sidekiq 5.1.1 → 6.5.9

data/lib/sidekiq/launcher.rb

@@ -1,24 +1,33 @@
-# encoding: utf-8
 # frozen_string_literal: true
-require 'sidekiq/manager'
-require 'sidekiq/fetch'
-require 'sidekiq/scheduled'
+
+require "sidekiq/manager"
+require "sidekiq/fetch"
+require "sidekiq/scheduled"
+require "sidekiq/ring_buffer"
 
 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
+    include Sidekiq::Component
+
+    STATS_TTL = 5 * 365 * 24 * 60 * 60 # 5 years
+
+    PROCTITLES = [
+      proc { "sidekiq" },
+      proc { Sidekiq::VERSION },
+      proc { |me, data| data["tag"] },
+      proc { |me, data| "[#{Processor::WORK_STATE.size} of #{data["concurrency"]} busy]" },
+      proc { |me, data| "stopping" if me.stopping? }
+    ]
 
     attr_accessor :manager, :poller, :fetcher
 
     def initialize(options)
+      @config = options
+      options[:fetch] ||= BasicFetch.new(options)
       @manager = Sidekiq::Manager.new(options)
-      @poller = Sidekiq::Scheduled::Poller.new
+      @poller = Sidekiq::Scheduled::Poller.new(options)
       @done = false
-      @options = options
     end
 
     def run
@@ -35,11 +44,9 @@ module Sidekiq
       @poller.terminate
     end
 
-    # Shuts down the process.  This method does not
-    # return until all work is complete and cleaned up.
-    # It can take up to the timeout to complete.
+    # Shuts down this Sidekiq instance. Waits up to the deadline for all jobs to complete.
     def stop
-      deadline = Time.now + @options[:timeout]
+      deadline = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC) + @config[:timeout]
 
       @done = true
       @manager.quiet
@@ -47,10 +54,10 @@ module Sidekiq
 
       @manager.stop(deadline)
 
-      # Requeue everything in case there was a worker who grabbed work while stopped
+      # Requeue everything in case there was a thread which fetched a job while the process was stopped.
       # This call is a no-op in Sidekiq but necessary for Sidekiq Pro.
-      strategy = (@options[:fetch] || Sidekiq::BasicFetch)
-      strategy.bulk_requeue([], @options)
+      strategy = @config[:fetch]
+      strategy.bulk_requeue([], @config)
 
       clear_heartbeat
     end
@@ -61,107 +68,200 @@ module Sidekiq
 
     private unless $TESTING
 
+    BEAT_PAUSE = 5
+
+    def start_heartbeat
+      loop do
+        heartbeat
+        sleep BEAT_PAUSE
+      end
+      logger.info("Heartbeat stopping...")
+    end
+
+    def clear_heartbeat
+      flush_stats
+
+      # 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.
+      redis do |conn|
+        conn.pipelined do |pipeline|
+          pipeline.srem("processes", [identity])
+          pipeline.unlink("#{identity}:work")
+        end
+      end
+    rescue
+      # best effort, ignore network errors
+    end
+
     def heartbeat
-      results = Sidekiq::CLI::PROCTITLES.map {|x| x.(self, to_data) }
-      results.compact!
-      $0 = results.join(' ')
+      $0 = PROCTITLES.map { |proc| proc.call(self, to_data) }.compact.join(" ")
 
       ❤
     end
 
+    def flush_stats
+      fails = Processor::FAILURE.reset
+      procd = Processor::PROCESSED.reset
+      return if fails + procd == 0
+
+      nowdate = Time.now.utc.strftime("%Y-%m-%d")
+      begin
+        Sidekiq.redis do |conn|
+          conn.pipelined do |pipeline|
+            pipeline.incrby("stat:processed", procd)
+            pipeline.incrby("stat:processed:#{nowdate}", procd)
+            pipeline.expire("stat:processed:#{nowdate}", STATS_TTL)
+
+            pipeline.incrby("stat:failed", fails)
+            pipeline.incrby("stat:failed:#{nowdate}", fails)
+            pipeline.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
+
     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::WORK_STATE.dup
 
-        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:#{nowdate}", procd)
-            conn.incrby("stat:failed".freeze, fails)
-            conn.incrby("stat:failed:#{nowdate}", fails)
-            conn.del(workers_key)
-            Processor::WORKER_STATE.each_pair do |tid, hash|
-              conn.hset(workers_key, tid, Sidekiq.dump_json(hash))
+        nowdate = Time.now.utc.strftime("%Y-%m-%d")
+
+        redis do |conn|
+          conn.multi do |transaction|
+            transaction.incrby("stat:processed", procd)
+            transaction.incrby("stat:processed:#{nowdate}", procd)
+            transaction.expire("stat:processed:#{nowdate}", STATS_TTL)
+
+            transaction.incrby("stat:failed", fails)
+            transaction.incrby("stat:failed:#{nowdate}", fails)
+            transaction.expire("stat:failed:#{nowdate}", STATS_TTL)
+          end
+
+          # work is the current set of executing jobs
+          work_key = "#{key}:work"
+          conn.pipelined do |transaction|
+            transaction.unlink(work_key)
+            curstate.each_pair do |tid, hash|
+              transaction.hset(work_key, tid, Sidekiq.dump_json(hash))
             end
-            conn.expire(workers_key, 60)
+            transaction.expire(work_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', to_json, 'busy', Processor::WORKER_STATE.size, 'beat', Time.now.to_f, 'quiet', @done)
-            conn.expire(key, 60)
-            conn.rpop("#{key}-signals")
-          end
-        end
+        _, exists, _, _, msg = redis { |conn|
+          conn.multi { |transaction|
+            transaction.sadd("processes", [key])
+            transaction.exists?(key)
+            transaction.hmset(key, "info", to_json,
+              "busy", curstate.size,
+              "beat", Time.now.to_f,
+              "rtt_us", rtt,
+              "quiet", @done.to_s,
+              "rss", kb)
+            transaction.expire(key, 60)
+            transaction.rpop("#{key}-signals")
+          }
+        }
 
         # first heartbeat or recovering from an outage and need to reestablish our heartbeat
-        fire_event(:heartbeat) if !exists
+        fire_event(:heartbeat) unless exists
+        fire_event(:beat, oneshot: false)
 
         return unless msg
 
-        ::Process.kill(msg, $$)
+        ::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
-      while true
-        heartbeat
-        sleep 5
+    # 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
+
+    def check_rtt
+      a = b = 0
+      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 }
+        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-saturated; reduce your concurrency and/or see https://github.com/mperham/sidekiq/discussions/5039
+        EOM
+        RTT_READINGS.reset
+      end
+      rtt
     end
 
-    def to_data
-      @data ||= begin
-        {
-          'hostname' => hostname,
-          'started_at' => Time.now.to_f,
-          'pid' => $$,
-          'tag' => @options[:tag] || '',
-          'concurrency' => @options[:concurrency],
-          'queues' => @options[:queues].uniq,
-          'labels' => @options[:labels],
-          'identity' => identity,
-        }
-      end
+    MEMORY_GRABBER = case RUBY_PLATFORM
+    when /linux/
+      ->(pid) {
+        IO.readlines("/proc/#{$$}/status").each do |line|
+          next unless line.start_with?("VmRSS:")
+          break line.split[1].to_i
+        end
+      }
+    when /darwin|bsd/
+      ->(pid) {
+        `ps -o pid,rss -p #{pid}`.lines.last.split.last.to_i
+      }
+    else
+      ->(pid) { 0 }
     end
 
-    def to_json
-      @json ||= begin
-        # this data changes infrequently so dump it to a string
-        # now so we don't need to dump it every heartbeat.
-        Sidekiq.dump_json(to_data)
-      end
+    def memory_usage(pid)
+      MEMORY_GRABBER.call(pid)
     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")
-        end
-      end
-    rescue
-      # best effort, ignore network errors
+    def to_data
+      @data ||= {
+        "hostname" => hostname,
+        "started_at" => Time.now.to_f,
+        "pid" => ::Process.pid,
+        "tag" => @config[:tag] || "",
+        "concurrency" => @config[:concurrency],
+        "queues" => @config[:queues].uniq,
+        "labels" => @config[: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 +1,156 @@
+# 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
+
+    def self.add(k, v)
+      current[k] = v
+    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
+
+    LEVELS.each do |level, numeric_level|
+      define_method("#{level}?") do
+        local_level.nil? ? super() : local_level <= numeric_level
+      end
+    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

@@ -1,13 +1,10 @@
-# encoding: utf-8
 # frozen_string_literal: true
-require 'sidekiq/util'
-require 'sidekiq/processor'
-require 'sidekiq/fetch'
-require 'thread'
-require 'set'
 
-module Sidekiq
+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.
@@ -23,43 +20,37 @@ module Sidekiq
   # the shutdown process.  The other tasks are performed by other threads.
   #
   class Manager
-    include Util
+    include Sidekiq::Component
 
     attr_reader :workers
-    attr_reader :options
 
-    def initialize(options={})
+    def initialize(options = {})
+      @config = 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(@config, &method(:processor_result))
       end
       @plock = Mutex.new
     end
 
     def start
-      @workers.each do |x|
-        x.start
-      end
+      @workers.each(&:start)
     end
 
     def quiet
       return if @done
       @done = true
 
-      logger.info { "Terminating quiet workers" }
-      @workers.each { |x| x.terminate }
+      logger.info { "Terminating quiet threads" }
+      @workers.each(&:terminate)
       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, reverse: true)
@@ -70,29 +61,18 @@ module Sidekiq
       sleep PAUSE_TIME
       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
+      logger.info { "Pausing to allow jobs to finish..." }
+      wait_for(deadline) { @workers.empty? }
       return if @workers.empty?
 
       hard_shutdown
     end
 
-    def processor_stopped(processor)
-      @plock.synchronize do
-        @workers.delete(processor)
-      end
-    end
-
-    def processor_died(processor, reason)
+    def processor_result(processor, reason = nil)
       @plock.synchronize do
         @workers.delete(processor)
         unless @done
-          p = Processor.new(self)
+          p = Processor.new(@config, &method(:processor_result))
           @workers << p
           p.start
         end
@@ -106,7 +86,7 @@ module Sidekiq
     private
 
     def hard_shutdown
-      # We've reached the timeout and we still have busy workers.
+      # We've reached the timeout and we still have busy threads.
       # They must die but their jobs shall live on.
       cleanup = nil
       @plock.synchronize do
@@ -114,25 +94,43 @@ 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}" }
+        logger.warn { "Terminating #{cleanup.size} busy threads" }
+        logger.debug { "Jobs still in progress #{jobs.inspect}" }
 
         # Re-enqueue unfinished jobs
         # NOTE: You may notice that we may push a job back to redis before
-        # the worker thread is terminated. This is ok because Sidekiq's
+        # the thread is terminated. This is ok because Sidekiq's
         # 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.bulk_requeue(jobs, @options)
+        strategy = @config[:fetch]
+        strategy.bulk_requeue(jobs, @config)
       end
 
       cleanup.each do |processor|
         processor.kill
       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
 
+    # hack for quicker development / testing environment #2774
+    PAUSE_TIME = $stdout.tty? ? 0.1 : 0.5
+
+    # Wait for the orblock to be true or the deadline passed.
+    def wait_for(deadline, &condblock)
+      remaining = deadline - ::Process.clock_gettime(::Process::CLOCK_MONOTONIC)
+      while remaining > PAUSE_TIME
+        return if condblock.call
+        sleep PAUSE_TIME
+        remaining = deadline - ::Process.clock_gettime(::Process::CLOCK_MONOTONIC)
+      end
+    end
   end
 end