sidekiq 6.4.1 → 7.0.0

data/lib/sidekiq/launcher.rb

@@ -1,13 +1,14 @@
 # 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 Util
+    include Sidekiq::Component
 
     STATS_TTL = 5 * 365 * 24 * 60 * 60 # 5 years
 
@@ -15,50 +16,53 @@ module Sidekiq
       proc { "sidekiq" },
       proc { Sidekiq::VERSION },
       proc { |me, data| data["tag"] },
-      proc { |me, data| "[#{Processor::WORKER_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)
-      options[:fetch] ||= BasicFetch.new(options)
-      @manager = Sidekiq::Manager.new(options)
-      @poller = Sidekiq::Scheduled::Poller.new
+    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
-      @options = options
     end
 
     def run
+      Sidekiq.freeze!
       @thread = safe_thread("heartbeat", &method(:start_heartbeat))
       @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 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 = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC) + @options[:timeout]
-
-      @done = true
-      @manager.quiet
-      @poller.terminate
+      deadline = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC) + @config[:timeout]
 
-      @manager.stop(deadline)
+      quiet
+      stoppers = @managers.map do |mgr|
+        Thread.new do
+          mgr.stop(deadline)
+        end
+      end
 
-      # 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]
-      strategy.bulk_requeue([], @options)
+      fire_event(:shutdown, reverse: true)
+      stoppers.each(&:join)
 
       clear_heartbeat
     end
@@ -69,24 +73,26 @@ module Sidekiq
 
     private unless $TESTING
 
-    BEAT_PAUSE = 5
+    BEAT_PAUSE = 10
 
     def start_heartbeat
       loop do
         heartbeat
         sleep BEAT_PAUSE
       end
-      Sidekiq.logger.info("Heartbeat stopping...")
+      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.
-      Sidekiq.redis do |conn|
+      redis do |conn|
         conn.pipelined do |pipeline|
-          pipeline.srem("processes", identity)
-          pipeline.unlink("#{identity}:workers")
+          pipeline.srem("processes", [identity])
+          pipeline.unlink("#{identity}:work")
         end
       end
     rescue
@@ -94,19 +100,19 @@ module Sidekiq
     end
 
     def heartbeat
-      $0 = PROCTITLES.map { |proc| proc.call(self, to_data) }.compact.join(" ")
+      $0 = PROCTITLES.map { |proc| proc.call(self, to_data) }.compact.join(" ") unless @embedded
 
       ❤
     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)
@@ -118,40 +124,27 @@ 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::WORKER_STATE.dup
-
-        workers_key = "#{key}:workers"
-        nowdate = Time.now.utc.strftime("%Y-%m-%d")
-
-        Sidekiq.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)
-
-            transaction.unlink(workers_key)
+        flush_stats
+
+        curstate = Processor::WORK_STATE.dup
+        redis do |conn|
+          # 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(workers_key, tid, Sidekiq.dump_json(hash))
+              transaction.hset(work_key, tid, Sidekiq.dump_json(hash))
             end
-            transaction.expire(workers_key, 60)
+            transaction.expire(work_key, 60)
           end
         end
 
@@ -160,15 +153,15 @@ module Sidekiq
         fails = procd = 0
         kb = memory_usage(::Process.pid)
 
-        _, exists, _, _, msg = Sidekiq.redis { |conn|
+        _, exists, _, _, msg = 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,
               "rtt_us", rtt,
-              "quiet", @done,
+              "quiet", @done.to_s,
               "rss", kb)
             transaction.expire(key, 60)
             transaction.rpop("#{key}-signals")
@@ -176,7 +169,8 @@ 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
 
@@ -198,7 +192,7 @@ module Sidekiq
 
     def check_rtt
       a = b = 0
-      Sidekiq.redis do |x|
+      redis do |x|
         a = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC, :microsecond)
         x.ping
         b = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC, :microsecond)
@@ -209,12 +203,12 @@ module Sidekiq
       # 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
+        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
+          CPU-saturated; reduce your concurrency and/or see https://github.com/mperham/sidekiq/discussions/5039
         EOM
         RTT_READINGS.reset
       end
@@ -246,10 +240,10 @@ module Sidekiq
         "hostname" => hostname,
         "started_at" => Time.now.to_f,
         "pid" => ::Process.pid,
-        "tag" => @options[:tag] || "",
-        "concurrency" => @options[:concurrency],
-        "queues" => @options[:queues].uniq,
-        "labels" => @options[:labels],
+        "tag" => @config[:tag] || "",
+        "concurrency" => @config.total_concurrency,
+        "queues" => @config.capsules.values.map { |cap| cap.queues }.flatten.uniq,
+        "labels" => @config[:labels].to_a,
         "identity" => identity
       }
     end

data/lib/sidekiq/logger.rb

@@ -18,7 +18,7 @@ module Sidekiq
     end
 
     def self.add(k, v)
-      Thread.current[:sidekiq_context][k] = v
+      current[k] = v
     end
   end
 
@@ -31,28 +31,14 @@ 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
 
-    def debug?
-      level <= 0
-    end
-
-    def info?
-      level <= 1
-    end
-
-    def warn?
-      level <= 2
-    end
-
-    def error?
-      level <= 3
-    end
-
-    def fatal?
-      level <= 4
+    LEVELS.each do |level, numeric_level|
+      define_method("#{level}?") do
+        local_level.nil? ? super() : local_level <= numeric_level
+      end
     end
 
     def local_level
@@ -84,36 +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

@@ -1,8 +1,6 @@
 # frozen_string_literal: true
 
-require "sidekiq/util"
 require "sidekiq/processor"
-require "sidekiq/fetch"
 require "set"
 
 module Sidekiq
@@ -21,43 +19,38 @@ 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
+    attr_reader :capsule
 
-    def initialize(options = {})
-      logger.debug { options.inspect }
-      @options = options
-      @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(self, options)
+        @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 }
-      fire_event(:quiet, reverse: true)
+      logger.info { "Terminating quiet threads for #{capsule.name} capsule" }
+      @workers.each(&:terminate)
     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
@@ -65,24 +58,20 @@ module Sidekiq
       sleep PAUSE_TIME
       return if @workers.empty?
 
-      logger.info { "Pausing to allow workers to finish..." }
+      logger.info { "Pausing to allow jobs to finish..." }
       wait_for(deadline) { @workers.empty? }
       return if @workers.empty?
 
       hard_shutdown
+    ensure
+      capsule.stop
     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, options)
+          p = Processor.new(@config, &method(:processor_result))
           @workers << p
           p.start
         end
@@ -96,7 +85,7 @@ 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
@@ -106,17 +95,16 @@ module Sidekiq
       if cleanup.size > 0
         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]
-        strategy.bulk_requeue(jobs, @options)
+        capsule.fetcher.bulk_requeue(jobs)
       end
 
       cleanup.each do |processor|
@@ -129,5 +117,18 @@ module Sidekiq
       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

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)
+            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