sidekiq 6.1.0 → 7.0.0

data/lib/sidekiq/job_retry.rb

@@ -1,10 +1,8 @@
 # frozen_string_literal: true
 
-require "sidekiq/scheduled"
-require "sidekiq/api"
-
 require "zlib"
 require "base64"
+require "sidekiq/component"
 
 module Sidekiq
   ##
@@ -25,18 +23,19 @@ module Sidekiq
   #
   # A job looks like:
   #
-  #     { 'class' => 'HardWorker', 'args' => [1, 2, 'foo'], 'retry' => true }
+  #     { 'class' => 'HardJob', 'args' => [1, 2, 'foo'], 'retry' => true }
   #
   # The 'retry' option also accepts a number (in place of 'true'):
   #
-  #     { 'class' => 'HardWorker', 'args' => [1, 2, 'foo'], 'retry' => 5 }
+  #     { 'class' => 'HardJob', 'args' => [1, 2, 'foo'], 'retry' => 5 }
   #
   # The job will be retried this number of times before giving up. (If simply
   # 'true', Sidekiq retries 25 times)
   #
-  # We'll add a bit more data to the job to support retries:
+  # Relevant options for job retries:
   #
-  #  * 'queue' - the queue to use
+  #  * 'queue' - the queue for the initial job
+  #  * 'retry_queue' - if job retries should be pushed to a different (e.g. lower priority) queue
   #  * 'retry_count' - number of times we've retried so far.
   #  * 'error_message' - the message from the exception
   #  * 'error_class' - the exception class
@@ -52,28 +51,31 @@ module Sidekiq
   #
   #   Sidekiq.options[:max_retries] = 7
   #
-  # or limit the number of retries for a particular worker with:
+  # or limit the number of retries for a particular job and send retries to
+  # a low priority queue with:
   #
-  #    class MyWorker
-  #      include Sidekiq::Worker
-  #      sidekiq_options :retry => 10
+  #    class MyJob
+  #      include Sidekiq::Job
+  #      sidekiq_options retry: 10, retry_queue: 'low'
   #    end
   #
   class JobRetry
     class Handled < ::RuntimeError; end
+
     class Skip < Handled; end
 
-    include Sidekiq::Util
+    include Sidekiq::Component
 
     DEFAULT_MAX_RETRY_ATTEMPTS = 25
 
-    def initialize(options = {})
-      @max_retries = Sidekiq.options.merge(options).fetch(:max_retries, DEFAULT_MAX_RETRY_ATTEMPTS)
+    def initialize(capsule)
+      @config = @capsule = capsule
+      @max_retries = Sidekiq.default_configuration[:max_retries] || DEFAULT_MAX_RETRY_ATTEMPTS
     end
 
     # The global retry handler requires only the barest of data.
     # We want to be able to retry as much as possible so we don't
-    # require the worker to be instantiated.
+    # require the job to be instantiated.
     def global(jobstr, queue)
       yield
     rescue Handled => ex
@@ -87,9 +89,9 @@ module Sidekiq
 
       msg = Sidekiq.load_json(jobstr)
       if msg["retry"]
-        attempt_retry(nil, msg, queue, e)
+        process_retry(nil, msg, queue, e)
       else
-        Sidekiq.death_handlers.each do |handler|
+        @capsule.config.death_handlers.each do |handler|
           handler.call(msg, e)
         rescue => handler_ex
           handle_exception(handler_ex, {context: "Error calling death handler", job: msg})
@@ -100,14 +102,14 @@ module Sidekiq
     end
 
     # The local retry support means that any errors that occur within
-    # this block can be associated with the given worker instance.
+    # this block can be associated with the given job instance.
     # This is required to support the `sidekiq_retries_exhausted` block.
     #
     # Note that any exception from the block is wrapped in the Skip
     # exception so the global block does not reprocess the error.  The
     # Skip exception is unwrapped within Sidekiq::Processor#process before
     # calling the handle_exception handlers.
-    def local(worker, jobstr, queue)
+    def local(jobinst, jobstr, queue)
       yield
     rescue Handled => ex
       raise ex
@@ -120,11 +122,11 @@ module Sidekiq
 
       msg = Sidekiq.load_json(jobstr)
       if msg["retry"].nil?
-        msg["retry"] = worker.class.get_sidekiq_options["retry"]
+        msg["retry"] = jobinst.class.get_sidekiq_options["retry"]
       end
 
       raise e unless msg["retry"]
-      attempt_retry(worker, msg, queue, e)
+      process_retry(jobinst, msg, queue, e)
       # We've handled this error associated with this job, don't
       # need to handle it at the global level
       raise Skip
@@ -132,10 +134,10 @@ module Sidekiq
 
     private
 
-    # Note that +worker+ can be nil here if an error is raised before we can
-    # instantiate the worker instance.  All access must be guarded and
+    # Note that +jobinst+ can be nil here if an error is raised before we can
+    # instantiate the job instance.  All access must be guarded and
     # best effort.
-    def attempt_retry(worker, msg, queue, exception)
+    def process_retry(jobinst, msg, queue, exception)
       max_retry_attempts = retry_attempts_from(msg["retry"], @max_retries)
 
       msg["queue"] = (msg["retry_queue"] || queue)
@@ -166,24 +168,54 @@ module Sidekiq
         msg["error_backtrace"] = compress_backtrace(lines)
       end
 
-      if count < max_retry_attempts
-        delay = delay_for(worker, count, exception)
-        # Logging here can break retries if the logging device raises ENOSPC #3979
-        # logger.debug { "Failure! Retry #{count} in #{delay} seconds" }
-        retry_at = Time.now.to_f + delay
-        payload = Sidekiq.dump_json(msg)
-        Sidekiq.redis do |conn|
-          conn.zadd("retry", retry_at.to_s, payload)
-        end
-      else
-        # Goodbye dear message, you (re)tried your best I'm sure.
-        retries_exhausted(worker, msg, exception)
+      # Goodbye dear message, you (re)tried your best I'm sure.
+      return retries_exhausted(jobinst, msg, exception) if count >= max_retry_attempts
+
+      strategy, delay = delay_for(jobinst, count, exception)
+      case strategy
+      when :discard
+        return # poof!
+      when :kill
+        return retries_exhausted(jobinst, msg, exception)
+      end
+
+      # Logging here can break retries if the logging device raises ENOSPC #3979
+      # logger.debug { "Failure! Retry #{count} in #{delay} seconds" }
+      jitter = rand(10) * (count + 1)
+      retry_at = Time.now.to_f + delay + jitter
+      payload = Sidekiq.dump_json(msg)
+      redis do |conn|
+        conn.zadd("retry", retry_at.to_s, payload)
       end
     end
 
-    def retries_exhausted(worker, msg, exception)
+    # returns (strategy, seconds)
+    def delay_for(jobinst, count, exception)
+      rv = begin
+        # sidekiq_retry_in can return two different things:
+        # 1. When to retry next, as an integer of seconds
+        # 2. A symbol which re-routes the job elsewhere, e.g. :discard, :kill, :default
+        jobinst&.sidekiq_retry_in_block&.call(count, exception)
+      rescue Exception => e
+        handle_exception(e, {context: "Failure scheduling retry using the defined `sidekiq_retry_in` in #{jobinst.class.name}, falling back to default"})
+        nil
+      end
+
+      delay = (count**4) + 15
+      if Integer === rv && rv > 0
+        delay = rv
+      elsif rv == :discard
+        return [:discard, nil] # do nothing, job goes poof
+      elsif rv == :kill
+        return [:kill, nil]
+      end
+
+      [:default, delay]
+    end
+
+    def retries_exhausted(jobinst, msg, exception)
       begin
-        block = worker&.sidekiq_retries_exhausted_block
+        block = jobinst&.sidekiq_retries_exhausted_block
         block&.call(msg, exception)
       rescue => e
         handle_exception(e, {context: "Error calling retries_exhausted", job: msg})
@@ -191,7 +223,7 @@ module Sidekiq
 
       send_to_morgue(msg) unless msg["dead"] == false
 
-      Sidekiq.death_handlers.each do |handler|
+      @capsule.config.death_handlers.each do |handler|
         handler.call(msg, exception)
       rescue => e
         handle_exception(e, {context: "Error calling death handler", job: msg})
@@ -201,7 +233,15 @@ module Sidekiq
     def send_to_morgue(msg)
       logger.info { "Adding dead #{msg["class"]} job #{msg["jid"]}" }
       payload = Sidekiq.dump_json(msg)
-      DeadSet.new.kill(payload, notify_failure: false)
+      now = Time.now.to_f
+
+      redis do |conn|
+        conn.multi do |xa|
+          xa.zadd("dead", now.to_s, payload)
+          xa.zremrangebyscore("dead", "-inf", now - @capsule.config[:dead_timeout_in_seconds])
+          xa.zremrangebyrank("dead", 0, - @capsule.config[:dead_max_jobs])
+        end
+      end
     end
 
     def retry_attempts_from(msg_retry, default)
@@ -212,26 +252,6 @@ module Sidekiq
       end
     end
 
-    def delay_for(worker, count, exception)
-      if worker&.sidekiq_retry_in_block
-        custom_retry_in = retry_in(worker, count, exception).to_i
-        return custom_retry_in if custom_retry_in > 0
-      end
-      seconds_to_delay(count)
-    end
-
-    # delayed_job uses the same basic formula
-    def seconds_to_delay(count)
-      (count**4) + 15 + (rand(30) * (count + 1))
-    end
-
-    def retry_in(worker, count, exception)
-      worker.sidekiq_retry_in_block.call(count, exception)
-    rescue Exception => e
-      handle_exception(e, {context: "Failure scheduling retry using the defined `sidekiq_retry_in` in #{worker.class.name}, falling back to default"})
-      nil
-    end
-
     def exception_caused_by_shutdown?(e, checked_causes = [])
       return false unless e.cause
 

data/lib/sidekiq/job_util.rb

@@ -0,0 +1,71 @@
+require "securerandom"
+require "time"
+
+module Sidekiq
+  module JobUtil
+    # These functions encapsulate various job utilities.
+
+    TRANSIENT_ATTRIBUTES = %w[]
+
+    def validate(item)
+      raise(ArgumentError, "Job must be a Hash with 'class' and 'args' keys: `#{item}`") unless item.is_a?(Hash) && item.key?("class") && item.key?("args")
+      raise(ArgumentError, "Job args must be an Array: `#{item}`") unless item["args"].is_a?(Array)
+      raise(ArgumentError, "Job class must be either a Class or String representation of the class name: `#{item}`") unless item["class"].is_a?(Class) || item["class"].is_a?(String)
+      raise(ArgumentError, "Job 'at' must be a Numeric timestamp: `#{item}`") if item.key?("at") && !item["at"].is_a?(Numeric)
+      raise(ArgumentError, "Job tags must be an Array: `#{item}`") if item["tags"] && !item["tags"].is_a?(Array)
+    end
+
+    def verify_json(item)
+      job_class = item["wrapped"] || item["class"]
+      if Sidekiq::Config::DEFAULTS[:on_complex_arguments] == :raise
+        msg = <<~EOM
+          Job arguments to #{job_class} must be native JSON types, see https://github.com/mperham/sidekiq/wiki/Best-Practices.
+          To disable this error, add `Sidekiq.strict_args!(false)` to your initializer.
+        EOM
+        raise(ArgumentError, msg) unless json_safe?(item)
+      elsif Sidekiq::Config::DEFAULTS[:on_complex_arguments] == :warn
+        warn <<~EOM unless json_safe?(item)
+          Job arguments to #{job_class} do not serialize to JSON safely. This will raise an error in
+          Sidekiq 7.0. See https://github.com/mperham/sidekiq/wiki/Best-Practices or raise an error today
+          by calling `Sidekiq.strict_args!` during Sidekiq initialization.
+        EOM
+      end
+    end
+
+    def normalize_item(item)
+      validate(item)
+
+      # merge in the default sidekiq_options for the item's class and/or wrapped element
+      # this allows ActiveJobs to control sidekiq_options too.
+      defaults = normalized_hash(item["class"])
+      defaults = defaults.merge(item["wrapped"].get_sidekiq_options) if item["wrapped"].respond_to?(:get_sidekiq_options)
+      item = defaults.merge(item)
+
+      raise(ArgumentError, "Job must include a valid queue name") if item["queue"].nil? || item["queue"] == ""
+
+      # remove job attributes which aren't necessary to persist into Redis
+      TRANSIENT_ATTRIBUTES.each { |key| item.delete(key) }
+
+      item["jid"] ||= SecureRandom.hex(12)
+      item["class"] = item["class"].to_s
+      item["queue"] = item["queue"].to_s
+      item["created_at"] ||= Time.now.to_f
+      item
+    end
+
+    def normalized_hash(item_class)
+      if item_class.is_a?(Class)
+        raise(ArgumentError, "Message must include a Sidekiq::Job class, not class name: #{item_class.ancestors.inspect}") unless item_class.respond_to?(:get_sidekiq_options)
+        item_class.get_sidekiq_options
+      else
+        Sidekiq.default_job_options
+      end
+    end
+
+    private
+
+    def json_safe?(item)
+      JSON.parse(JSON.dump(item["args"])) == item["args"]
+    end
+  end
+end

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]
+      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 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
@@ -69,22 +73,26 @@ module Sidekiq
 
     private unless $TESTING
 
+    BEAT_PAUSE = 10
+
     def start_heartbeat
       loop do
         heartbeat
-        sleep 5
+        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|
-        conn.pipelined do
-          conn.srem("processes", identity)
-          conn.unlink("#{identity}:workers")
+      redis do |conn|
+        conn.pipelined do |pipeline|
+          pipeline.srem("processes", [identity])
+          pipeline.unlink("#{identity}:work")
         end
       end
     rescue
@@ -92,81 +100,77 @@ 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|
-          conn.pipelined do
-            conn.incrby("stat:processed", procd)
-            conn.incrby("stat:processed:#{nowdate}", procd)
-            conn.expire("stat:processed:#{nowdate}", STATS_TTL)
-
-            conn.incrby("stat:failed", fails)
-            conn.incrby("stat:failed:#{nowdate}", fails)
-            conn.expire("stat:failed:#{nowdate}", STATS_TTL)
+        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}")
+        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
-            conn.incrby("stat:processed", procd)
-            conn.incrby("stat:processed:#{nowdate}", procd)
-            conn.expire("stat:processed:#{nowdate}", STATS_TTL)
-
-            conn.incrby("stat:failed", fails)
-            conn.incrby("stat:failed:#{nowdate}", fails)
-            conn.expire("stat:failed:#{nowdate}", STATS_TTL)
-
-            conn.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|
-              conn.hset(workers_key, tid, Sidekiq.dump_json(hash))
+              transaction.hset(work_key, tid, Sidekiq.dump_json(hash))
             end
-            conn.expire(workers_key, 60)
+            transaction.expire(work_key, 60)
           end
         end
 
-        fails = procd = 0
+        rtt = check_rtt
 
-        _, 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, "quiet", @done)
-            conn.expire(key, 60)
-            conn.rpop("#{key}-signals")
+        fails = procd = 0
+        kb = memory_usage(::Process.pid)
+
+        _, 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) unless exists
+        fire_event(:heartbeat) unless exists > 0
+        fire_event(:beat, oneshot: false)
 
         return unless msg
 
@@ -180,27 +184,74 @@ module Sidekiq
       end
     end
 
-    def to_data
-      @data ||= begin
-        {
-          "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
-        }
+    # 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
+      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
+
+    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 memory_usage(pid)
+      MEMORY_GRABBER.call(pid)
+    end
+
+    def to_data
+      @data ||= {
+        "hostname" => hostname,
+        "started_at" => Time.now.to_f,
+        "pid" => ::Process.pid,
+        "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
 
     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
+      # 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