sidekiq 6.5.1 → 7.0.9

data/lib/sidekiq/launcher.rb

@@ -1,12 +1,12 @@
 # frozen_string_literal: true
 
 require "sidekiq/manager"
-require "sidekiq/fetch"
+require "sidekiq/capsule"
 require "sidekiq/scheduled"
 require "sidekiq/ring_buffer"
 
 module Sidekiq
-  # The Launcher starts the Manager and Poller threads and provides the process heartbeat.
+  # The Launcher starts the Capsule Managers, the Poller thread and provides the process heartbeat.
   class Launcher
     include Sidekiq::Component
 
@@ -16,48 +16,56 @@ module Sidekiq
       proc { "sidekiq" },
       proc { Sidekiq::VERSION },
       proc { |me, data| data["tag"] },
-      proc { |me, data| "[#{Processor::WORK_STATE.size} of #{data["concurrency"]} busy]" },
+      proc { |me, data| "[#{Processor::WORK_STATE.size} of #{me.config.total_concurrency} busy]" },
       proc { |me, data| "stopping" if me.stopping? }
     ]
 
-    attr_accessor :manager, :poller, :fetcher
+    attr_accessor :managers, :poller
 
-    def initialize(options)
-      @config = options
-      options[:fetch] ||= BasicFetch.new(options)
-      @manager = Sidekiq::Manager.new(options)
-      @poller = Sidekiq::Scheduled::Poller.new(options)
+    def initialize(config, embedded: false)
+      @config = config
+      @embedded = embedded
+      @managers = config.capsules.values.map do |cap|
+        Sidekiq::Manager.new(cap)
+      end
+      @poller = Sidekiq::Scheduled::Poller.new(@config)
       @done = false
     end
 
-    def run
-      @thread = safe_thread("heartbeat", &method(:start_heartbeat))
+    # Start this Sidekiq instance. If an embedding process already
+    # has a heartbeat thread, caller can use `async_beat: false`
+    # and instead have thread call Launcher#heartbeat every N seconds.
+    def run(async_beat: true)
+      Sidekiq.freeze!
+      logger.debug { @config.merge!({}) }
+      @thread = safe_thread("heartbeat", &method(:start_heartbeat)) if async_beat
       @poller.start
-      @manager.start
+      @managers.each(&:start)
     end
 
     # Stops this instance from processing any more jobs,
-    #
     def quiet
+      return if @done
+
       @done = true
-      @manager.quiet
+      @managers.each(&:quiet)
       @poller.terminate
+      fire_event(:quiet, reverse: true)
     end
 
     # Shuts down this Sidekiq instance. Waits up to the deadline for all jobs to complete.
     def stop
       deadline = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC) + @config[:timeout]
 
-      @done = true
-      @manager.quiet
-      @poller.terminate
-
-      @manager.stop(deadline)
+      quiet
+      stoppers = @managers.map do |mgr|
+        Thread.new do
+          mgr.stop(deadline)
+        end
+      end
 
-      # 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 = @config[:fetch]
-      strategy.bulk_requeue([], @config)
+      fire_event(:shutdown, reverse: true)
+      stoppers.each(&:join)
 
       clear_heartbeat
     end
@@ -66,25 +74,39 @@ module Sidekiq
       @done
     end
 
+    # If embedding Sidekiq, you can have the process heartbeat
+    # call this method to regularly heartbeat rather than creating
+    # a separate thread.
+    def heartbeat
+      ❤
+    end
+
     private unless $TESTING
 
-    BEAT_PAUSE = 5
+    BEAT_PAUSE = 10
 
     def start_heartbeat
       loop do
-        heartbeat
+        beat
         sleep BEAT_PAUSE
       end
       logger.info("Heartbeat stopping...")
     end
 
+    def beat
+      $0 = PROCTITLES.map { |proc| proc.call(self, to_data) }.compact.join(" ") unless @embedded
+      ❤
+    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.srem("processes", [identity])
           pipeline.unlink("#{identity}:work")
         end
       end
@@ -92,20 +114,14 @@ module Sidekiq
       # best effort, ignore network errors
     end
 
-    def heartbeat
-      $0 = PROCTITLES.map { |proc| proc.call(self, to_data) }.compact.join(" ")
-
-      ❤
-    end
-
-    def self.flush_stats
+    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|
+        redis do |conn|
           conn.pipelined do |pipeline|
             pipeline.incrby("stat:processed", procd)
             pipeline.incrby("stat:processed:#{nowdate}", procd)
@@ -117,35 +133,19 @@ module Sidekiq
           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}")
+        logger.warn("Unable to flush stats: #{ex}")
       end
     end
-    at_exit(&method(:flush_stats))
 
     def ❤
       key = identity
       fails = procd = 0
 
       begin
-        fails = Processor::FAILURE.reset
-        procd = Processor::PROCESSED.reset
-        curstate = Processor::WORK_STATE.dup
-
-        nowdate = Time.now.utc.strftime("%Y-%m-%d")
+        flush_stats
 
+        curstate = Processor::WORK_STATE.dup
         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|
@@ -162,10 +162,10 @@ module Sidekiq
         fails = procd = 0
         kb = memory_usage(::Process.pid)
 
-        _, exists, _, _, msg = redis { |conn|
+        _, exists, _, _, signal = redis { |conn|
           conn.multi { |transaction|
-            transaction.sadd("processes", key)
-            transaction.exists?(key)
+            transaction.sadd("processes", [key])
+            transaction.exists(key)
             transaction.hmset(key, "info", to_json,
               "busy", curstate.size,
               "beat", Time.now.to_f,
@@ -178,11 +178,10 @@ module Sidekiq
         }
 
         # first heartbeat or recovering from an outage and need to reestablish our heartbeat
-        fire_event(:heartbeat) unless exists
+        fire_event(:heartbeat) unless exists > 0
+        fire_event(:beat, oneshot: false)
 
-        return unless msg
-
-        ::Process.kill(msg, ::Process.pid)
+        ::Process.kill(signal, ::Process.pid) if signal && !@embedded
       rescue => e
         # ignore all redis/network issues
         logger.error("heartbeat: #{e}")
@@ -216,7 +215,7 @@ module Sidekiq
           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
+          CPU-saturated; reduce your concurrency and/or see https://github.com/sidekiq/sidekiq/discussions/5039
         EOM
         RTT_READINGS.reset
       end
@@ -249,13 +248,20 @@ module Sidekiq
         "started_at" => Time.now.to_f,
         "pid" => ::Process.pid,
         "tag" => @config[:tag] || "",
-        "concurrency" => @config[:concurrency],
-        "queues" => @config[:queues].uniq,
-        "labels" => @config[:labels],
-        "identity" => identity
+        "concurrency" => @config.total_concurrency,
+        "queues" => @config.capsules.values.flat_map { |cap| cap.queues }.uniq,
+        "weights" => to_weights,
+        "labels" => @config[:labels].to_a,
+        "identity" => identity,
+        "version" => Sidekiq::VERSION,
+        "embedded" => @embedded
       }
     end
 
+    def to_weights
+      @config.capsules.values.map(&:weights)
+    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.

data/lib/sidekiq/logger.rb

@@ -31,7 +31,7 @@ module Sidekiq
       "fatal" => 4
     }
     LEVELS.default_proc = proc do |_, level|
-      Sidekiq.logger.warn("Invalid log level: #{level.inspect}")
+      puts("Invalid log level: #{level.inspect}")
       nil
     end
 
@@ -70,36 +70,11 @@ module Sidekiq
     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

data/lib/sidekiq/manager.rb

@@ -1,7 +1,6 @@
 # frozen_string_literal: true
 
 require "sidekiq/processor"
-require "sidekiq/fetch"
 require "set"
 
 module Sidekiq
@@ -23,19 +22,19 @@ module Sidekiq
     include Sidekiq::Component
 
     attr_reader :workers
+    attr_reader :capsule
 
-    def initialize(options = {})
-      @config = options
-      logger.debug { options.inspect }
-      @count = options[:concurrency] || 10
+    def initialize(capsule)
+      @config = @capsule = capsule
+      @count = capsule.concurrency
       raise ArgumentError, "Concurrency of #{@count} is not supported" if @count < 1
 
       @done = false
       @workers = Set.new
+      @plock = Mutex.new
       @count.times do
         @workers << Processor.new(@config, &method(:processor_result))
       end
-      @plock = Mutex.new
     end
 
     def start
@@ -46,14 +45,12 @@ module Sidekiq
       return if @done
       @done = true
 
-      logger.info { "Terminating quiet threads" }
+      logger.info { "Terminating quiet threads for #{capsule.name} capsule" }
       @workers.each(&:terminate)
-      fire_event(:quiet, reverse: true)
     end
 
     def stop(deadline)
       quiet
-      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
@@ -66,6 +63,8 @@ module Sidekiq
       return if @workers.empty?
 
       hard_shutdown
+    ensure
+      capsule.stop
     end
 
     def processor_result(processor, reason = nil)
@@ -105,8 +104,7 @@ 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 = @config[:fetch]
-        strategy.bulk_requeue(jobs, @config)
+        capsule.fetcher.bulk_requeue(jobs)
       end
 
       cleanup.each do |processor|

data/lib/sidekiq/metrics/query.rb

@@ -0,0 +1,153 @@
+require "sidekiq"
+require "date"
+require "set"
+
+require "sidekiq/metrics/shared"
+
+module Sidekiq
+  module Metrics
+    # Allows caller to query for Sidekiq execution metrics within Redis.
+    # Caller sets a set of attributes to act as filters. {#fetch} will call
+    # Redis and return a Hash of results.
+    #
+    # NB: all metrics and times/dates are UTC only. We specifically do not
+    # support timezones.
+    class Query
+      def initialize(pool: nil, now: Time.now)
+        @time = now.utc
+        @pool = pool || Sidekiq.default_configuration.redis_pool
+        @klass = nil
+      end
+
+      # Get metric data for all jobs from the last hour
+      def top_jobs(minutes: 60)
+        result = Result.new
+
+        time = @time
+        redis_results = @pool.with do |conn|
+          conn.pipelined do |pipe|
+            minutes.times do |idx|
+              key = "j|#{time.strftime("%Y%m%d")}|#{time.hour}:#{time.min}"
+              pipe.hgetall key
+              result.prepend_bucket time
+              time -= 60
+            end
+          end
+        end
+
+        time = @time
+        redis_results.each do |hash|
+          hash.each do |k, v|
+            kls, metric = k.split("|")
+            result.job_results[kls].add_metric metric, time, v.to_i
+          end
+          time -= 60
+        end
+
+        result.marks = fetch_marks(result.starts_at..result.ends_at)
+
+        result
+      end
+
+      def for_job(klass, minutes: 60)
+        result = Result.new
+
+        time = @time
+        redis_results = @pool.with do |conn|
+          conn.pipelined do |pipe|
+            minutes.times do |idx|
+              key = "j|#{time.strftime("%Y%m%d")}|#{time.hour}:#{time.min}"
+              pipe.hmget key, "#{klass}|ms", "#{klass}|p", "#{klass}|f"
+              result.prepend_bucket time
+              time -= 60
+            end
+          end
+        end
+
+        time = @time
+        @pool.with do |conn|
+          redis_results.each do |(ms, p, f)|
+            result.job_results[klass].add_metric "ms", time, ms.to_i if ms
+            result.job_results[klass].add_metric "p", time, p.to_i if p
+            result.job_results[klass].add_metric "f", time, f.to_i if f
+            result.job_results[klass].add_hist time, Histogram.new(klass).fetch(conn, time).reverse
+            time -= 60
+          end
+        end
+
+        result.marks = fetch_marks(result.starts_at..result.ends_at)
+
+        result
+      end
+
+      class Result < Struct.new(:starts_at, :ends_at, :size, :buckets, :job_results, :marks)
+        def initialize
+          super
+          self.buckets = []
+          self.marks = []
+          self.job_results = Hash.new { |h, k| h[k] = JobResult.new }
+        end
+
+        def prepend_bucket(time)
+          buckets.unshift time.strftime("%H:%M")
+          self.ends_at ||= time
+          self.starts_at = time
+        end
+      end
+
+      class JobResult < Struct.new(:series, :hist, :totals)
+        def initialize
+          super
+          self.series = Hash.new { |h, k| h[k] = Hash.new(0) }
+          self.hist = Hash.new { |h, k| h[k] = [] }
+          self.totals = Hash.new(0)
+        end
+
+        def add_metric(metric, time, value)
+          totals[metric] += value
+          series[metric][time.strftime("%H:%M")] += value
+
+          # Include timing measurements in seconds for convenience
+          add_metric("s", time, value / 1000.0) if metric == "ms"
+        end
+
+        def add_hist(time, hist_result)
+          hist[time.strftime("%H:%M")] = hist_result
+        end
+
+        def total_avg(metric = "ms")
+          completed = totals["p"] - totals["f"]
+          totals[metric].to_f / completed
+        end
+
+        def series_avg(metric = "ms")
+          series[metric].each_with_object(Hash.new(0)) do |(bucket, value), result|
+            completed = series.dig("p", bucket) - series.dig("f", bucket)
+            result[bucket] = (completed == 0) ? 0 : value.to_f / completed
+          end
+        end
+      end
+
+      class MarkResult < Struct.new(:time, :label)
+        def bucket
+          time.strftime("%H:%M")
+        end
+      end
+
+      private
+
+      def fetch_marks(time_range)
+        [].tap do |result|
+          marks = @pool.with { |c| c.hgetall("#{@time.strftime("%Y%m%d")}-marks") }
+
+          marks.each do |timestamp, label|
+            time = Time.parse(timestamp)
+            if time_range.cover? time
+              result << MarkResult.new(time, label)
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/lib/sidekiq/metrics/shared.rb

@@ -0,0 +1,95 @@
+require "concurrent"
+
+module Sidekiq
+  module Metrics
+    # This is the only dependency on concurrent-ruby in Sidekiq but it's
+    # mandatory for thread-safety until MRI supports atomic operations on values.
+    Counter = ::Concurrent::AtomicFixnum
+
+    # Implements space-efficient but statistically useful histogram storage.
+    # A precise time histogram stores every time. Instead we break times into a set of
+    # known buckets and increment counts of the associated time bucket. Even if we call
+    # the histogram a million times, we'll still only store 26 buckets.
+    # NB: needs to be thread-safe or resiliant to races.
+    #
+    # To store this data, we use Redis' BITFIELD command to store unsigned 16-bit counters
+    # per bucket per klass per minute. It's unlikely that most people will be executing more
+    # than 1000 job/sec for a full minute of a specific type.
+    class Histogram
+      include Enumerable
+
+      # This number represents the maximum milliseconds for this bucket.
+      # 20 means all job executions up to 20ms, e.g. if a job takes
+      # 280ms, it'll increment bucket[7]. Note we can track job executions
+      # up to about 5.5 minutes. After that, it's assumed you're probably
+      # not too concerned with its performance.
+      BUCKET_INTERVALS = [
+        20, 30, 45, 65, 100,
+        150, 225, 335, 500, 750,
+        1100, 1700, 2500, 3800, 5750,
+        8500, 13000, 20000, 30000, 45000,
+        65000, 100000, 150000, 225000, 335000,
+        1e20 # the "maybe your job is too long" bucket
+      ].freeze
+      LABELS = [
+        "20ms", "30ms", "45ms", "65ms", "100ms",
+        "150ms", "225ms", "335ms", "500ms", "750ms",
+        "1.1s", "1.7s", "2.5s", "3.8s", "5.75s",
+        "8.5s", "13s", "20s", "30s", "45s",
+        "65s", "100s", "150s", "225s", "335s",
+        "Slow"
+      ].freeze
+      FETCH = "GET u16 #0 GET u16 #1 GET u16 #2 GET u16 #3 \
+        GET u16 #4 GET u16 #5 GET u16 #6 GET u16 #7 \
+        GET u16 #8 GET u16 #9 GET u16 #10 GET u16 #11 \
+        GET u16 #12 GET u16 #13 GET u16 #14 GET u16 #15 \
+        GET u16 #16 GET u16 #17 GET u16 #18 GET u16 #19 \
+        GET u16 #20 GET u16 #21 GET u16 #22 GET u16 #23 \
+        GET u16 #24 GET u16 #25".split
+      HISTOGRAM_TTL = 8 * 60 * 60
+
+      def each
+        buckets.each { |counter| yield counter.value }
+      end
+
+      def label(idx)
+        LABELS[idx]
+      end
+
+      attr_reader :buckets
+      def initialize(klass)
+        @klass = klass
+        @buckets = Array.new(BUCKET_INTERVALS.size) { Counter.new }
+      end
+
+      def record_time(ms)
+        index_to_use = BUCKET_INTERVALS.each_index do |idx|
+          break idx if ms < BUCKET_INTERVALS[idx]
+        end
+
+        @buckets[index_to_use].increment
+      end
+
+      def fetch(conn, now = Time.now)
+        window = now.utc.strftime("%d-%H:%-M")
+        key = "#{@klass}-#{window}"
+        conn.bitfield(key, *FETCH)
+      end
+
+      def persist(conn, now = Time.now)
+        buckets, @buckets = @buckets, []
+        window = now.utc.strftime("%d-%H:%-M")
+        key = "#{@klass}-#{window}"
+        cmd = [key, "OVERFLOW", "SAT"]
+        buckets.each_with_index do |counter, idx|
+          val = counter.value
+          cmd << "INCRBY" << "u16" << "##{idx}" << val.to_s if val > 0
+        end
+
+        conn.bitfield(*cmd) if cmd.size > 3
+        conn.expire(key, HISTOGRAM_TTL)
+        key
+      end
+    end
+  end
+end