sidekiq 5.2.9 → 6.4.1

data/lib/sidekiq/job_retry.rb

@@ -1,6 +1,10 @@
 # frozen_string_literal: true
-require 'sidekiq/scheduled'
-require 'sidekiq/api'
+
+require "sidekiq/scheduled"
+require "sidekiq/api"
+
+require "zlib"
+require "base64"
 
 module Sidekiq
   ##
@@ -30,9 +34,10 @@ module Sidekiq
   # 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
@@ -48,15 +53,17 @@ 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 worker and send retries to
+  # a low priority queue with:
   #
   #    class MyWorker
   #      include Sidekiq::Worker
-  #      sidekiq_options :retry => 10
+  #      sidekiq_options retry: 10, retry_queue: 'low'
   #    end
   #
   class JobRetry
     class Handled < ::RuntimeError; end
+
     class Skip < Handled; end
 
     include Sidekiq::Util
@@ -70,7 +77,7 @@ module Sidekiq
     # 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.
-    def global(msg, queue)
+    def global(jobstr, queue)
       yield
     rescue Handled => ex
       raise ex
@@ -81,22 +88,20 @@ module Sidekiq
       # ignore, will be pushed back onto queue during hard_shutdown
       raise Sidekiq::Shutdown if exception_caused_by_shutdown?(e)
 
-      if msg['retry']
+      msg = Sidekiq.load_json(jobstr)
+      if msg["retry"]
         attempt_retry(nil, msg, queue, e)
       else
         Sidekiq.death_handlers.each do |handler|
-          begin
-            handler.call(msg, e)
-          rescue => handler_ex
-            handle_exception(handler_ex, { context: "Error calling death handler", job: msg })
-          end
+          handler.call(msg, e)
+        rescue => handler_ex
+          handle_exception(handler_ex, {context: "Error calling death handler", job: msg})
         end
       end
 
       raise Handled
     end
 
-
     # The local retry support means that any errors that occur within
     # this block can be associated with the given worker instance.
     # This is required to support the `sidekiq_retries_exhausted` block.
@@ -105,7 +110,7 @@ module Sidekiq
     # 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, msg, queue)
+    def local(worker, jobstr, queue)
       yield
     rescue Handled => ex
       raise ex
@@ -116,11 +121,12 @@ module Sidekiq
       # ignore, will be pushed back onto queue during hard_shutdown
       raise Sidekiq::Shutdown if exception_caused_by_shutdown?(e)
 
-      if msg['retry'] == nil
-        msg['retry'] = worker.class.get_sidekiq_options['retry']
+      msg = Sidekiq.load_json(jobstr)
+      if msg["retry"].nil?
+        msg["retry"] = worker.class.get_sidekiq_options["retry"]
       end
 
-      raise e unless msg['retry']
+      raise e unless msg["retry"]
       attempt_retry(worker, msg, queue, e)
       # We've handled this error associated with this job, don't
       # need to handle it at the global level
@@ -133,13 +139,9 @@ module Sidekiq
     # instantiate the worker instance.  All access must be guarded and
     # best effort.
     def attempt_retry(worker, msg, queue, exception)
-      max_retry_attempts = retry_attempts_from(msg['retry'], @max_retries)
+      max_retry_attempts = retry_attempts_from(msg["retry"], @max_retries)
 
-      msg['queue'] = if msg['retry_queue']
-        msg['retry_queue']
-      else
-        queue
-      end
+      msg["queue"] = (msg["retry_queue"] || queue)
 
       m = exception_message(exception)
       if m.respond_to?(:scrub!)
@@ -147,32 +149,34 @@ module Sidekiq
         m.scrub!
       end
 
-      msg['error_message'] = m
-      msg['error_class'] = exception.class.name
-      count = if msg['retry_count']
-        msg['retried_at'] = Time.now.to_f
-        msg['retry_count'] += 1
+      msg["error_message"] = m
+      msg["error_class"] = exception.class.name
+      count = if msg["retry_count"]
+        msg["retried_at"] = Time.now.to_f
+        msg["retry_count"] += 1
       else
-        msg['failed_at'] = Time.now.to_f
-        msg['retry_count'] = 0
+        msg["failed_at"] = Time.now.to_f
+        msg["retry_count"] = 0
       end
 
-      if msg['backtrace'] == true
-        msg['error_backtrace'] = exception.backtrace
-      elsif !msg['backtrace']
-        # do nothing
-      elsif msg['backtrace'].to_i != 0
-        msg['error_backtrace'] = exception.backtrace[0...msg['backtrace'].to_i]
+      if msg["backtrace"]
+        lines = if msg["backtrace"] == true
+          exception.backtrace
+        else
+          exception.backtrace[0...msg["backtrace"].to_i]
+        end
+
+        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" }
+        # 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)
+          conn.zadd("retry", retry_at.to_s, payload)
         end
       else
         # Goodbye dear message, you (re)tried your best I'm sure.
@@ -182,25 +186,23 @@ module Sidekiq
 
     def retries_exhausted(worker, msg, exception)
       begin
-        block = worker && worker.sidekiq_retries_exhausted_block
-        block.call(msg, exception) if block
+        block = worker&.sidekiq_retries_exhausted_block
+        block&.call(msg, exception)
       rescue => e
-        handle_exception(e, { context: "Error calling retries_exhausted", job: msg })
+        handle_exception(e, {context: "Error calling retries_exhausted", job: msg})
       end
 
+      send_to_morgue(msg) unless msg["dead"] == false
+
       Sidekiq.death_handlers.each do |handler|
-        begin
-          handler.call(msg, exception)
-        rescue => e
-          handle_exception(e, { context: "Error calling death handler", job: msg })
-        end
+        handler.call(msg, exception)
+      rescue => e
+        handle_exception(e, {context: "Error calling death handler", job: msg})
       end
-
-      send_to_morgue(msg) unless msg['dead'] == false
     end
 
     def send_to_morgue(msg)
-      logger.info { "Adding dead #{msg['class']} job #{msg['jid']}" }
+      logger.info { "Adding dead #{msg["class"]} job #{msg["jid"]}" }
       payload = Sidekiq.dump_json(msg)
       DeadSet.new.kill(payload, notify_failure: false)
     end
@@ -214,25 +216,19 @@ module Sidekiq
     end
 
     def delay_for(worker, count, exception)
-      if worker && worker.sidekiq_retry_in_block
+      jitter = rand(10) * (count + 1)
+      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
+        return custom_retry_in + jitter 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))
+      (count**4) + 15 + jitter
     end
 
     def retry_in(worker, count, exception)
-      begin
-        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
+      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 = [])
@@ -249,14 +245,17 @@ module Sidekiq
     # Extract message from exception.
     # Set a default if the message raises an error
     def exception_message(exception)
-      begin
-        # App code can stuff all sorts of crazy binary data into the error message
-        # that won't convert to JSON.
-        exception.message.to_s[0, 10_000]
-      rescue
-        "!!! ERROR MESSAGE THREW AN ERROR !!!".dup
-      end
+      # App code can stuff all sorts of crazy binary data into the error message
+      # that won't convert to JSON.
+      exception.message.to_s[0, 10_000]
+    rescue
+      +"!!! ERROR MESSAGE THREW AN ERROR !!!"
     end
 
+    def compress_backtrace(backtrace)
+      serialized = Sidekiq.dump_json(backtrace)
+      compressed = Zlib::Deflate.deflate(serialized)
+      Base64.encode64(compressed)
+    end
   end
 end

data/lib/sidekiq/job_util.rb

@@ -0,0 +1,65 @@
+require "securerandom"
+require "time"
+
+module Sidekiq
+  module JobUtil
+    # These functions encapsulate various job utilities.
+    # They must be simple and free from side effects.
+
+    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)
+
+      if Sidekiq.options[:on_complex_arguments] == :raise
+        msg = <<~EOM
+          Job arguments to #{item["class"]} must be native JSON types, see https://github.com/mperham/sidekiq/wiki/Best-Practices.
+          To disable this error, remove `Sidekiq.strict_args!` from your initializer.
+        EOM
+        raise(ArgumentError, msg) unless json_safe?(item)
+      elsif Sidekiq.options[:on_complex_arguments] == :warn
+        Sidekiq.logger.warn <<~EOM unless json_safe?(item)
+          Job arguments to #{item["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"] == ""
+
+      item["class"] = item["class"].to_s
+      item["queue"] = item["queue"].to_s
+      item["jid"] ||= SecureRandom.hex(12)
+      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::Worker class, not class name: #{item_class.ancestors.inspect}") unless item_class.respond_to?(:get_sidekiq_options)
+        item_class.get_sidekiq_options
+      else
+        Sidekiq.default_worker_options
+      end
+    end
+
+    private
+
+    def json_safe?(item)
+      JSON.parse(JSON.dump(item["args"])) == item["args"]
+    end
+  end
+end

data/lib/sidekiq/launcher.rb

@@ -1,21 +1,28 @@
 # frozen_string_literal: true
-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
 
-    attr_accessor :manager, :poller, :fetcher
+    STATS_TTL = 5 * 365 * 24 * 60 * 60 # 5 years
 
-    STATS_TTL = 5*365*24*60*60
+    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
@@ -50,7 +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
@@ -62,17 +69,66 @@ module Sidekiq
 
     private unless $TESTING
 
+    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 |pipeline|
+          pipeline.srem("processes", identity)
+          pipeline.unlink("#{identity}:workers")
+        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 self.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
+    at_exit(&method(:flush_stats))
+
     def ❤
       key = identity
       fails = procd = 0
+
       begin
         fails = Processor::FAILURE.reset
         procd = Processor::PROCESSED.reset
@@ -80,94 +136,128 @@ module Sidekiq
 
         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.multi do |transaction|
+            transaction.incrby("stat:processed", procd)
+            transaction.incrby("stat:processed:#{nowdate}", procd)
+            transaction.expire("stat:processed:#{nowdate}", STATS_TTL)
 
-            conn.incrby("stat:failed", fails)
-            conn.incrby("stat:failed:#{nowdate}", fails)
-            conn.expire("stat:failed:#{nowdate}", STATS_TTL)
+            transaction.incrby("stat:failed", fails)
+            transaction.incrby("stat:failed:#{nowdate}", fails)
+            transaction.expire("stat:failed:#{nowdate}", STATS_TTL)
 
-            conn.del(workers_key)
+            transaction.unlink(workers_key)
             curstate.each_pair do |tid, hash|
-              conn.hset(workers_key, tid, Sidekiq.dump_json(hash))
+              transaction.hset(workers_key, tid, Sidekiq.dump_json(hash))
             end
-            conn.expire(workers_key, 60)
+            transaction.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', to_json, 'busy', curstate.size, 'beat', Time.now.to_f, 'quiet', @done)
-            conn.expire(key, 60)
-            conn.rpop("#{key}-signals")
-          end
-        end
+        _, exists, _, _, msg = Sidekiq.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,
+              "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
 
         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.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
+      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 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" => @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