sidekiq 3.4.1 → 7.3.0

data/lib/sidekiq/monitor.rb

@@ -0,0 +1,146 @@
+#!/usr/bin/env ruby
+
+require "fileutils"
+require "sidekiq/api"
+
+class Sidekiq::Monitor
+  class Status
+    VALID_SECTIONS = %w[all version overview processes queues]
+    COL_PAD = 2
+
+    def display(section = nil)
+      section ||= "all"
+      unless VALID_SECTIONS.include? section
+        puts "I don't know how to check the status of '#{section}'!"
+        puts "Try one of these: #{VALID_SECTIONS.join(", ")}"
+        return
+      end
+      send(section)
+    end
+
+    def all
+      version
+      puts
+      overview
+      puts
+      processes
+      puts
+      queues
+    end
+
+    def version
+      puts "Sidekiq #{Sidekiq::VERSION}"
+      puts Time.now.utc
+    end
+
+    def overview
+      puts "---- Overview ----"
+      puts "  Processed: #{delimit stats.processed}"
+      puts "     Failed: #{delimit stats.failed}"
+      puts "       Busy: #{delimit stats.workers_size}"
+      puts "   Enqueued: #{delimit stats.enqueued}"
+      puts "    Retries: #{delimit stats.retry_size}"
+      puts "  Scheduled: #{delimit stats.scheduled_size}"
+      puts "       Dead: #{delimit stats.dead_size}"
+    end
+
+    def processes
+      puts "---- Processes (#{process_set.size}) ----"
+      process_set.each_with_index do |process, index|
+        # Keep compatibility with legacy versions since we don't want to break sidekiqmon during rolling upgrades or downgrades.
+        #
+        # Before:
+        #   ["default", "critical"]
+        #
+        # After:
+        #   {"default" => 1, "critical" => 10}
+        queues =
+          if process["weights"]
+            process["weights"].sort_by { |queue| queue[0] }.map { |capsule| capsule.map { |name, weight| (weight > 0) ? "#{name}: #{weight}" : name }.join(", ") }
+          else
+            process["queues"].sort
+          end
+
+        puts "#{process["identity"]} #{tags_for(process)}"
+        puts "  Started: #{Time.at(process["started_at"])} (#{time_ago(process["started_at"])})"
+        puts "  Threads: #{process["concurrency"]} (#{process["busy"]} busy)"
+        puts "   Queues: #{split_multiline(queues, pad: 11)}"
+        puts "  Version: #{process["version"] || "Unknown"}" if process["version"] != Sidekiq::VERSION
+        puts "" unless (index + 1) == process_set.size
+      end
+    end
+
+    def queues
+      puts "---- Queues (#{queue_data.size}) ----"
+      columns = {
+        name: [:ljust, (["name"] + queue_data.map(&:name)).map(&:length).max + COL_PAD],
+        size: [:rjust, (["size"] + queue_data.map(&:size)).map(&:length).max + COL_PAD],
+        latency: [:rjust, (["latency"] + queue_data.map(&:latency)).map(&:length).max + COL_PAD]
+      }
+      columns.each { |col, (dir, width)| print col.to_s.upcase.public_send(dir, width) }
+      puts
+      queue_data.each do |q|
+        columns.each do |col, (dir, width)|
+          print q.send(col).public_send(dir, width)
+        end
+        puts
+      end
+    end
+
+    private
+
+    def delimit(number)
+      number.to_s.reverse.scan(/.{1,3}/).join(",").reverse
+    end
+
+    def split_multiline(values, opts = {})
+      return "none" unless values
+      pad = opts[:pad] || 0
+      max_length = opts[:max_length] || (80 - pad)
+      out = []
+      line = ""
+      values.each do |value|
+        if (line.length + value.length) > max_length
+          out << line
+          line = " " * pad
+        end
+        line << value + ", "
+      end
+      out << line[0..-3]
+      out.join("\n")
+    end
+
+    def tags_for(process)
+      tags = [
+        process["tag"],
+        process["labels"],
+        ((process["quiet"] == "true") ? "quiet" : nil)
+      ].flatten.compact
+      tags.any? ? "[#{tags.join("] [")}]" : nil
+    end
+
+    def time_ago(timestamp)
+      seconds = Time.now - Time.at(timestamp)
+      return "just now" if seconds < 60
+      return "a minute ago" if seconds < 120
+      return "#{seconds.floor / 60} minutes ago" if seconds < 3600
+      return "an hour ago" if seconds < 7200
+      "#{seconds.floor / 60 / 60} hours ago"
+    end
+
+    QUEUE_STRUCT = Struct.new(:name, :size, :latency)
+    def queue_data
+      @queue_data ||= Sidekiq::Queue.all.map { |q|
+        QUEUE_STRUCT.new(q.name, q.size.to_s, sprintf("%#.2f", q.latency))
+      }
+    end
+
+    def process_set
+      @process_set ||= Sidekiq::ProcessSet.new
+    end
+
+    def stats
+      @stats ||= Sidekiq::Stats.new
+    end
+  end
+end

data/lib/sidekiq/paginator.rb

@@ -1,8 +1,9 @@
+# frozen_string_literal: true
+
 module Sidekiq
   module Paginator
-
-    def page(key, pageidx=1, page_size=25, opts=nil)
-      current_page = pageidx.to_i < 1 ? 1 : pageidx.to_i
+    def page(key, pageidx = 1, page_size = 25, opts = nil)
+      current_page = (pageidx.to_i < 1) ? 1 : pageidx.to_i
       pageidx = current_page - 1
       total_size = 0
       items = []
@@ -11,26 +12,31 @@ module Sidekiq
 
       Sidekiq.redis do |conn|
         type = conn.type(key)
+        rev = opts && opts[:reverse]
 
         case type
-        when 'zset'
-          rev = opts && opts[:reverse]
-          total_size, items = conn.multi do
-            conn.zcard(key)
+        when "zset"
+          total_size, items = conn.multi { |transaction|
+            transaction.zcard(key)
             if rev
-              conn.zrevrange(key, starting, ending, :with_scores => true)
+              transaction.zrange(key, starting, ending, "REV", "withscores")
             else
-              conn.zrange(key, starting, ending, :with_scores => true)
+              transaction.zrange(key, starting, ending, "withscores")
             end
-          end
+          }
           [current_page, total_size, items]
-        when 'list'
-          total_size, items = conn.multi do
-            conn.llen(key)
-            conn.lrange(key, starting, ending)
-          end
+        when "list"
+          total_size, items = conn.multi { |transaction|
+            transaction.llen(key)
+            if rev
+              transaction.lrange(key, -ending - 1, -starting - 1)
+            else
+              transaction.lrange(key, starting, ending)
+            end
+          }
+          items.reverse! if rev
           [current_page, total_size, items]
-        when 'none'
+        when "none"
           [1, 0, []]
         else
           raise "can't page a #{type}"
@@ -38,5 +44,12 @@ module Sidekiq
       end
     end
 
+    def page_items(items, pageidx = 1, page_size = 25)
+      current_page = (pageidx.to_i < 1) ? 1 : pageidx.to_i
+      pageidx = current_page - 1
+      starting = pageidx * page_size
+      items = items.to_a
+      [current_page, items.size, items[starting, page_size]]
+    end
   end
 end

data/lib/sidekiq/processor.rb

@@ -1,155 +1,291 @@
-require 'sidekiq/util'
-require 'sidekiq/actor'
+# frozen_string_literal: true
 
-require 'sidekiq/middleware/server/retry_jobs'
-require 'sidekiq/middleware/server/logging'
+require "sidekiq/fetch"
+require "sidekiq/job_logger"
+require "sidekiq/job_retry"
 
 module Sidekiq
   ##
-  # The Processor receives a message from the Manager and actually
-  # processes it.  It instantiates the worker, runs the middleware
-  # chain and then calls Sidekiq::Worker#perform.
+  # The Processor is a standalone thread which:
+  #
+  # 1. fetches a job from Redis
+  # 2. executes the job
+  #   a. instantiate the job class
+  #   b. run the middleware chain
+  #   c. call #perform
+  #
+  # A Processor can exit due to shutdown or due to
+  # an error during job execution.
+  #
+  # If an error occurs in the job execution, the
+  # Processor calls the Manager to create a new one
+  # to replace itself and exits.
+  #
   class Processor
-    # To prevent a memory leak, ensure that stats expire. However, they should take up a minimal amount of storage
-    # so keep them around for a long time
-    STATS_TIMEOUT = 24 * 60 * 60 * 365 * 5
-
-    include Util
-    include Actor
-
-    def self.default_middleware
-      Middleware::Chain.new do |m|
-        m.add Middleware::Server::Logging
-        m.add Middleware::Server::RetryJobs
-        if defined?(::ActiveRecord::Base)
-          require 'sidekiq/middleware/server/active_record'
-          m.add Sidekiq::Middleware::Server::ActiveRecord
-        end
-      end
+    include Sidekiq::Component
+
+    attr_reader :thread
+    attr_reader :job
+    attr_reader :capsule
+
+    def initialize(capsule, &block)
+      @config = @capsule = capsule
+      @callback = block
+      @down = false
+      @done = false
+      @job = nil
+      @thread = nil
+      @reloader = Sidekiq.default_configuration[:reloader]
+      @job_logger = (capsule.config[:job_logger] || Sidekiq::JobLogger).new(capsule.config)
+      @retrier = Sidekiq::JobRetry.new(capsule)
     end
 
-    attr_accessor :proxy_id
+    def terminate(wait = false)
+      @done = true
+      return unless @thread
+      @thread.value if wait
+    end
 
-    def initialize(boss)
-      @boss = boss
+    def kill(wait = false)
+      @done = true
+      return unless @thread
+      # unlike the other actors, terminate does not wait
+      # for the thread to finish because we don't know how
+      # long the job will take to finish.  Instead we
+      # provide a `kill` method to call after the shutdown
+      # timeout passes.
+      @thread.raise ::Sidekiq::Shutdown
+      @thread.value if wait
     end
 
-    def process(work)
-      msgstr = work.message
-      queue = work.queue_name
+    def stopping?
+      @done
+    end
 
-      @boss.async.real_thread(proxy_id, Thread.current)
+    def start
+      @thread ||= safe_thread("#{config.name}/processor", &method(:run))
+    end
 
-      ack = true
-      begin
-        msg = Sidekiq.load_json(msgstr)
-        klass  = msg['class'].constantize
-        worker = klass.new
-        worker.jid = msg['jid']
-
-        stats(worker, msg, queue) do
-          Sidekiq.server_middleware.invoke(worker, msg, queue) do
-            execute_job(worker, cloned(msg['args']))
-          end
-        end
-      rescue Sidekiq::Shutdown
-        # Had to force kill this job because it didn't finish
-        # within the timeout.  Don't acknowledge the work since
-        # we didn't properly finish it.
-        ack = false
-      rescue Exception => ex
-        handle_exception(ex, msg || { :message => msgstr })
-        raise
-      ensure
-        work.acknowledge if ack
-      end
+    private unless $TESTING
+
+    def run
+      # By setting this thread-local, Sidekiq.redis will access +Sidekiq::Capsule#redis_pool+
+      # instead of the global pool in +Sidekiq::Config#redis_pool+.
+      Thread.current[:sidekiq_capsule] = @capsule
 
-      @boss.async.processor_done(current_actor)
+      process_one until @done
+      @callback.call(self)
+    rescue Sidekiq::Shutdown
+      @callback.call(self)
+    rescue Exception => ex
+      @callback.call(self, ex)
     end
 
-    def inspect
-      "<Processor##{object_id.to_s(16)}>"
+    def process_one(&block)
+      @job = fetch
+      process(@job) if @job
+      @job = nil
     end
 
-    def execute_job(worker, cloned_args)
-      worker.perform(*cloned_args)
+    def get_one
+      uow = capsule.fetcher.retrieve_work
+      if @down
+        logger.info { "Redis is online, #{::Process.clock_gettime(::Process::CLOCK_MONOTONIC) - @down} sec downtime" }
+        @down = nil
+      end
+      uow
+    rescue Sidekiq::Shutdown
+    rescue => ex
+      handle_fetch_exception(ex)
     end
 
-    private
+    def fetch
+      j = get_one
+      if j && @done
+        j.requeue
+        nil
+      else
+        j
+      end
+    end
 
-    def thread_identity
-      @str ||= Thread.current.object_id.to_s(36)
+    def handle_fetch_exception(ex)
+      unless @down
+        @down = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC)
+        logger.error("Error fetching job: #{ex}")
+        handle_exception(ex)
+      end
+      sleep(1)
+      nil
     end
 
-    def stats(worker, msg, queue)
-      # Do not conflate errors from the job with errors caused by updating
-      # stats so calling code can react appropriately
-      retry_and_suppress_exceptions do
-        hash = Sidekiq.dump_json({:queue => queue, :payload => msg, :run_at => Time.now.to_i })
-        Sidekiq.redis do |conn|
-          conn.multi do
-            conn.hmset("#{identity}:workers", thread_identity, hash)
-            conn.expire("#{identity}:workers", 60*60*4)
+    def dispatch(job_hash, queue, jobstr)
+      # since middleware can mutate the job hash
+      # we need to clone it to report the original
+      # job structure to the Web UI
+      # or to push back to redis when retrying.
+      # To avoid costly and, most of the time, useless cloning here,
+      # we pass original String of JSON to respected methods
+      # to re-parse it there if we need access to the original, untouched job
+
+      @job_logger.prepare(job_hash) do
+        @retrier.global(jobstr, queue) do
+          @job_logger.call(job_hash, queue) do
+            stats(jobstr, 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 = Object.const_get(job_hash["class"])
+                inst = klass.new
+                inst.jid = job_hash["jid"]
+                inst._context = self
+                @retrier.local(inst, jobstr, queue) do
+                  yield inst
+                end
+              end
+            end
           end
         end
       end
+    end
+
+    IGNORE_SHUTDOWN_INTERRUPTS = {Sidekiq::Shutdown => :never}
+    private_constant :IGNORE_SHUTDOWN_INTERRUPTS
+    ALLOW_SHUTDOWN_INTERRUPTS = {Sidekiq::Shutdown => :immediate}
+    private_constant :ALLOW_SHUTDOWN_INTERRUPTS
+
+    def process(uow)
+      jobstr = uow.job
+      queue = uow.queue_name
 
+      # Treat malformed JSON as a special case: job goes straight to the morgue.
+      job_hash = nil
       begin
-        yield
-      rescue Exception
-        retry_and_suppress_exceptions do
-          failed = "stat:failed:#{Time.now.utc.to_date}"
-          Sidekiq.redis do |conn|
-            conn.multi do
-              conn.incrby("stat:failed", 1)
-              conn.incrby(failed, 1)
-              conn.expire(failed, STATS_TIMEOUT)
-            end
+        job_hash = Sidekiq.load_json(jobstr)
+      rescue => ex
+        handle_exception(ex, {context: "Invalid JSON for job", jobstr: jobstr})
+        now = Time.now.to_f
+        redis do |conn|
+          conn.multi do |xa|
+            xa.zadd("dead", now.to_s, jobstr)
+            xa.zremrangebyscore("dead", "-inf", now - @capsule.config[:dead_timeout_in_seconds])
+            xa.zremrangebyrank("dead", 0, - @capsule.config[:dead_max_jobs])
           end
         end
-        raise
-      ensure
-        retry_and_suppress_exceptions do
-          processed = "stat:processed:#{Time.now.utc.to_date}"
-          Sidekiq.redis do |conn|
-            conn.multi do
-              conn.hdel("#{identity}:workers", thread_identity)
-              conn.incrby("stat:processed", 1)
-              conn.incrby(processed, 1)
-              conn.expire(processed, STATS_TIMEOUT)
+        return uow.acknowledge
+      end
+
+      ack = false
+      Thread.handle_interrupt(IGNORE_SHUTDOWN_INTERRUPTS) do
+        Thread.handle_interrupt(ALLOW_SHUTDOWN_INTERRUPTS) do
+          dispatch(job_hash, queue, jobstr) do |inst|
+            config.server_middleware.invoke(inst, job_hash, queue) do
+              execute_job(inst, job_hash["args"])
             end
           end
+          ack = true
+        rescue Sidekiq::Shutdown
+          # Had to force kill this job because it didn't finish
+          # within the timeout.  Don't acknowledge the work since
+          # we didn't properly finish it.
+        rescue Sidekiq::JobRetry::Skip => s
+          # Skip means we handled this error elsewhere. We don't
+          # need to log or report the error.
+          ack = true
+          raise s
+        rescue Sidekiq::JobRetry::Handled => h
+          # this is the common case: job raised error and Sidekiq::JobRetry::Handled
+          # signals that we created a retry successfully.  We can acknowledge the job.
+          ack = true
+          e = h.cause || h
+          handle_exception(e, {context: "Job raised exception", job: job_hash})
+          raise e
+        rescue Exception => ex
+          # Unexpected error!  This is very bad and indicates an exception that got past
+          # the retry subsystem (e.g. network partition).  We won't acknowledge the job
+          # so it can be rescued when using Sidekiq Pro.
+          handle_exception(ex, {context: "Internal exception!", job: job_hash, jobstr: jobstr})
+          raise ex
+        end
+      ensure
+        if ack
+          uow.acknowledge
         end
       end
     end
 
-    # Deep clone the arguments passed to the worker so that if
-    # the message fails, what is pushed back onto Redis hasn't
-    # been mutated by the worker.
-    def cloned(ary)
-      Marshal.load(Marshal.dump(ary))
+    def execute_job(inst, cloned_args)
+      inst.perform(*cloned_args)
     end
 
-    # If an exception occurs in the block passed to this method, that block will be retried up to max_retries times.
-    # All exceptions will be swallowed and logged.
-    def retry_and_suppress_exceptions(max_retries = 5)
-      retry_count = 0
-      begin
-        yield
-      rescue => e
-        retry_count += 1
-        if retry_count <= max_retries
-          Sidekiq.logger.debug {"Suppressing and retrying error: #{e.inspect}"}
-          pause_for_recovery(retry_count)
-          retry
-        else
-          handle_exception(e, { :message => "Exhausted #{max_retries} retries"})
-        end
+    # 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 += amount }
+      end
+
+      def reset
+        @lock.synchronize {
+          val = @value
+          @value = 0
+          val
+        }
+      end
+    end
+
+    # jruby's Hash implementation is not threadsafe, so we wrap it in a mutex here
+    class SharedWorkState
+      def initialize
+        @work_state = {}
+        @lock = Mutex.new
+      end
+
+      def set(tid, hash)
+        @lock.synchronize { @work_state[tid] = hash }
+      end
+
+      def delete(tid)
+        @lock.synchronize { @work_state.delete(tid) }
+      end
+
+      def dup
+        @lock.synchronize { @work_state.dup }
+      end
+
+      def size
+        @lock.synchronize { @work_state.size }
+      end
+
+      def clear
+        @lock.synchronize { @work_state.clear }
       end
     end
 
-    def pause_for_recovery(retry_count)
-      sleep(retry_count)
+    PROCESSED = Counter.new
+    FAILURE = Counter.new
+    WORK_STATE = SharedWorkState.new
+
+    def stats(jobstr, queue)
+      WORK_STATE.set(tid, {queue: queue, payload: jobstr, run_at: Time.now.to_i})
+
+      begin
+        yield
+      rescue Exception
+        FAILURE.incr
+        raise
+      ensure
+        WORK_STATE.delete(tid)
+        PROCESSED.incr
+      end
     end
   end
 end