sidekiq 4.2.10 → 6.1.2

data/lib/sidekiq/extensions/generic_proxy.rb

@@ -1,10 +1,13 @@
 # frozen_string_literal: true
-require 'yaml'
+
+require "yaml"
 
 module Sidekiq
   module Extensions
+    SIZE_LIMIT = 8_192
+
     class Proxy < BasicObject
-      def initialize(performable, target, options={})
+      def initialize(performable, target, options = {})
         @performable = performable
         @target = target
         @opts = options
@@ -17,9 +20,12 @@ module Sidekiq
         # to JSON and then deserialized on the other side back into a
         # Ruby object.
         obj = [@target, name, args]
-        @performable.client_push({ 'class' => @performable, 'args' => [::YAML.dump(obj)] }.merge(@opts))
+        marshalled = ::YAML.dump(obj)
+        if marshalled.size > SIZE_LIMIT
+          ::Sidekiq.logger.warn { "#{@target}.#{name} job argument is #{marshalled.bytesize} bytes, you should refactor it to reduce the size" }
+        end
+        @performable.client_push({"class" => @performable, "args" => [marshalled]}.merge(@opts))
       end
     end
-
   end
 end

data/lib/sidekiq/fetch.rb

@@ -1,5 +1,6 @@
 # frozen_string_literal: true
-require 'sidekiq'
+
+require "sidekiq"
 
 module Sidekiq
   class BasicFetch
@@ -7,27 +8,29 @@ module Sidekiq
     # can check if the process is shutting down.
     TIMEOUT = 2
 
-    UnitOfWork = Struct.new(:queue, :job) do
+    UnitOfWork = Struct.new(:queue, :job) {
       def acknowledge
         # nothing to do
       end
 
       def queue_name
-        queue.sub(/.*queue:/, ''.freeze)
+        queue.delete_prefix("queue:")
       end
 
       def requeue
         Sidekiq.redis do |conn|
-          conn.rpush("queue:#{queue_name}", job)
+          conn.rpush(queue, job)
         end
       end
-    end
+    }
 
     def initialize(options)
-      @strictly_ordered_queues = !!options[:strict]
-      @queues = options[:queues].map { |q| "queue:#{q}" }
+      raise ArgumentError, "missing queue list" unless options[:queues]
+      @options = options
+      @strictly_ordered_queues = !!@options[:strict]
+      @queues = @options[:queues].map { |q| "queue:#{q}" }
       if @strictly_ordered_queues
-        @queues = @queues.uniq
+        @queues.uniq!
         @queues << TIMEOUT
       end
     end
@@ -37,38 +40,20 @@ module Sidekiq
       UnitOfWork.new(*work) if work
     end
 
-    # Creating the Redis#brpop command takes into account any
-    # configured queue weights. By default Redis#brpop returns
-    # data from the first queue that has pending elements. We
-    # recreate the queue command each time we invoke Redis#brpop
-    # to honor weights and avoid queue starvation.
-    def queues_cmd
-      if @strictly_ordered_queues
-        @queues
-      else
-        queues = @queues.shuffle.uniq
-        queues << TIMEOUT
-        queues
-      end
-    end
-
-
-    # By leaving this as a class method, it can be pluggable and used by the Manager actor. Making it
-    # an instance method will make it async to the Fetcher actor
-    def self.bulk_requeue(inprogress, options)
+    def bulk_requeue(inprogress, options)
       return if inprogress.empty?
 
       Sidekiq.logger.debug { "Re-queueing terminated jobs" }
       jobs_to_requeue = {}
       inprogress.each do |unit_of_work|
-        jobs_to_requeue[unit_of_work.queue_name] ||= []
-        jobs_to_requeue[unit_of_work.queue_name] << unit_of_work.job
+        jobs_to_requeue[unit_of_work.queue] ||= []
+        jobs_to_requeue[unit_of_work.queue] << unit_of_work.job
       end
 
       Sidekiq.redis do |conn|
         conn.pipelined do
           jobs_to_requeue.each do |queue, jobs|
-            conn.rpush("queue:#{queue}", jobs)
+            conn.rpush(queue, jobs)
           end
         end
       end
@@ -77,5 +62,19 @@ module Sidekiq
       Sidekiq.logger.warn("Failed to requeue #{inprogress.size} jobs: #{ex.message}")
     end
 
+    # Creating the Redis#brpop command takes into account any
+    # configured queue weights. By default Redis#brpop returns
+    # data from the first queue that has pending elements. We
+    # recreate the queue command each time we invoke Redis#brpop
+    # to honor weights and avoid queue starvation.
+    def queues_cmd
+      if @strictly_ordered_queues
+        @queues
+      else
+        queues = @queues.shuffle!.uniq
+        queues << TIMEOUT
+        queues
+      end
+    end
   end
 end

data/lib/sidekiq/job_logger.rb

@@ -0,0 +1,63 @@
+# frozen_string_literal: true
+
+module Sidekiq
+  class JobLogger
+    def initialize(logger = Sidekiq.logger)
+      @logger = logger
+    end
+
+    def call(item, queue)
+      start = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC)
+      @logger.info("start")
+
+      yield
+
+      with_elapsed_time_context(start) do
+        @logger.info("done")
+      end
+    rescue Exception
+      with_elapsed_time_context(start) do
+        @logger.info("fail")
+      end
+
+      raise
+    end
+
+    def prepare(job_hash, &block)
+      level = job_hash["log_level"]
+      if level
+        @logger.log_at(level) do
+          Sidekiq::Context.with(job_hash_context(job_hash), &block)
+        end
+      else
+        Sidekiq::Context.with(job_hash_context(job_hash), &block)
+      end
+    end
+
+    def job_hash_context(job_hash)
+      # If we're using a wrapper class, like ActiveJob, use the "wrapped"
+      # attribute to expose the underlying thing.
+      h = {
+        class: job_hash["wrapped"] || job_hash["class"],
+        jid: job_hash["jid"]
+      }
+      h[:bid] = job_hash["bid"] if job_hash["bid"]
+      h[:tags] = job_hash["tags"] if job_hash["tags"]
+      h
+    end
+
+    def with_elapsed_time_context(start, &block)
+      Sidekiq::Context.with(elapsed_time_context(start), &block)
+    end
+
+    def elapsed_time_context(start)
+      {elapsed: elapsed(start).to_s}
+    end
+
+    private
+
+    def elapsed(start)
+      (::Process.clock_gettime(::Process::CLOCK_MONOTONIC) - start).round(3)
+    end
+  end
+end

data/lib/sidekiq/job_retry.rb

@@ -0,0 +1,262 @@
+# frozen_string_literal: true
+
+require "sidekiq/scheduled"
+require "sidekiq/api"
+
+require "zlib"
+require "base64"
+
+module Sidekiq
+  ##
+  # Automatically retry jobs that fail in Sidekiq.
+  # Sidekiq's retry support assumes a typical development lifecycle:
+  #
+  #   0. Push some code changes with a bug in it.
+  #   1. Bug causes job processing to fail, Sidekiq's middleware captures
+  #      the job and pushes it onto a retry queue.
+  #   2. Sidekiq retries jobs in the retry queue multiple times with
+  #      an exponential delay, the job continues to fail.
+  #   3. After a few days, a developer deploys a fix. The job is
+  #      reprocessed successfully.
+  #   4. Once retries are exhausted, Sidekiq will give up and move the
+  #      job to the Dead Job Queue (aka morgue) where it must be dealt with
+  #      manually in the Web UI.
+  #   5. After 6 months on the DJQ, Sidekiq will discard the job.
+  #
+  # A job looks like:
+  #
+  #     { 'class' => 'HardWorker', '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 }
+  #
+  # 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:
+  #
+  #  * 'queue' - the queue to use
+  #  * 'retry_count' - number of times we've retried so far.
+  #  * 'error_message' - the message from the exception
+  #  * 'error_class' - the exception class
+  #  * 'failed_at' - the first time it failed
+  #  * 'retried_at' - the last time it was retried
+  #  * 'backtrace' - the number of lines of error backtrace to store
+  #
+  # We don't store the backtrace by default as that can add a lot of overhead
+  # to the job and everyone is using an error service, right?
+  #
+  # The default number of retries is 25 which works out to about 3 weeks
+  # You can change the default maximum number of retries in your initializer:
+  #
+  #   Sidekiq.options[:max_retries] = 7
+  #
+  # or limit the number of retries for a particular worker with:
+  #
+  #    class MyWorker
+  #      include Sidekiq::Worker
+  #      sidekiq_options :retry => 10
+  #    end
+  #
+  class JobRetry
+    class Handled < ::RuntimeError; end
+    class Skip < Handled; end
+
+    include Sidekiq::Util
+
+    DEFAULT_MAX_RETRY_ATTEMPTS = 25
+
+    def initialize(options = {})
+      @max_retries = Sidekiq.options.merge(options).fetch(: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.
+    def global(jobstr, queue)
+      yield
+    rescue Handled => ex
+      raise ex
+    rescue Sidekiq::Shutdown => ey
+      # ignore, will be pushed back onto queue during hard_shutdown
+      raise ey
+    rescue Exception => e
+      # ignore, will be pushed back onto queue during hard_shutdown
+      raise Sidekiq::Shutdown if exception_caused_by_shutdown?(e)
+
+      msg = Sidekiq.load_json(jobstr)
+      if msg["retry"]
+        attempt_retry(nil, msg, queue, e)
+      else
+        Sidekiq.death_handlers.each do |handler|
+          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.
+    #
+    # 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)
+      yield
+    rescue Handled => ex
+      raise ex
+    rescue Sidekiq::Shutdown => ey
+      # ignore, will be pushed back onto queue during hard_shutdown
+      raise ey
+    rescue Exception => e
+      # ignore, will be pushed back onto queue during hard_shutdown
+      raise Sidekiq::Shutdown if exception_caused_by_shutdown?(e)
+
+      msg = Sidekiq.load_json(jobstr)
+      if msg["retry"].nil?
+        msg["retry"] = worker.class.get_sidekiq_options["retry"]
+      end
+
+      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
+      raise Skip
+    end
+
+    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
+    # best effort.
+    def attempt_retry(worker, msg, queue, exception)
+      max_retry_attempts = retry_attempts_from(msg["retry"], @max_retries)
+
+      msg["queue"] = (msg["retry_queue"] || queue)
+
+      m = exception_message(exception)
+      if m.respond_to?(:scrub!)
+        m.force_encoding("utf-8")
+        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
+      else
+        msg["failed_at"] = Time.now.to_f
+        msg["retry_count"] = 0
+      end
+
+      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" }
+        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)
+      end
+    end
+
+    def retries_exhausted(worker, msg, exception)
+      begin
+        block = worker&.sidekiq_retries_exhausted_block
+        block&.call(msg, exception)
+      rescue => e
+        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|
+        handler.call(msg, exception)
+      rescue => e
+        handle_exception(e, {context: "Error calling death handler", job: msg})
+      end
+    end
+
+    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)
+    end
+
+    def retry_attempts_from(msg_retry, default)
+      if msg_retry.is_a?(Integer)
+        msg_retry
+      else
+        default
+      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
+
+      # Handle circular causes
+      checked_causes << e.object_id
+      return false if checked_causes.include?(e.cause.object_id)
+
+      e.cause.instance_of?(Sidekiq::Shutdown) ||
+        exception_caused_by_shutdown?(e.cause, checked_causes)
+    end
+
+    # Extract message from exception.
+    # Set a default if the message raises an error
+    def exception_message(exception)
+      # 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