sidekiq 6.1.1 → 6.5.10

data/lib/sidekiq/launcher.rb

@@ -3,11 +3,12 @@
 require "sidekiq/manager"
 require "sidekiq/fetch"
 require "sidekiq/scheduled"
+require "sidekiq/ring_buffer"
 
 module Sidekiq
   # The Launcher starts the Manager and Poller threads and provides the process heartbeat.
   class Launcher
-    include Util
+    include Sidekiq::Component
 
     STATS_TTL = 5 * 365 * 24 * 60 * 60 # 5 years
 
@@ -15,18 +16,18 @@ module Sidekiq
       proc { "sidekiq" },
       proc { Sidekiq::VERSION },
       proc { |me, data| data["tag"] },
-      proc { |me, data| "[#{Processor::WORKER_STATE.size} of #{data["concurrency"]} busy]" },
+      proc { |me, data| "[#{Processor::WORK_STATE.size} of #{data["concurrency"]} busy]" },
       proc { |me, data| "stopping" if me.stopping? }
     ]
 
     attr_accessor :manager, :poller, :fetcher
 
     def initialize(options)
+      @config = options
       options[:fetch] ||= BasicFetch.new(options)
       @manager = Sidekiq::Manager.new(options)
-      @poller = Sidekiq::Scheduled::Poller.new
+      @poller = Sidekiq::Scheduled::Poller.new(options)
       @done = false
-      @options = options
     end
 
     def run
@@ -43,11 +44,9 @@ module Sidekiq
       @poller.terminate
     end
 
-    # Shuts down the process.  This method does not
-    # return until all work is complete and cleaned up.
-    # It can take up to the timeout to complete.
+    # Shuts down this Sidekiq instance. Waits up to the deadline for all jobs to complete.
     def stop
-      deadline = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC) + @options[:timeout]
+      deadline = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC) + @config[:timeout]
 
       @done = true
       @manager.quiet
@@ -55,10 +54,10 @@ module Sidekiq
 
       @manager.stop(deadline)
 
-      # Requeue everything in case there was a worker who grabbed work while stopped
+      # Requeue everything in case there was a thread which fetched a job while the process was stopped.
       # This call is a no-op in Sidekiq but necessary for Sidekiq Pro.
-      strategy = @options[:fetch]
-      strategy.bulk_requeue([], @options)
+      strategy = @config[:fetch]
+      strategy.bulk_requeue([], @config)
 
       clear_heartbeat
     end
@@ -69,22 +68,26 @@ 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...")
+      logger.info("Heartbeat stopping...")
     end
 
     def clear_heartbeat
+      flush_stats
+
       # 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")
+      redis do |conn|
+        conn.pipelined do |pipeline|
+          pipeline.srem("processes", [identity])
+          pipeline.unlink("#{identity}:work")
         end
       end
     rescue
@@ -97,7 +100,7 @@ module Sidekiq
       ❤
     end
 
-    def self.flush_stats
+    def flush_stats
       fails = Processor::FAILURE.reset
       procd = Processor::PROCESSED.reset
       return if fails + procd == 0
@@ -105,14 +108,14 @@ module Sidekiq
       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)
+          conn.pipelined do |pipeline|
+            pipeline.incrby("stat:processed", procd)
+            pipeline.incrby("stat:processed:#{nowdate}", procd)
+            pipeline.expire("stat:processed:#{nowdate}", STATS_TTL)
+
+            pipeline.incrby("stat:failed", fails)
+            pipeline.incrby("stat:failed:#{nowdate}", fails)
+            pipeline.expire("stat:failed:#{nowdate}", STATS_TTL)
           end
         end
       rescue => ex
@@ -121,7 +124,6 @@ module Sidekiq
         Sidekiq.logger.warn("Unable to flush stats: #{ex}")
       end
     end
-    at_exit(&method(:flush_stats))
 
     def ❤
       key = identity
@@ -130,43 +132,55 @@ module Sidekiq
       begin
         fails = Processor::FAILURE.reset
         procd = Processor::PROCESSED.reset
-        curstate = Processor::WORKER_STATE.dup
+        curstate = Processor::WORK_STATE.dup
 
-        workers_key = "#{key}:workers"
         nowdate = Time.now.utc.strftime("%Y-%m-%d")
 
-        Sidekiq.redis do |conn|
-          conn.multi do
-            conn.incrby("stat:processed", procd)
-            conn.incrby("stat:processed:#{nowdate}", procd)
-            conn.expire("stat:processed:#{nowdate}", STATS_TTL)
+        redis do |conn|
+          conn.multi do |transaction|
+            transaction.incrby("stat:processed", procd)
+            transaction.incrby("stat:processed:#{nowdate}", procd)
+            transaction.expire("stat:processed:#{nowdate}", STATS_TTL)
 
-            conn.incrby("stat:failed", fails)
-            conn.incrby("stat:failed:#{nowdate}", fails)
-            conn.expire("stat:failed:#{nowdate}", STATS_TTL)
+            transaction.incrby("stat:failed", fails)
+            transaction.incrby("stat:failed:#{nowdate}", fails)
+            transaction.expire("stat:failed:#{nowdate}", STATS_TTL)
+          end
 
-            conn.unlink(workers_key)
+          # work is the current set of executing jobs
+          work_key = "#{key}:work"
+          conn.pipelined do |transaction|
+            transaction.unlink(work_key)
             curstate.each_pair do |tid, hash|
-              conn.hset(workers_key, tid, Sidekiq.dump_json(hash))
+              transaction.hset(work_key, tid, Sidekiq.dump_json(hash))
             end
-            conn.expire(workers_key, 60)
+            transaction.expire(work_key, 60)
           end
         end
 
-        fails = procd = 0
+        rtt = check_rtt
 
-        _, 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, "quiet", @done)
-            conn.expire(key, 60)
-            conn.rpop("#{key}-signals")
+        fails = procd = 0
+        kb = memory_usage(::Process.pid)
+
+        _, exists, _, _, msg = redis { |conn|
+          conn.multi { |transaction|
+            transaction.sadd("processes", [key])
+            transaction.exists?(key)
+            transaction.hmset(key, "info", to_json,
+              "busy", curstate.size,
+              "beat", Time.now.to_f,
+              "rtt_us", rtt,
+              "quiet", @done.to_s,
+              "rss", kb)
+            transaction.expire(key, 60)
+            transaction.rpop("#{key}-signals")
           }
         }
 
         # first heartbeat or recovering from an outage and need to reestablish our heartbeat
         fire_event(:heartbeat) unless exists
+        fire_event(:beat, oneshot: false)
 
         return unless msg
 
@@ -180,27 +194,74 @@ module Sidekiq
       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
+      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 }
+        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-saturated; reduce your concurrency and/or 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" => @config[:tag] || "",
+        "concurrency" => @config[:concurrency],
+        "queues" => @config[:queues].uniq,
+        "labels" => @config[: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

@@ -6,15 +6,20 @@ require "time"
 module Sidekiq
   module Context
     def self.with(hash)
+      orig_context = current.dup
       current.merge!(hash)
       yield
     ensure
-      hash.each_key { |key| current.delete(key) }
+      Thread.current[:sidekiq_context] = orig_context
     end
 
     def self.current
       Thread.current[:sidekiq_context] ||= {}
     end
+
+    def self.add(k, v)
+      current[k] = v
+    end
   end
 
   module LoggingUtils
@@ -30,24 +35,10 @@ module Sidekiq
       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
+    LEVELS.each do |level, numeric_level|
+      define_method("#{level}?") do
+        local_level.nil? ? super() : local_level <= numeric_level
+      end
     end
 
     def local_level
@@ -89,7 +80,7 @@ module Sidekiq
       return true if @logdev.nil? || severity < level
 
       if message.nil?
-        if block_given?
+        if block
           message = yield
         else
           message = progname

data/lib/sidekiq/manager.rb

@@ -1,6 +1,5 @@
 # frozen_string_literal: true
 
-require "sidekiq/util"
 require "sidekiq/processor"
 require "sidekiq/fetch"
 require "set"
@@ -21,43 +20,37 @@ module Sidekiq
   # the shutdown process.  The other tasks are performed by other threads.
   #
   class Manager
-    include Util
+    include Sidekiq::Component
 
     attr_reader :workers
-    attr_reader :options
 
     def initialize(options = {})
+      @config = options
       logger.debug { options.inspect }
-      @options = options
       @count = options[:concurrency] || 10
       raise ArgumentError, "Concurrency of #{@count} is not supported" if @count < 1
 
       @done = false
       @workers = Set.new
       @count.times do
-        @workers << Processor.new(self, options)
+        @workers << Processor.new(@config, &method(:processor_result))
       end
       @plock = Mutex.new
     end
 
     def start
-      @workers.each do |x|
-        x.start
-      end
+      @workers.each(&:start)
     end
 
     def quiet
       return if @done
       @done = true
 
-      logger.info { "Terminating quiet workers" }
-      @workers.each { |x| x.terminate }
+      logger.info { "Terminating quiet threads" }
+      @workers.each(&:terminate)
       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)
@@ -68,29 +61,18 @@ module Sidekiq
       sleep PAUSE_TIME
       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
+      logger.info { "Pausing to allow jobs to finish..." }
+      wait_for(deadline) { @workers.empty? }
       return if @workers.empty?
 
       hard_shutdown
     end
 
-    def processor_stopped(processor)
-      @plock.synchronize do
-        @workers.delete(processor)
-      end
-    end
-
-    def processor_died(processor, reason)
+    def processor_result(processor, reason = nil)
       @plock.synchronize do
         @workers.delete(processor)
         unless @done
-          p = Processor.new(self, options)
+          p = Processor.new(@config, &method(:processor_result))
           @workers << p
           p.start
         end
@@ -104,7 +86,7 @@ module Sidekiq
     private
 
     def hard_shutdown
-      # We've reached the timeout and we still have busy workers.
+      # We've reached the timeout and we still have busy threads.
       # They must die but their jobs shall live on.
       cleanup = nil
       @plock.synchronize do
@@ -114,22 +96,41 @@ module Sidekiq
       if cleanup.size > 0
         jobs = cleanup.map { |p| p.job }.compact
 
-        logger.warn { "Terminating #{cleanup.size} busy worker threads" }
-        logger.warn { "Work still in progress #{jobs.inspect}" }
+        logger.warn { "Terminating #{cleanup.size} busy threads" }
+        logger.debug { "Jobs still in progress #{jobs.inspect}" }
 
         # Re-enqueue unfinished jobs
         # NOTE: You may notice that we may push a job back to redis before
-        # the worker thread is terminated. This is ok because Sidekiq's
+        # the thread is terminated. This is ok because Sidekiq's
         # 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]
-        strategy.bulk_requeue(jobs, @options)
+        strategy = @config[:fetch]
+        strategy.bulk_requeue(jobs, @config)
       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
+
+    # hack for quicker development / testing environment #2774
+    PAUSE_TIME = $stdout.tty? ? 0.1 : 0.5
+
+    # Wait for the orblock to be true or the deadline passed.
+    def wait_for(deadline, &condblock)
+      remaining = deadline - ::Process.clock_gettime(::Process::CLOCK_MONOTONIC)
+      while remaining > PAUSE_TIME
+        return if condblock.call
+        sleep PAUSE_TIME
+        remaining = deadline - ::Process.clock_gettime(::Process::CLOCK_MONOTONIC)
+      end
     end
   end
 end

data/lib/sidekiq/metrics/deploy.rb

@@ -0,0 +1,47 @@
+require "sidekiq"
+require "time"
+
+# This file is designed to be required within the user's
+# deployment script; it should need a bare minimum of dependencies.
+#
+#   require "sidekiq/metrics/deploy"
+#   gitdesc = `git log -1 --format="%h %s"`.strip
+#   d = Sidekiq::Metrics::Deploy.new
+#   d.mark(label: gitdesc)
+#
+# Note that you cannot mark more than once per minute. This is a feature, not a bug.
+module Sidekiq
+  module Metrics
+    class Deploy
+      MARK_TTL = 90 * 24 * 60 * 60 # 90 days
+
+      def initialize(pool = Sidekiq.redis_pool)
+        @pool = pool
+      end
+
+      def mark(at: Time.now, label: "")
+        # we need to round the timestamp so that we gracefully
+        # handle an excepted common error in marking deploys:
+        # having every process mark its deploy, leading
+        # to N marks for each deploy. Instead we round the time
+        # to the minute so that multple marks within that minute
+        # will all naturally rollup into one mark per minute.
+        whence = at.utc
+        floor = Time.utc(whence.year, whence.month, whence.mday, whence.hour, whence.min, 0)
+        datecode = floor.strftime("%Y%m%d")
+        key = "#{datecode}-marks"
+        @pool.with do |c|
+          c.pipelined do |pipe|
+            pipe.hsetnx(key, floor.iso8601, label)
+            pipe.expire(key, MARK_TTL)
+          end
+        end
+      end
+
+      def fetch(date = Time.now.utc.to_date)
+        datecode = date.strftime("%Y%m%d")
+        @pool.with { |c| c.hgetall("#{datecode}-marks") }
+      end
+    end
+  end
+end