sidekiq 6.0.0 → 6.4.0

data/lib/sidekiq/job_retry.rb

@@ -3,6 +3,9 @@
 require "sidekiq/scheduled"
 require "sidekiq/api"
 
+require "zlib"
+require "base64"
+
 module Sidekiq
   ##
   # Automatically retry jobs that fail in Sidekiq.
@@ -31,9 +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
@@ -49,15 +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
@@ -71,7 +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
@@ -82,6 +88,7 @@ module Sidekiq
       # 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
@@ -103,7 +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
@@ -114,6 +121,7 @@ module Sidekiq
       # 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
@@ -151,12 +159,14 @@ module Sidekiq
         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
@@ -182,13 +192,13 @@ module Sidekiq
         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
-
-      send_to_morgue(msg) unless msg["dead"] == false
     end
 
     def send_to_morgue(msg)
@@ -206,16 +216,12 @@ module Sidekiq
     end
 
     def delay_for(worker, count, exception)
+      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)
@@ -245,5 +251,11 @@ module Sidekiq
     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

@@ -16,12 +16,13 @@ module 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? },
+      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
@@ -56,7 +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
@@ -68,10 +69,12 @@ module Sidekiq
 
     private unless $TESTING
 
+    BEAT_PAUSE = 5
+
     def start_heartbeat
       loop do
         heartbeat
-        sleep 5
+        sleep BEAT_PAUSE
       end
       Sidekiq.logger.info("Heartbeat stopping...")
     end
@@ -83,7 +86,7 @@ module Sidekiq
       Sidekiq.redis do |conn|
         conn.pipelined do
           conn.srem("processes", identity)
-          conn.del("#{identity}:workers")
+          conn.unlink("#{identity}:workers")
         end
       end
     rescue
@@ -96,6 +99,32 @@ module Sidekiq
       ❤
     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 = identity
       fails = procd = 0
@@ -118,7 +147,7 @@ module Sidekiq
             conn.incrby("stat:failed:#{nowdate}", fails)
             conn.expire("stat:failed:#{nowdate}", STATS_TTL)
 
-            conn.del(workers_key)
+            conn.unlink(workers_key)
             curstate.each_pair do |tid, hash|
               conn.hset(workers_key, tid, Sidekiq.dump_json(hash))
             end
@@ -126,18 +155,24 @@ module Sidekiq
           end
         end
 
+        rtt = check_rtt
+
         fails = procd = 0
+        kb = memory_usage(::Process.pid)
 
         _, exists, _, _, msg = Sidekiq.redis { |conn|
-          res = conn.multi {
+          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.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")
           }
-
-          res
         }
 
         # first heartbeat or recovering from an outage and need to reestablish our heartbeat
@@ -148,34 +183,81 @@ module Sidekiq
         ::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 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
+      Sidekiq.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 }
+        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
+
+    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" => @options[:tag] || "",
+        "concurrency" => @options[:concurrency],
+        "queues" => @options[:queues].uniq,
+        "labels" => @options[:labels],
+        "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

data/lib/sidekiq/logger.rb

@@ -4,22 +4,110 @@ require "logger"
 require "time"
 
 module Sidekiq
-  class Logger < ::Logger
-    def initialize(*args)
-      super
+  module Context
+    def self.with(hash)
+      orig_context = current.dup
+      current.merge!(hash)
+      yield
+    ensure
+      Thread.current[:sidekiq_context] = orig_context
+    end
 
-      self.formatter = Sidekiq.log_formatter
+    def self.current
+      Thread.current[:sidekiq_context] ||= {}
+    end
+  end
+
+  module LoggingUtils
+    LEVELS = {
+      "debug" => 0,
+      "info" => 1,
+      "warn" => 2,
+      "error" => 3,
+      "fatal" => 4
+    }
+    LEVELS.default_proc = proc do |_, level|
+      Sidekiq.logger.warn("Invalid log level: #{level.inspect}")
+      nil
+    end
+
+    def debug?
+      level <= 0
+    end
+
+    def info?
+      level <= 1
+    end
+
+    def warn?
+      level <= 2
+    end
+
+    def error?
+      level <= 3
+    end
+
+    def fatal?
+      level <= 4
     end
 
-    def with_context(hash)
-      ctx.merge!(hash)
+    def local_level
+      Thread.current[:sidekiq_log_level]
+    end
+
+    def local_level=(level)
+      case level
+      when Integer
+        Thread.current[:sidekiq_log_level] = level
+      when Symbol, String
+        Thread.current[:sidekiq_log_level] = LEVELS[level.to_s]
+      when nil
+        Thread.current[:sidekiq_log_level] = nil
+      else
+        raise ArgumentError, "Invalid log level: #{level.inspect}"
+      end
+    end
+
+    def level
+      local_level || super
+    end
+
+    # Change the thread-local level for the duration of the given block.
+    def log_at(level)
+      old_local_level = local_level
+      self.local_level = level
       yield
     ensure
-      hash.keys.each { |key| ctx.delete(key) }
+      self.local_level = old_local_level
     end
 
-    def ctx
-      Thread.current[:sidekiq_context] ||= {}
+    # Redefined to check severity against #level, and thus the thread-local level, rather than +@level+.
+    # FIXME: Remove when the minimum Ruby version supports overriding Logger#level.
+    def add(severity, message = nil, progname = nil, &block)
+      severity ||= ::Logger::UNKNOWN
+      progname ||= @progname
+
+      return true if @logdev.nil? || severity < level
+
+      if message.nil?
+        if block
+          message = yield
+        else
+          message = progname
+          progname = @progname
+        end
+      end
+
+      @logdev.write format_message(format_severity(severity), Time.now, progname, message)
+    end
+  end
+
+  class Logger < ::Logger
+    include LoggingUtils
+
+    def initialize(*args, **kwargs)
+      super
+      self.formatter = Sidekiq.log_formatter
     end
 
     module Formatters
@@ -29,11 +117,20 @@ module Sidekiq
         end
 
         def ctx
-          Thread.current[:sidekiq_context] ||= {}
+          Sidekiq::Context.current
         end
 
         def format_context
-          " " + ctx.compact.map { |k, v| "#{k}=#{v}" }.join(" ") if ctx.any?
+          if ctx.any?
+            " " + ctx.compact.map { |k, v|
+              case v
+              when Array
+                "#{k}=#{v.join(",")}"
+              else
+                "#{k}=#{v}"
+              end
+            }.join(" ")
+          end
         end
       end
 
@@ -56,7 +153,7 @@ module Sidekiq
             pid: ::Process.pid,
             tid: tid,
             lvl: severity,
-            msg: message,
+            msg: message
           }
           c = ctx
           hash["ctx"] = c unless c.empty?

data/lib/sidekiq/manager.rb

@@ -35,7 +35,7 @@ module Sidekiq
       @done = false
       @workers = Set.new
       @count.times do
-        @workers << Processor.new(self)
+        @workers << Processor.new(self, options)
       end
       @plock = Mutex.new
     end
@@ -55,9 +55,6 @@ module Sidekiq
       fire_event(:quiet, reverse: true)
     end
 
-    # hack for quicker development / testing environment #2774
-    PAUSE_TIME = STDOUT.tty? ? 0.1 : 0.5
-
     def stop(deadline)
       quiet
       fire_event(:shutdown, reverse: true)
@@ -69,12 +66,7 @@ module Sidekiq
       return if @workers.empty?
 
       logger.info { "Pausing to allow workers to finish..." }
-      remaining = deadline - ::Process.clock_gettime(::Process::CLOCK_MONOTONIC)
-      while remaining > PAUSE_TIME
-        return if @workers.empty?
-        sleep PAUSE_TIME
-        remaining = deadline - ::Process.clock_gettime(::Process::CLOCK_MONOTONIC)
-      end
+      wait_for(deadline) { @workers.empty? }
       return if @workers.empty?
 
       hard_shutdown
@@ -90,7 +82,7 @@ module Sidekiq
       @plock.synchronize do
         @workers.delete(processor)
         unless @done
-          p = Processor.new(self)
+          p = Processor.new(self, options)
           @workers << p
           p.start
         end
@@ -123,13 +115,19 @@ module Sidekiq
         # contract says that jobs are run AT LEAST once. Process termination
         # is delayed until we're certain the jobs are back in Redis because
         # it is worse to lose a job than to run it twice.
-        strategy = (@options[:fetch] || Sidekiq::BasicFetch)
+        strategy = @options[:fetch]
         strategy.bulk_requeue(jobs, @options)
       end
 
       cleanup.each do |processor|
         processor.kill
       end
+
+      # when this method returns, we immediately call `exit` which may not give
+      # the remaining threads time to run `ensure` blocks, etc. We pause here up
+      # to 3 seconds to give threads a minimal amount of time to run `ensure` blocks.
+      deadline = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC) + 3
+      wait_for(deadline) { @workers.empty? }
     end
   end
 end