sidekiq 4.2.4 → 6.2.1

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 +1,263 @@
+# 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

data/lib/sidekiq/launcher.rb

@@ -1,20 +1,28 @@
 # frozen_string_literal: true
-# encoding: utf-8
-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
 
+    STATS_TTL = 5 * 365 * 24 * 60 * 60 # 5 years
+
+    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
@@ -39,7 +47,7 @@ module Sidekiq
     # return until all work is complete and cleaned up.
     # It can take up to the timeout to complete.
     def stop
-      deadline = Time.now + @options[:timeout]
+      deadline = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC) + @options[:timeout]
 
       @done = true
       @manager.quiet
@@ -49,7 +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
@@ -61,104 +69,195 @@ module Sidekiq
 
     private unless $TESTING
 
-    JVM_RESERVED_SIGNALS = ['USR1', 'USR2'] # Don't Process#kill if we get these signals via the API
+    def start_heartbeat
+      loop do
+        heartbeat
+        sleep 5
+      end
+      Sidekiq.logger.info("Heartbeat stopping...")
+    end
 
-    def heartbeat(k, data, json)
-      results = Sidekiq::CLI::PROCTITLES.map {|x| x.(self, data) }
-      results.compact!
-      $0 = results.join(' ')
+    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.unlink("#{identity}:workers")
+        end
+      end
+    rescue
+      # best effort, ignore network errors
+    end
+
+    def heartbeat
+      $0 = PROCTITLES.map { |proc| proc.call(self, to_data) }.compact.join(" ")
 
-      ❤(k, json)
+      ❤
+    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
+            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)
+          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, 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
+        curstate = Processor::WORKER_STATE.dup
+
+        workers_key = "#{key}:workers"
+        nowdate = Time.now.utc.strftime("%Y-%m-%d")
 
-        workers_key = "#{key}:workers".freeze
-        nowdate = Time.now.utc.strftime("%Y-%m-%d".freeze)
         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.del(workers_key)
-            Processor::WORKER_STATE.each_pair do |tid, hash|
+            conn.expire("stat:failed:#{nowdate}", STATS_TTL)
+
+            conn.unlink(workers_key)
+            curstate.each_pair do |tid, hash|
               conn.hset(workers_key, tid, Sidekiq.dump_json(hash))
             end
             conn.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', json, 'busy', Processor::WORKER_STATE.size, 'beat', Time.now.to_f, 'quiet', @done)
+        _, 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,
+              "rtt_us", rtt,
+              "quiet", @done,
+              "rss", kb)
             conn.expire(key, 60)
             conn.rpop("#{key}-signals")
-          end
-        end
+          }
+        }
 
         # 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
 
-        if JVM_RESERVED_SIGNALS.include?(msg)
-          Sidekiq::CLI.instance.handle_signal(msg)
-        else
-          ::Process.kill(msg, $$)
-        end
+        ::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.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)
+    # 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
 
-      while true
-        heartbeat(k, data, json)
-        sleep 5
+    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.
+        EOM
+        RTT_READINGS.reset
+      end
+      rtt
     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")
+    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 ||= 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
+        }
       end
-    rescue
-      # best effort, ignore network errors
     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
   end
 end