sidekiq 4.2.4 → 5.2.0

data/lib/sidekiq/job_retry.rb

@@ -0,0 +1,241 @@
+# frozen_string_literal: true
+require 'sidekiq/scheduled'
+require 'sidekiq/api'
+
+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 Skip < ::RuntimeError; 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(msg, queue)
+      yield
+    rescue Skip => 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)
+
+      raise e unless msg['retry']
+      attempt_retry(nil, msg, queue, e)
+      raise e
+    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, msg, queue)
+      yield
+    rescue Skip => 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)
+
+      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'] = if msg['retry_queue']
+        msg['retry_queue']
+      else
+        queue
+      end
+
+      # App code can stuff all sorts of crazy binary data into the error message
+      # that won't convert to JSON.
+      m = exception.message.to_s[0, 10_000]
+      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'] == 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]
+      end
+
+      if count < max_retry_attempts
+        delay = delay_for(worker, count, exception)
+        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)
+      logger.debug { "Retries exhausted for job" }
+      begin
+        block = worker && worker.sidekiq_retries_exhausted_block
+        block.call(msg, exception) if block
+      rescue => e
+        handle_exception(e, { context: "Error calling retries_exhausted", job: msg })
+      end
+
+      Sidekiq.death_handlers.each do |handler|
+        begin
+          handler.call(msg, exception)
+        rescue => e
+          handle_exception(e, { context: "Error calling death handler", job: msg })
+        end
+      end
+
+      send_to_morgue(msg) unless msg['dead'] == false
+    end
+
+    def send_to_morgue(msg)
+      Sidekiq.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 && 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)
+      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
+    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
+
+  end
+end

data/lib/sidekiq/launcher.rb

@@ -1,5 +1,4 @@
 # frozen_string_literal: true
-# encoding: utf-8
 require 'sidekiq/manager'
 require 'sidekiq/fetch'
 require 'sidekiq/scheduled'
@@ -13,6 +12,8 @@ module Sidekiq
     include Util
 
     attr_accessor :manager, :poller, :fetcher
+    
+    STATS_TTL = 5*365*24*60*60
 
     def initialize(options)
       @manager = Sidekiq::Manager.new(options)
@@ -61,30 +62,33 @@ module Sidekiq
 
     private unless $TESTING
 
-    JVM_RESERVED_SIGNALS = ['USR1', 'USR2'] # Don't Process#kill if we get these signals via the API
-
-    def heartbeat(k, data, json)
-      results = Sidekiq::CLI::PROCTITLES.map {|x| x.(self, data) }
+    def heartbeat
+      results = Sidekiq::CLI::PROCTITLES.map {|x| x.(self, to_data) }
       results.compact!
       $0 = results.join(' ')
 
-      ❤(k, json)
+      ❤
     end
 
-    def ❤(key, json)
+    def ❤
+      key = identity
       fails = procd = 0
       begin
-        Processor::FAILURE.update {|curr| fails = curr; 0 }
-        Processor::PROCESSED.update {|curr| procd = curr; 0 }
+        fails = Processor::FAILURE.reset
+        procd = Processor::PROCESSED.reset
 
-        workers_key = "#{key}:workers".freeze
-        nowdate = Time.now.utc.strftime("%Y-%m-%d".freeze)
+        workers_key = "#{key}:workers"
+        nowdate = Time.now.utc.strftime("%Y-%m-%d")
         Sidekiq.redis do |conn|
           conn.multi do
-            conn.incrby("stat:processed".freeze, procd)
+            conn.incrby("stat:processed", procd)
             conn.incrby("stat:processed:#{nowdate}", procd)
-            conn.incrby("stat:failed".freeze, fails)
+            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.del(workers_key)
             Processor::WORKER_STATE.each_pair do |tid, hash|
               conn.hset(workers_key, tid, Sidekiq.dump_json(hash))
@@ -98,7 +102,7 @@ module Sidekiq
           conn.multi do
             conn.sadd('processes', key)
             conn.exists(key)
-            conn.hmset(key, 'info', json, 'busy', Processor::WORKER_STATE.size, 'beat', Time.now.to_f, 'quiet', @done)
+            conn.hmset(key, 'info', to_json, 'busy', Processor::WORKER_STATE.size, 'beat', Time.now.to_f, 'quiet', @done)
             conn.expire(key, 60)
             conn.rpop("#{key}-signals")
           end
@@ -109,43 +113,47 @@ module Sidekiq
 
         return unless msg
 
-        if JVM_RESERVED_SIGNALS.include?(msg)
-          Sidekiq::CLI.instance.handle_signal(msg)
-        else
-          ::Process.kill(msg, $$)
-        end
+        ::Process.kill(msg, $$)
       rescue => e
         # ignore all redis/network issues
         logger.error("heartbeat: #{e.message}")
         # don't lose the counts if there was a network issue
-        Processor::PROCESSED.increment(procd)
-        Processor::FAILURE.increment(fails)
+        Processor::PROCESSED.incr(procd)
+        Processor::FAILURE.incr(fails)
       end
     end
 
     def start_heartbeat
-      k = identity
-      data = {
-        'hostname' => hostname,
-        'started_at' => Time.now.to_f,
-        'pid' => $$,
-        'tag' => @options[:tag] || '',
-        'concurrency' => @options[:concurrency],
-        'queues' => @options[:queues].uniq,
-        'labels' => @options[:labels],
-        'identity' => k,
-      }
-      # this data doesn't change so dump it to a string
-      # now so we don't need to dump it every heartbeat.
-      json = Sidekiq.dump_json(data)
-
       while true
-        heartbeat(k, data, json)
+        heartbeat
         sleep 5
       end
       Sidekiq.logger.info("Heartbeat stopping...")
     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
+    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
+    end
+
     def clear_heartbeat
       # Remove record from Redis since we are shutting down.
       # Note we don't stop the heartbeat thread; if the process

data/lib/sidekiq/logging.rb

@@ -11,7 +11,7 @@ module Sidekiq
 
       # Provide a call() method that returns the formatted message.
       def call(severity, time, program_name, message)
-        "#{time.utc.iso8601(3)} #{::Process.pid} TID-#{Thread.current.object_id.to_s(36)}#{context} #{severity}: #{message}\n"
+        "#{time.utc.iso8601(3)} #{::Process.pid} TID-#{Sidekiq::Logging.tid}#{context} #{severity}: #{message}\n"
       end
 
       def context
@@ -22,10 +22,26 @@ module Sidekiq
 
     class WithoutTimestamp < Pretty
       def call(severity, time, program_name, message)
-        "#{::Process.pid} TID-#{Thread.current.object_id.to_s(36)}#{context} #{severity}: #{message}\n"
+        "#{::Process.pid} TID-#{Sidekiq::Logging.tid}#{context} #{severity}: #{message}\n"
       end
     end
 
+    def self.tid
+      Thread.current['sidekiq_tid'] ||= (Thread.current.object_id ^ ::Process.pid).to_s(36)
+    end
+
+    def self.job_hash_context(job_hash)
+      # If we're using a wrapper class, like ActiveJob, use the "wrapped"
+      # attribute to expose the underlying thing.
+      klass = job_hash['wrapped'] || job_hash["class"]
+      bid = job_hash['bid']
+      "#{klass} JID-#{job_hash['jid']}#{" BID-#{bid}" if bid}"
+    end
+
+    def self.with_job_hash_context(job_hash, &block)
+      with_context(job_hash_context(job_hash), &block)
+    end
+
     def self.with_context(msg)
       Thread.current[:sidekiq_context] ||= []
       Thread.current[:sidekiq_context] << msg

data/lib/sidekiq/manager.rb

@@ -1,5 +1,4 @@
 # frozen_string_literal: true
-# encoding: utf-8
 require 'sidekiq/util'
 require 'sidekiq/processor'
 require 'sidekiq/fetch'
@@ -10,7 +9,7 @@ module Sidekiq
 
   ##
   # The Manager is the central coordination point in Sidekiq, controlling
-  # the lifecycle of the Processors and feeding them jobs as necessary.
+  # the lifecycle of the Processors.
   #
   # Tasks:
   #
@@ -54,7 +53,7 @@ module Sidekiq
 
       logger.info { "Terminating quiet workers" }
       @workers.each { |x| x.terminate }
-      fire_event(:quiet, true)
+      fire_event(:quiet, reverse: true)
     end
 
     # hack for quicker development / testing environment #2774
@@ -62,7 +61,7 @@ module Sidekiq
 
     def stop(deadline)
       quiet
-      fire_event(:shutdown, true)
+      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

data/lib/sidekiq/middleware/server/active_record.rb

@@ -1,7 +1,17 @@
+# frozen_string_literal: true
 module Sidekiq
   module Middleware
     module Server
       class ActiveRecord
+
+        def initialize
+          # With Rails 5+ we must use the Reloader **always**.
+          # The reloader handles code loading and db connection management.
+          if defined?(::Rails) && ::Rails::VERSION::MAJOR >= 5
+            raise ArgumentError, "Rails 5 no longer needs or uses the ActiveRecord middleware."
+          end
+        end
+
         def call(*args)
           yield
         ensure

data/lib/sidekiq/processor.rb

@@ -1,9 +1,9 @@
 # frozen_string_literal: true
 require 'sidekiq/util'
 require 'sidekiq/fetch'
+require 'sidekiq/job_logger'
+require 'sidekiq/job_retry'
 require 'thread'
-require 'concurrent/map'
-require 'concurrent/atomic/atomic_fixnum'
 
 module Sidekiq
   ##
@@ -37,6 +37,8 @@ module Sidekiq
       @thread = nil
       @strategy = (mgr.options[:fetch] || Sidekiq::BasicFetch).new(mgr.options)
       @reloader = Sidekiq.options[:reloader]
+      @logging = (mgr.options[:job_logger] || Sidekiq::JobLogger).new
+      @retrier = Sidekiq::JobRetry.new
     end
 
     def terminate(wait=false)
@@ -107,36 +109,63 @@ module Sidekiq
       if !@down
         @down = Time.now
         logger.error("Error fetching job: #{ex}")
-        ex.backtrace.each do |bt|
-          logger.error(bt)
-        end
+        handle_exception(ex)
       end
       sleep(1)
       nil
     end
 
+    def dispatch(job_hash, queue)
+      # since middleware can mutate the job hash
+      # we clone here so we report the original
+      # job structure to the Web UI
+      pristine = cloned(job_hash)
+
+      Sidekiq::Logging.with_job_hash_context(job_hash) do
+        @retrier.global(pristine, queue) do
+          @logging.call(job_hash, queue) do
+            stats(pristine, queue) do
+              # Rails 5 requires a Reloader to wrap code execution.  In order to
+              # constantize the worker and instantiate an instance, we have to call
+              # the Reloader.  It handles code loading, db connection management, etc.
+              # Effectively this block denotes a "unit of work" to Rails.
+              @reloader.call do
+                klass  = constantize(job_hash['class'])
+                worker = klass.new
+                worker.jid = job_hash['jid']
+                @retrier.local(worker, pristine, queue) do
+                  yield worker
+                end
+              end
+            end
+          end
+        end
+      end
+    end
+
     def process(work)
       jobstr = work.job
       queue = work.queue_name
 
       ack = false
       begin
-        @reloader.call do
-          job = Sidekiq.load_json(jobstr)
-          klass  = job['class'.freeze].constantize
-          worker = klass.new
-          worker.jid = job['jid'.freeze]
-
-          stats(worker, job, queue) do
-            Sidekiq.server_middleware.invoke(worker, job, queue) do
-              # Only ack if we either attempted to start this job or
-              # successfully completed it. This prevents us from
-              # losing jobs if a middleware raises an exception before yielding
-              ack = true
-              execute_job(worker, cloned(job['args'.freeze]))
-            end
-          end
+        # Treat malformed JSON as a special case: job goes straight to the morgue.
+        job_hash = nil
+        begin
+          job_hash = Sidekiq.load_json(jobstr)
+        rescue => ex
+          handle_exception(ex, { :context => "Invalid JSON for job", :jobstr => jobstr })
+          # we can't notify because the job isn't a valid hash payload.
+          DeadSet.new.kill(jobstr, notify_failure: false)
           ack = true
+          raise
+        end
+
+        ack = true
+        dispatch(job_hash, queue) do |worker|
+          Sidekiq.server_middleware.invoke(worker, job_hash, queue) do
+            execute_job(worker, cloned(job_hash['args']))
+          end
         end
       rescue Sidekiq::Shutdown
         # Had to force kill this job because it didn't finish
@@ -144,8 +173,9 @@ module Sidekiq
         # we didn't properly finish it.
         ack = false
       rescue Exception => ex
-        handle_exception(ex, { :context => "Job raised exception", :job => job, :jobstr => jobstr })
-        raise
+        e = ex.is_a?(::Sidekiq::JobRetry::Skip) && ex.cause ? ex.cause : ex
+        handle_exception(e, { :context => "Job raised exception", :job => job_hash, :jobstr => jobstr })
+        raise e
       ensure
         work.acknowledge if ack
       end
@@ -155,34 +185,61 @@ module Sidekiq
       worker.perform(*cloned_args)
     end
 
-    def thread_identity
-      @str ||= Thread.current.object_id.to_s(36)
+    # Ruby doesn't provide atomic counters out of the box so we'll
+    # implement something simple ourselves.
+    # https://bugs.ruby-lang.org/issues/14706
+    class Counter
+      def initialize
+        @value = 0
+        @lock = Mutex.new
+      end
+
+      def incr(amount=1)
+        @lock.synchronize { @value = @value + amount }
+      end
+
+      def reset
+        @lock.synchronize { val = @value; @value = 0; val }
+      end
     end
 
-    WORKER_STATE = Concurrent::Map.new
-    PROCESSED = Concurrent::AtomicFixnum.new
-    FAILURE = Concurrent::AtomicFixnum.new
+    PROCESSED = Counter.new
+    FAILURE = Counter.new
+    # This is mutable global state but because each thread is storing
+    # its own unique key/value, there's no thread-safety issue AFAIK.
+    WORKER_STATE = {}
 
-    def stats(worker, job, queue)
-      tid = thread_identity
-      WORKER_STATE[tid] = {:queue => queue, :payload => cloned(job), :run_at => Time.now.to_i }
+    def stats(job_hash, queue)
+      tid = Sidekiq::Logging.tid
+      WORKER_STATE[tid] = {:queue => queue, :payload => job_hash, :run_at => Time.now.to_i }
 
       begin
         yield
       rescue Exception
-        FAILURE.increment
+        FAILURE.incr
         raise
       ensure
         WORKER_STATE.delete(tid)
-        PROCESSED.increment
+        PROCESSED.incr
       end
     end
 
     # Deep clone the arguments passed to the worker so that if
     # the job fails, what is pushed back onto Redis hasn't
     # been mutated by the worker.
-    def cloned(ary)
-      Marshal.load(Marshal.dump(ary))
+    def cloned(thing)
+      Marshal.load(Marshal.dump(thing))
+    end
+
+    def constantize(str)
+      names = str.split('::')
+      names.shift if names.empty? || names.first.empty?
+
+      names.inject(Object) do |constant, name|
+        # the false flag limits search for name to under the constant namespace
+        #   which mimics Rails' behaviour
+        constant.const_defined?(name, false) ? constant.const_get(name, false) : constant.const_missing(name)
+      end
     end
 
   end