sidekiq 4.2.4 → 6.4.0

data/lib/sidekiq/launcher.rb

@@ -1,20 +1,28 @@
 # frozen_string_literal: true
-# encoding: utf-8
-require 'sidekiq/manager'
-require 'sidekiq/fetch'
-require 'sidekiq/scheduled'
+
+require "sidekiq/manager"
+require "sidekiq/fetch"
+require "sidekiq/scheduled"
 
 module Sidekiq
-  # The Launcher is a very simple Actor whose job is to
-  # start, monitor and stop the core Actors in Sidekiq.
-  # If any of these actors die, the Sidekiq process exits
-  # immediately.
+  # The Launcher starts the Manager and Poller threads and provides the process heartbeat.
   class Launcher
     include Util
 
+    STATS_TTL = 5 * 365 * 24 * 60 * 60 # 5 years
+
+    PROCTITLES = [
+      proc { "sidekiq" },
+      proc { Sidekiq::VERSION },
+      proc { |me, data| data["tag"] },
+      proc { |me, data| "[#{Processor::WORKER_STATE.size} of #{data["concurrency"]} busy]" },
+      proc { |me, data| "stopping" if me.stopping? }
+    ]
+
     attr_accessor :manager, :poller, :fetcher
 
     def initialize(options)
+      options[:fetch] ||= BasicFetch.new(options)
       @manager = Sidekiq::Manager.new(options)
       @poller = Sidekiq::Scheduled::Poller.new
       @done = false
@@ -39,7 +47,7 @@ module Sidekiq
     # return until all work is complete and cleaned up.
     # It can take up to the timeout to complete.
     def stop
-      deadline = Time.now + @options[:timeout]
+      deadline = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC) + @options[:timeout]
 
       @done = true
       @manager.quiet
@@ -49,7 +57,7 @@ module Sidekiq
 
       # Requeue everything in case there was a worker who grabbed work while stopped
       # This call is a no-op in Sidekiq but necessary for Sidekiq Pro.
-      strategy = (@options[:fetch] || Sidekiq::BasicFetch)
+      strategy = @options[:fetch]
       strategy.bulk_requeue([], @options)
 
       clear_heartbeat
@@ -61,104 +69,195 @@ module Sidekiq
 
     private unless $TESTING
 
-    JVM_RESERVED_SIGNALS = ['USR1', 'USR2'] # Don't Process#kill if we get these signals via the API
+    BEAT_PAUSE = 5
+
+    def start_heartbeat
+      loop do
+        heartbeat
+        sleep BEAT_PAUSE
+      end
+      Sidekiq.logger.info("Heartbeat stopping...")
+    end
+
+    def clear_heartbeat
+      # Remove record from Redis since we are shutting down.
+      # Note we don't stop the heartbeat thread; if the process
+      # doesn't actually exit, it'll reappear in the Web UI.
+      Sidekiq.redis do |conn|
+        conn.pipelined do
+          conn.srem("processes", identity)
+          conn.unlink("#{identity}:workers")
+        end
+      end
+    rescue
+      # best effort, ignore network errors
+    end
+
+    def heartbeat
+      $0 = PROCTITLES.map { |proc| proc.call(self, to_data) }.compact.join(" ")
+
+      ❤
+    end
+
+    def self.flush_stats
+      fails = Processor::FAILURE.reset
+      procd = Processor::PROCESSED.reset
+      return if fails + procd == 0
 
-    def heartbeat(k, data, json)
-      results = Sidekiq::CLI::PROCTITLES.map {|x| x.(self, data) }
-      results.compact!
-      $0 = results.join(' ')
+      nowdate = Time.now.utc.strftime("%Y-%m-%d")
+      begin
+        Sidekiq.redis do |conn|
+          conn.pipelined do
+            conn.incrby("stat:processed", procd)
+            conn.incrby("stat:processed:#{nowdate}", procd)
+            conn.expire("stat:processed:#{nowdate}", STATS_TTL)
 
-      ❤(k, json)
+            conn.incrby("stat:failed", fails)
+            conn.incrby("stat:failed:#{nowdate}", fails)
+            conn.expire("stat:failed:#{nowdate}", STATS_TTL)
+          end
+        end
+      rescue => ex
+        # we're exiting the process, things might be shut down so don't
+        # try to handle the exception
+        Sidekiq.logger.warn("Unable to flush stats: #{ex}")
+      end
     end
+    at_exit(&method(:flush_stats))
 
-    def ❤(key, json)
+    def ❤
+      key = identity
       fails = procd = 0
+
       begin
-        Processor::FAILURE.update {|curr| fails = curr; 0 }
-        Processor::PROCESSED.update {|curr| procd = curr; 0 }
+        fails = Processor::FAILURE.reset
+        procd = Processor::PROCESSED.reset
+        curstate = Processor::WORKER_STATE.dup
+
+        workers_key = "#{key}:workers"
+        nowdate = Time.now.utc.strftime("%Y-%m-%d")
 
-        workers_key = "#{key}:workers".freeze
-        nowdate = Time.now.utc.strftime("%Y-%m-%d".freeze)
         Sidekiq.redis do |conn|
           conn.multi do
-            conn.incrby("stat:processed".freeze, procd)
+            conn.incrby("stat:processed", procd)
             conn.incrby("stat:processed:#{nowdate}", procd)
-            conn.incrby("stat:failed".freeze, fails)
+            conn.expire("stat:processed:#{nowdate}", STATS_TTL)
+
+            conn.incrby("stat:failed", fails)
             conn.incrby("stat:failed:#{nowdate}", fails)
-            conn.del(workers_key)
-            Processor::WORKER_STATE.each_pair do |tid, hash|
+            conn.expire("stat:failed:#{nowdate}", STATS_TTL)
+
+            conn.unlink(workers_key)
+            curstate.each_pair do |tid, hash|
               conn.hset(workers_key, tid, Sidekiq.dump_json(hash))
             end
             conn.expire(workers_key, 60)
           end
         end
+
+        rtt = check_rtt
+
         fails = procd = 0
+        kb = memory_usage(::Process.pid)
 
-        _, exists, _, _, msg = Sidekiq.redis do |conn|
-          conn.multi do
-            conn.sadd('processes', key)
-            conn.exists(key)
-            conn.hmset(key, 'info', json, 'busy', Processor::WORKER_STATE.size, 'beat', Time.now.to_f, 'quiet', @done)
+        _, exists, _, _, msg = Sidekiq.redis { |conn|
+          conn.multi {
+            conn.sadd("processes", key)
+            conn.exists?(key)
+            conn.hmset(key, "info", to_json,
+              "busy", curstate.size,
+              "beat", Time.now.to_f,
+              "rtt_us", rtt,
+              "quiet", @done,
+              "rss", kb)
             conn.expire(key, 60)
             conn.rpop("#{key}-signals")
-          end
-        end
+          }
+        }
 
         # first heartbeat or recovering from an outage and need to reestablish our heartbeat
-        fire_event(:heartbeat) if !exists
+        fire_event(:heartbeat) unless exists
 
         return unless msg
 
-        if JVM_RESERVED_SIGNALS.include?(msg)
-          Sidekiq::CLI.instance.handle_signal(msg)
-        else
-          ::Process.kill(msg, $$)
-        end
+        ::Process.kill(msg, ::Process.pid)
       rescue => e
         # ignore all redis/network issues
-        logger.error("heartbeat: #{e.message}")
+        logger.error("heartbeat: #{e}")
         # don't lose the counts if there was a network issue
-        Processor::PROCESSED.increment(procd)
-        Processor::FAILURE.increment(fails)
+        Processor::PROCESSED.incr(procd)
+        Processor::FAILURE.incr(fails)
       end
     end
 
-    def start_heartbeat
-      k = identity
-      data = {
-        'hostname' => hostname,
-        'started_at' => Time.now.to_f,
-        'pid' => $$,
-        'tag' => @options[:tag] || '',
-        'concurrency' => @options[:concurrency],
-        'queues' => @options[:queues].uniq,
-        'labels' => @options[:labels],
-        'identity' => k,
-      }
-      # this data doesn't change so dump it to a string
-      # now so we don't need to dump it every heartbeat.
-      json = Sidekiq.dump_json(data)
+    # We run the heartbeat every five seconds.
+    # Capture five samples of RTT, log a warning if each sample
+    # is above our warning threshold.
+    RTT_READINGS = RingBuffer.new(5)
+    RTT_WARNING_LEVEL = 50_000
 
-      while true
-        heartbeat(k, data, json)
-        sleep 5
+    def check_rtt
+      a = b = 0
+      Sidekiq.redis do |x|
+        a = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC, :microsecond)
+        x.ping
+        b = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC, :microsecond)
       end
-      Sidekiq.logger.info("Heartbeat stopping...")
+      rtt = b - a
+      RTT_READINGS << rtt
+      # Ideal RTT for Redis is < 1000µs
+      # Workable is < 10,000µs
+      # Log a warning if it's a disaster.
+      if RTT_READINGS.all? { |x| x > RTT_WARNING_LEVEL }
+        Sidekiq.logger.warn <<~EOM
+          Your Redis network connection is performing extremely poorly.
+          Last RTT readings were #{RTT_READINGS.buffer.inspect}, ideally these should be < 1000.
+          Ensure Redis is running in the same AZ or datacenter as Sidekiq.
+          If these values are close to 100,000, that means your Sidekiq process may be
+          CPU overloaded; see https://github.com/mperham/sidekiq/discussions/5039
+        EOM
+        RTT_READINGS.reset
+      end
+      rtt
     end
 
-    def clear_heartbeat
-      # Remove record from Redis since we are shutting down.
-      # Note we don't stop the heartbeat thread; if the process
-      # doesn't actually exit, it'll reappear in the Web UI.
-      Sidekiq.redis do |conn|
-        conn.pipelined do
-          conn.srem('processes', identity)
-          conn.del("#{identity}:workers")
+    MEMORY_GRABBER = case RUBY_PLATFORM
+    when /linux/
+      ->(pid) {
+        IO.readlines("/proc/#{$$}/status").each do |line|
+          next unless line.start_with?("VmRSS:")
+          break line.split[1].to_i
         end
-      end
-    rescue
-      # best effort, ignore network errors
+      }
+    when /darwin|bsd/
+      ->(pid) {
+        `ps -o pid,rss -p #{pid}`.lines.last.split.last.to_i
+      }
+    else
+      ->(pid) { 0 }
+    end
+
+    def memory_usage(pid)
+      MEMORY_GRABBER.call(pid)
+    end
+
+    def to_data
+      @data ||= {
+        "hostname" => hostname,
+        "started_at" => Time.now.to_f,
+        "pid" => ::Process.pid,
+        "tag" => @options[:tag] || "",
+        "concurrency" => @options[:concurrency],
+        "queues" => @options[:queues].uniq,
+        "labels" => @options[:labels],
+        "identity" => identity
+      }
     end
 
+    def to_json
+      # this data changes infrequently so dump it to a string
+      # now so we don't need to dump it every heartbeat.
+      @json ||= Sidekiq.dump_json(to_data)
+    end
   end
 end

data/lib/sidekiq/logger.rb

@@ -0,0 +1,166 @@
+# frozen_string_literal: true
+
+require "logger"
+require "time"
+
+module Sidekiq
+  module Context
+    def self.with(hash)
+      orig_context = current.dup
+      current.merge!(hash)
+      yield
+    ensure
+      Thread.current[:sidekiq_context] = orig_context
+    end
+
+    def self.current
+      Thread.current[:sidekiq_context] ||= {}
+    end
+  end
+
+  module LoggingUtils
+    LEVELS = {
+      "debug" => 0,
+      "info" => 1,
+      "warn" => 2,
+      "error" => 3,
+      "fatal" => 4
+    }
+    LEVELS.default_proc = proc do |_, level|
+      Sidekiq.logger.warn("Invalid log level: #{level.inspect}")
+      nil
+    end
+
+    def debug?
+      level <= 0
+    end
+
+    def info?
+      level <= 1
+    end
+
+    def warn?
+      level <= 2
+    end
+
+    def error?
+      level <= 3
+    end
+
+    def fatal?
+      level <= 4
+    end
+
+    def local_level
+      Thread.current[:sidekiq_log_level]
+    end
+
+    def local_level=(level)
+      case level
+      when Integer
+        Thread.current[:sidekiq_log_level] = level
+      when Symbol, String
+        Thread.current[:sidekiq_log_level] = LEVELS[level.to_s]
+      when nil
+        Thread.current[:sidekiq_log_level] = nil
+      else
+        raise ArgumentError, "Invalid log level: #{level.inspect}"
+      end
+    end
+
+    def level
+      local_level || super
+    end
+
+    # Change the thread-local level for the duration of the given block.
+    def log_at(level)
+      old_local_level = local_level
+      self.local_level = level
+      yield
+    ensure
+      self.local_level = old_local_level
+    end
+
+    # Redefined to check severity against #level, and thus the thread-local level, rather than +@level+.
+    # FIXME: Remove when the minimum Ruby version supports overriding Logger#level.
+    def add(severity, message = nil, progname = nil, &block)
+      severity ||= ::Logger::UNKNOWN
+      progname ||= @progname
+
+      return true if @logdev.nil? || severity < level
+
+      if message.nil?
+        if block
+          message = yield
+        else
+          message = progname
+          progname = @progname
+        end
+      end
+
+      @logdev.write format_message(format_severity(severity), Time.now, progname, message)
+    end
+  end
+
+  class Logger < ::Logger
+    include LoggingUtils
+
+    def initialize(*args, **kwargs)
+      super
+      self.formatter = Sidekiq.log_formatter
+    end
+
+    module Formatters
+      class Base < ::Logger::Formatter
+        def tid
+          Thread.current["sidekiq_tid"] ||= (Thread.current.object_id ^ ::Process.pid).to_s(36)
+        end
+
+        def ctx
+          Sidekiq::Context.current
+        end
+
+        def format_context
+          if ctx.any?
+            " " + ctx.compact.map { |k, v|
+              case v
+              when Array
+                "#{k}=#{v.join(",")}"
+              else
+                "#{k}=#{v}"
+              end
+            }.join(" ")
+          end
+        end
+      end
+
+      class Pretty < Base
+        def call(severity, time, program_name, message)
+          "#{time.utc.iso8601(3)} pid=#{::Process.pid} tid=#{tid}#{format_context} #{severity}: #{message}\n"
+        end
+      end
+
+      class WithoutTimestamp < Pretty
+        def call(severity, time, program_name, message)
+          "pid=#{::Process.pid} tid=#{tid}#{format_context} #{severity}: #{message}\n"
+        end
+      end
+
+      class JSON < Base
+        def call(severity, time, program_name, message)
+          hash = {
+            ts: time.utc.iso8601(3),
+            pid: ::Process.pid,
+            tid: tid,
+            lvl: severity,
+            msg: message
+          }
+          c = ctx
+          hash["ctx"] = c unless c.empty?
+
+          Sidekiq.dump_json(hash) << "\n"
+        end
+      end
+    end
+  end
+end

data/lib/sidekiq/manager.rb

@@ -1,16 +1,14 @@
 # frozen_string_literal: true
-# encoding: utf-8
-require 'sidekiq/util'
-require 'sidekiq/processor'
-require 'sidekiq/fetch'
-require 'thread'
-require 'set'
 
-module Sidekiq
+require "sidekiq/util"
+require "sidekiq/processor"
+require "sidekiq/fetch"
+require "set"
 
+module Sidekiq
   ##
   # The Manager is the central coordination point in Sidekiq, controlling
-  # the lifecycle of the Processors and feeding them jobs as necessary.
+  # the lifecycle of the Processors.
   #
   # Tasks:
   #
@@ -28,16 +26,16 @@ module Sidekiq
     attr_reader :workers
     attr_reader :options
 
-    def initialize(options={})
+    def initialize(options = {})
       logger.debug { options.inspect }
       @options = options
-      @count = options[:concurrency] || 25
+      @count = options[:concurrency] || 10
       raise ArgumentError, "Concurrency of #{@count} is not supported" if @count < 1
 
       @done = false
       @workers = Set.new
       @count.times do
-        @workers << Processor.new(self)
+        @workers << Processor.new(self, options)
       end
       @plock = Mutex.new
     end
@@ -54,15 +52,12 @@ module Sidekiq
 
       logger.info { "Terminating quiet workers" }
       @workers.each { |x| x.terminate }
-      fire_event(:quiet, true)
+      fire_event(:quiet, reverse: true)
     end
 
-    # hack for quicker development / testing environment #2774
-    PAUSE_TIME = STDOUT.tty? ? 0.1 : 0.5
-
     def stop(deadline)
       quiet
-      fire_event(:shutdown, true)
+      fire_event(:shutdown, reverse: true)
 
       # some of the shutdown events can be async,
       # we don't have any way to know when they're done but
@@ -71,12 +66,7 @@ module Sidekiq
       return if @workers.empty?
 
       logger.info { "Pausing to allow workers to finish..." }
-      remaining = deadline - Time.now
-      while remaining > PAUSE_TIME
-        return if @workers.empty?
-        sleep PAUSE_TIME
-        remaining = deadline - Time.now
-      end
+      wait_for(deadline) { @workers.empty? }
       return if @workers.empty?
 
       hard_shutdown
@@ -92,7 +82,7 @@ module Sidekiq
       @plock.synchronize do
         @workers.delete(processor)
         unless @done
-          p = Processor.new(self)
+          p = Processor.new(self, options)
           @workers << p
           p.start
         end
@@ -114,7 +104,7 @@ module Sidekiq
       end
 
       if cleanup.size > 0
-        jobs = cleanup.map {|p| p.job }.compact
+        jobs = cleanup.map { |p| p.job }.compact
 
         logger.warn { "Terminating #{cleanup.size} busy worker threads" }
         logger.warn { "Work still in progress #{jobs.inspect}" }
@@ -125,14 +115,19 @@ module Sidekiq
         # contract says that jobs are run AT LEAST once. Process termination
         # is delayed until we're certain the jobs are back in Redis because
         # it is worse to lose a job than to run it twice.
-        strategy = (@options[:fetch] || Sidekiq::BasicFetch)
+        strategy = @options[:fetch]
         strategy.bulk_requeue(jobs, @options)
       end
 
       cleanup.each do |processor|
         processor.kill
       end
-    end
 
+      # when this method returns, we immediately call `exit` which may not give
+      # the remaining threads time to run `ensure` blocks, etc. We pause here up
+      # to 3 seconds to give threads a minimal amount of time to run `ensure` blocks.
+      deadline = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC) + 3
+      wait_for(deadline) { @workers.empty? }
+    end
   end
 end

data/lib/sidekiq/middleware/chain.rb

@@ -1,4 +1,5 @@
 # frozen_string_literal: true
+
 module Sidekiq
   # Middleware is code configured to run before/after
   # a message is processed.  It is patterned after Rack
@@ -66,7 +67,6 @@ module Sidekiq
   module Middleware
     class Chain
       include Enumerable
-      attr_reader :entries
 
       def initialize_copy(copy)
         copy.instance_variable_set(:@entries, entries.dup)
@@ -77,21 +77,25 @@ module Sidekiq
       end
 
       def initialize
-        @entries = []
+        @entries = nil
         yield self if block_given?
       end
 
+      def entries
+        @entries ||= []
+      end
+
       def remove(klass)
         entries.delete_if { |entry| entry.klass == klass }
       end
 
       def add(klass, *args)
-        remove(klass) if exists?(klass)
+        remove(klass)
         entries << Entry.new(klass, *args)
       end
 
       def prepend(klass, *args)
-        remove(klass) if exists?(klass)
+        remove(klass)
         entries.insert(0, Entry.new(klass, *args))
       end
 
@@ -106,13 +110,17 @@ module Sidekiq
         i = entries.index { |entry| entry.klass == newklass }
         new_entry = i.nil? ? Entry.new(newklass, *args) : entries.delete_at(i)
         i = entries.index { |entry| entry.klass == oldklass } || entries.count - 1
-        entries.insert(i+1, new_entry)
+        entries.insert(i + 1, new_entry)
       end
 
       def exists?(klass)
         any? { |entry| entry.klass == klass }
       end
 
+      def empty?
+        @entries.nil? || @entries.empty?
+      end
+
       def retrieve
         map(&:make_new)
       end
@@ -122,8 +130,10 @@ module Sidekiq
       end
 
       def invoke(*args)
-        chain = retrieve.dup
-        traverse_chain = lambda do
+        return yield if empty?
+
+        chain = retrieve
+        traverse_chain = proc do
           if chain.empty?
             yield
           else
@@ -134,12 +144,14 @@ module Sidekiq
       end
     end
 
+    private
+
     class Entry
       attr_reader :klass
 
       def initialize(klass, *args)
         @klass = klass
-        @args  = args
+        @args = args
       end
 
       def make_new