 'tag_key',

data/lib/fluent/plugin/out_forward.rb

@@ -36,7 +36,6 @@ module Fluent::Plugin
     desc 'The transport protocol.'
     config_param :transport, :enum, list: [:tcp, :tls], default: :tcp
     # TODO: TLS session cache/tickets
-    # TODO: Connection keepalive
 
     desc 'The timeout time when sending event logs.'
     config_param :send_timeout, :time, default: 60
@@ -103,6 +102,10 @@ module Fluent::Plugin
     config_param :tls_client_private_key_path, :string, default: nil
     desc 'The client private key passphrase for TLS.'
     config_param :tls_client_private_key_passphrase, :string, default: nil, secret: true
+    desc "Enable keepalive connection."
+    config_param :keepalive, :bool, default: false
+    desc "Expired time of keepalive. Default value is nil, which means to keep connection as long as possible"
+    config_param :keepalive_timeout, :time, default: nil
 
     config_section :security, required: false, multi: false do
       desc 'The hostname'
@@ -151,6 +154,7 @@ module Fluent::Plugin
       @usock = nil
       @sock_ack_waiting = nil
       @sock_ack_waiting_mutex = nil
+      @keep_alive_watcher_interval = 5 # TODO
     end
 
     def configure(conf)
@@ -201,9 +205,9 @@ module Fluent::Plugin
 
         log.info "adding forwarding server '#{name}'", host: server.host, port: server.port, weight: server.weight, plugin_id: plugin_id
         if @heartbeat_type == :none
-          @nodes << NoneHeartbeatNode.new(self, server, failure: failure)
+          @nodes << NoneHeartbeatNode.new(self, server, failure: failure, keepalive: @keepalive, keepalive_timeout: @keepalive_timeout)
         else
-          node = Node.new(self, server, failure: failure)
+          node = Node.new(self, server, failure: failure, keepalive: @keepalive, keepalive_timeout: @keepalive_timeout)
           begin
             node.validate_host_resolution!
           rescue => e
@@ -227,6 +231,10 @@ module Fluent::Plugin
         raise Fluent::ConfigError, "forward output plugin requires at least one <server> is required"
       end
 
+      if !@keepalive && @keepalive_timeout
+        log.warn('The value of keepalive_timeout is ignored. if you want to use keepalive, please add `keepalive true` to your conf.')
+      end
+
       raise Fluent::ConfigError, "ack_response_timeout must be a positive integer" if @ack_response_timeout < 1
     end
 
@@ -279,6 +287,10 @@ module Fluent::Plugin
           end
         end
       end
+
+      if @keepalive && @keepalive_timeout
+        timer_execute(:out_forward_keep_alived_socket_watcher, @keep_alive_watcher_interval, &method(:on_purge_obsolete_socks))
+      end
     end
 
     def close
@@ -286,6 +298,10 @@ module Fluent::Plugin
         # close socket and ignore errors: this socket will not be used anyway.
         @usock.close rescue nil
       end
+
+      if @keepalive && @keepalive_timeout
+        @nodes.each(&:clear)
+      end
       super
     end
 
@@ -354,7 +370,10 @@ module Fluent::Plugin
           cert_path: @tls_client_cert_path,
           private_key_path: @tls_client_private_key_path,
           private_key_passphrase: @tls_client_private_key_passphrase,
-          linger_timeout: @send_timeout,
+
+          # Enabling SO_LINGER causes data loss on Windows
+          # https://github.com/fluent/fluentd/issues/1968
+          linger_timeout: Fluent.windows? ? nil : @send_timeout,
           send_timeout: @send_timeout,
           recv_timeout: @ack_response_timeout,
           &block
@@ -450,6 +469,10 @@ module Fluent::Plugin
       end
     end
 
+    def on_purge_obsolete_socks
+      @nodes.each(&:purge_obsolete_socks)
+    end
+
     # return chunk id to be committed
     def read_ack_from_sock(sock, unpacker)
       begin
@@ -484,8 +507,13 @@ module Fluent::Plugin
       log.error "unexpected error while receiving ack message", error: e
       log.error_backtrace
     ensure
-      info.sock.close_write rescue nil
-      info.sock.close rescue nil
+      if @keepalive
+        info.node.socket_cache.dec_ref_by_value(info.sock)
+      else
+        info.sock.close_write rescue nil
+        info.sock.close rescue nil
+      end
+
       @sock_ack_waiting_mutex.synchronize do
         @sock_ack_waiting.delete(info)
       end
@@ -513,6 +541,9 @@ module Fluent::Plugin
                 # (2) the node does support sending response but responses have not arrived for some reasons.
                 log.warn "no response from node. regard it as unavailable.", host: info.node.host, port: info.node.port
                 info.node.disable!
+                if @keepalive
+                  info.node.socket_cache.revoke_by_value(info.sock)
+                end
                 info.sock.close rescue nil
                 rollback_write(info.chunk_id, update_retry: false)
               else
@@ -538,7 +569,142 @@ module Fluent::Plugin
     end
 
     class Node
-      def initialize(sender, server, failure:)
+      class SocketCache
+        TimedSocket = Struct.new(:timeout, :sock, :ref)
+
+        def initialize(timeout, log)
+          @log = log
+          @timeout = timeout
+          @active_socks = {}
+          @inactive_socks = {}
+          @mutex = Mutex.new
+        end
+
+        def revoke(key = Thread.current.object_id)
+          @mutex.synchronize do
+            if @active_socks[key]
+              @inactive_socks[key] = @active_socks.delete(key)
+              @inactive_socks[key].ref = 0
+            end
+          end
+        end
+
+        def clear
+          @mutex.synchronize do
+            @inactive_socks.values.each do |s|
+              s.sock.close rescue nil
+            end
+            @inactive_socks.clear
+
+            @active_socks.values.each do |s|
+              s.sock.close rescue nil
+            end
+            @active_socks.clear
+          end
+        end
+
+        def purge_obsolete_socks
+          @mutex.synchronize do
+            @inactive_socks.keys.each do |k|
+              # 0 means sockets stored in this class received all acks
+              if @inactive_socks[k].ref <= 0
+                s = @inactive_socks.delete(k)
+                s.sock.close  rescue nil
+                @log.debug("purged obsolete socket #{s.sock}")
+              end
+            end
+
+            @active_socks.keys.each do |k|
+              if expired?(k) && @active_socks[k].ref <= 0
+                @inactive_socks[k] = @active_socks.delete(k)
+              end
+            end
+          end
+        end
+
+        # We expect that `yield` returns a unique object in this class
+        def fetch_or(key = Thread.current.object_id)
+          @mutex.synchronize do
+            unless @active_socks[key]
+              @active_socks[key] = TimedSocket.new(timeout, yield, 1)
+              @log.debug("connect new socket #{@active_socks[key]}")
+              return @active_socks[key].sock
+            end
+
+            if expired?(key)
+              # Do not close this socket here in case of it will be used by other place (e.g. wait for receiving ack)
+              @inactive_socks[key] = @active_socks.delete(key)
+              @log.debug("connection #{@inactive_socks[key]} is expired. reconnecting...")
+              @active_socks[key] = TimedSocket.new(timeout, yield, 0)
+            end
+
+            @active_socks[key].ref += 1;
+            @active_socks[key].sock
+          end
+        end
+
+        def dec_ref(key = Thread.current.object_id)
+          @mutex.synchronize do
+            if @active_socks[key]
+              @active_socks[key].ref -= 1
+            elsif @inactive_socks[key]
+              @inactive_socks[key].ref -= 1
+            else
+              @log.warn("Not found key for dec_ref: #{key}")
+            end
+          end
+        end
+
+        # This method is expected to be called in class which doesn't call #inc_ref
+        def dec_ref_by_value(val)
+          @mutex.synchronize do
+            sock = @active_socks.detect { |_, v| v.sock == val }
+            if sock
+              key = sock.first
+              @active_socks[key].ref -= 1
+              return
+            end
+
+            sock = @inactive_socks.detect { |_, v| v.sock == val }
+            if sock
+              key = sock.first
+              @inactive_socks[key].ref -= 1
+              return
+            else
+              @log.warn("Not found key for dec_ref_by_value: #{key}")
+            end
+          end
+        end
+
+        # This method is expected to be called in class which doesn't call #fetch_or
+        def revoke_by_value(val)
+          @mutex.synchronize do
+            sock = @active_socks.detect { |_, v| v.sock == val }
+            if sock
+              key = sock.first
+              @inactive_socks[key] = @active_socks.delete(key)
+              @inactive_socks[key].ref = 0
+            else
+              @log.debug("Not found for revoke_by_value :#{val}")
+            end
+          end
+        end
+
+        private
+
+        def timeout
+          @timeout && Time.now + @timeout
+        end
+
+        # This method is thread unsafe
+        def expired?(key = Thread.current.object_id)
+           @active_socks[key].timeout ? @active_socks[key].timeout < Time.now : false
+        end
+      end
+
+      # @param keepalive [Bool]
+      # @param keepalive_timeout [Integer | nil]
+      def initialize(sender, server, failure:, keepalive: false, keepalive_timeout: nil)
         @sender = sender
         @log = sender.log
         @compress = sender.compress
@@ -571,6 +737,11 @@ module Fluent::Plugin
         @resolved_host = nil
         @resolved_time = 0
         @resolved_once = false
+
+        @keepalive = keepalive
+        if @keepalive
+          @socket_cache = SocketCache.new(keepalive_timeout, @log)
+        end
       end
 
       attr_accessor :usock
@@ -578,6 +749,7 @@ module Fluent::Plugin
       attr_reader :name, :host, :port, :weight, :standby, :state
       attr_reader :sockaddr  # used by on_heartbeat
       attr_reader :failure, :available # for test
+      attr_reader :socket_cache        # for ack
 
       RequestInfo = Struct.new(:state, :shared_key_nonce, :auth)
 
@@ -598,15 +770,12 @@ module Fluent::Plugin
       end
 
       def verify_connection
-        sock = @sender.create_transfer_socket(resolved_host, port, @hostname)
-        begin
+        connect do |sock|
           ri = RequestInfo.new(@sender.security ? :helo : :established)
           if ri.state != :established
             establish_connection(sock, ri)
             raise if ri.state != :established
           end
-        ensure
-          sock.close
         end
       end
 
@@ -672,11 +841,16 @@ module Fluent::Plugin
       end
 
       def send_data(tag, chunk)
-        sock = @sender.create_transfer_socket(resolved_host, port, @hostname)
+        sock = connect
+
         begin
           send_data_actual(sock, tag, chunk)
         rescue
-          sock.close rescue nil
+          if @keepalive
+            @socket_cache.revoke
+          else
+            sock.close rescue nil
+          end
           raise
         end
 
@@ -684,12 +858,27 @@ module Fluent::Plugin
           return sock # to read ACK from socket
         end
 
-        sock.close_write rescue nil
-        sock.close rescue nil
+        if @keepalive
+          @socket_cache.dec_ref
+        else
+          sock.close_write rescue nil
+          sock.close rescue nil
+        end
         heartbeat(false)
         nil
       end
 
+      def clear
+        @keepalive && @socket_cache.clear
+      end
+
+      def purge_obsolete_socks
+        unless @keepalive
+          raise "Don not call this method without keepalive option"
+        end
+        @socket_cache.purge_obsolete_socks
+      end
+
       # FORWARD_TCP_HEARTBEAT_DATA = FORWARD_HEADER + ''.to_msgpack + [].to_msgpack
       def send_heartbeat
         begin
@@ -705,7 +894,7 @@ module Fluent::Plugin
 
         case @sender.heartbeat_type
         when :transport
-          @sender.create_transfer_socket(dest_addr, port, @hostname) do |sock|
+          connect(dest_addr) do |sock|
             ## don't send any data to not cause a compatibility problem
             # sock.write FORWARD_TCP_HEARTBEAT_DATA
 
@@ -880,6 +1069,32 @@ module Fluent::Plugin
           raise "BUG: unknown session state: #{ri.state}"
         end
       end
+
+      private
+
+      def connect(host = nil)
+        sock = if @keepalive
+                 @socket_cache.fetch_or { @sender.create_transfer_socket(host || resolved_host, port, @hostname) }
+               else
+                 @log.debug('connect new socket')
+                 @sender.create_transfer_socket(host || resolved_host, port, @hostname)
+               end
+
+        if block_given?
+          begin
+            yield(sock)
+          rescue
+            @socket_cache.revoke(sock) if @keepalive
+            raise
+          else
+            @socket_cache.dec_ref(sock) if @keepalive
+          ensure
+            sock.close unless @keepalive
+          end
+        else
+          sock
+        end
+      end
     end
 
     # Override Node to disable heartbeat