sidekiq 6.2.2 → 8.1.5

data/lib/sidekiq/redis_connection.rb

@@ -1,121 +1,92 @@
 # frozen_string_literal: true
 
 require "connection_pool"
-require "redis"
 require "uri"
+require "sidekiq/redis_client_adapter"
 
 module Sidekiq
-  class RedisConnection
+  module RedisConnection
     class << self
       def create(options = {})
-        symbolized_options = options.transform_keys(&:to_sym)
+        symbolized_options = deep_symbolize_keys(options)
+        symbolized_options[:url] ||= determine_redis_provider
+        symbolized_options[:password] = wrap(symbolized_options[:password]) if symbolized_options.key?(:password)
+        symbolized_options[:sentinel_password] = wrap(symbolized_options[:sentinel_password]) if symbolized_options.key?(:sentinel_password)
 
-        if !symbolized_options[:url] && (u = determine_redis_provider)
-          symbolized_options[:url] = u
-        end
+        logger = symbolized_options.delete(:logger)
+        logger&.info { "Sidekiq #{Sidekiq::VERSION} connecting to Redis with options #{scrub(symbolized_options)}" }
 
-        size = if symbolized_options[:size]
-          symbolized_options[:size]
-        elsif Sidekiq.server?
-          # Give ourselves plenty of connections.  pool is lazy
-          # so we won't create them until we need them.
-          Sidekiq.options[:concurrency] + 5
-        elsif ENV["RAILS_MAX_THREADS"]
-          Integer(ENV["RAILS_MAX_THREADS"])
-        else
-          5
-        end
+        raise "Sidekiq 7+ does not support Redis protocol 2" if symbolized_options[:protocol] == 2
 
-        verify_sizing(size, Sidekiq.options[:concurrency]) if Sidekiq.server?
+        safe = !!symbolized_options.delete(:cluster_safe)
+        raise ":nodes not allowed, Sidekiq is not safe to run on Redis Cluster" if !safe && symbolized_options.key?(:nodes)
 
-        pool_timeout = symbolized_options[:pool_timeout] || 1
-        log_info(symbolized_options)
+        size = symbolized_options.delete(:size) || 5
+        pool_timeout = symbolized_options.delete(:pool_timeout) || 1
+        symbolized_options.delete(:pool_name)
 
+        # Default timeout in redis-client is 1 second, which can be too aggressive
+        # if the Sidekiq process is CPU-bound. With 10-15 threads and a thread quantum of 100ms,
+        # it can be easy to get the occasional ReadTimeoutError. You can still provide
+        # a smaller timeout explicitly:
+        #     config.redis = { url: "...", timeout: 1 }
+        symbolized_options[:timeout] ||= 3
+
+        redis_config = Sidekiq::RedisClientAdapter.new(symbolized_options)
         ConnectionPool.new(timeout: pool_timeout, size: size) do
-          build_client(symbolized_options)
+          redis_config.new_client
         end
       end
 
       private
 
-      # Sidekiq needs a lot of concurrent Redis connections.
-      #
-      # We need a connection for each Processor.
-      # We need a connection for Pro's real-time change listener
-      # We need a connection to various features to call Redis every few seconds:
-      #   - the process heartbeat.
-      #   - enterprise's leader election
-      #   - enterprise's cron support
-      def verify_sizing(size, concurrency)
-        raise ArgumentError, "Your Redis connection pool is too small for Sidekiq to work. Your pool has #{size} connections but must have at least #{concurrency + 2}" if size < (concurrency + 2)
-      end
-
-      def build_client(options)
-        namespace = options[:namespace]
-
-        client = Redis.new client_opts(options)
-        if namespace
-          begin
-            require "redis/namespace"
-            Redis::Namespace.new(namespace, redis: client)
-          rescue LoadError
-            Sidekiq.logger.error("Your Redis configuration uses the namespace '#{namespace}' but the redis-namespace gem is not included in the Gemfile." \
-                                 "Add the gem to your Gemfile to continue using a namespace. Otherwise, remove the namespace parameter.")
-            exit(-127)
-          end
+      # Wrap hard-coded passwords in a Proc to avoid logging the value
+      def wrap(pwd)
+        if pwd.is_a?(String)
+          ->(username) { pwd }
         else
-          client
+          pwd
         end
       end
 
-      def client_opts(options)
-        opts = options.dup
-        if opts[:namespace]
-          opts.delete(:namespace)
-        end
-
-        if opts[:network_timeout]
-          opts[:timeout] = opts[:network_timeout]
-          opts.delete(:network_timeout)
+      def deep_symbolize_keys(object)
+        case object
+        when Hash
+          object.each_with_object({}) do |(key, value), result|
+            result[key.to_sym] = deep_symbolize_keys(value)
+          end
+        when Array
+          object.map { |e| deep_symbolize_keys(e) }
+        else
+          object
         end
-
-        opts[:driver] ||= Redis::Connection.drivers.last || "ruby"
-
-        # Issue #3303, redis-rb will silently retry an operation.
-        # This can lead to duplicate jobs if Sidekiq::Client's LPUSH
-        # is performed twice but I believe this is much, much rarer
-        # than the reconnect silently fixing a problem; we keep it
-        # on by default.
-        opts[:reconnect_attempts] ||= 1
-
-        opts
       end
 
-      def log_info(options)
+      def scrub(options)
         redacted = "REDACTED"
 
-        # deep clone so we can muck with these options all we want
-        #
-        # exclude SSL params from dump-and-load because some information isn't
-        # safely dumpable in current Rubies
-        keys = options.keys
-        keys.delete(:ssl_params)
+        # Deep clone so we can muck with these options all we want and exclude
+        # params from dump-and-load that may contain objects that Marshal is
+        # unable to safely dump.
+        keys = options.keys - [:logger, :ssl_params, :password, :sentinel_password]
         scrubbed_options = Marshal.load(Marshal.dump(options.slice(*keys)))
         if scrubbed_options[:url] && (uri = URI.parse(scrubbed_options[:url])) && uri.password
           uri.password = redacted
           scrubbed_options[:url] = uri.to_s
         end
-        if scrubbed_options[:password]
-          scrubbed_options[:password] = redacted
-        end
+        scrubbed_options[:password] = redacted if options.key?(:password)
+        scrubbed_options[:sentinel_password] = redacted if options.key?(:sentinel_password)
         scrubbed_options[:sentinels]&.each do |sentinel|
-          sentinel[:password] = redacted if sentinel[:password]
-        end
-        if Sidekiq.server?
-          Sidekiq.logger.info("Booting Sidekiq #{Sidekiq::VERSION} with redis options #{scrubbed_options}")
-        else
-          Sidekiq.logger.debug("#{Sidekiq::NAME} client with redis options #{scrubbed_options}")
+          if sentinel.is_a?(String)
+            if (uri = URI(sentinel)) && uri.password
+              uri.password = redacted
+              sentinel.replace(uri.to_s)
+            end
+          elsif sentinel[:password]
+            sentinel[:password] = redacted
+          end
         end
+        scrubbed_options
       end
 
       def determine_redis_provider
@@ -137,9 +108,7 @@ module Sidekiq
           EOM
         end
 
-        ENV[
-          p || "REDIS_URL"
-        ]
+        ENV[p.to_s] || ENV["REDIS_URL"]
       end
     end
   end

data/lib/sidekiq/ring_buffer.rb

@@ -0,0 +1,32 @@
+# frozen_string_literal: true
+
+require "forwardable"
+
+module Sidekiq
+  class RingBuffer
+    include Enumerable
+    extend Forwardable
+
+    def_delegators :@buf, :[], :each, :size
+
+    def initialize(size, default = 0)
+      @size = size
+      @buf = Array.new(size, default)
+      @index = 0
+    end
+
+    def <<(element)
+      @buf[@index % @size] = element
+      @index += 1
+      element
+    end
+
+    def buffer
+      @buf
+    end
+
+    def reset(default = 0)
+      @buf.fill(default)
+    end
+  end
+end

data/lib/sidekiq/scheduled.rb

@@ -1,37 +1,66 @@
 # frozen_string_literal: true
 
 require "sidekiq"
-require "sidekiq/util"
-require "sidekiq/api"
+require "sidekiq/component"
 
 module Sidekiq
   module Scheduled
     SETS = %w[retry schedule]
 
     class Enq
-      def enqueue_jobs(now = Time.now.to_f.to_s, sorted_sets = SETS)
+      include Sidekiq::Component
+
+      LUA_ZPOPBYSCORE = <<~LUA
+        local key, now = KEYS[1], ARGV[1]
+        local jobs = redis.call("zrange", key, "-inf", now, "byscore", "limit", 0, 1)
+        if jobs[1] then
+          redis.call("zrem", key, jobs[1])
+          return jobs[1]
+        end
+      LUA
+
+      def initialize(container)
+        @config = container
+        @client = Sidekiq::Client.new(config: container)
+        @done = false
+        @lua_zpopbyscore_sha = nil
+      end
+
+      def enqueue_jobs(sorted_sets = SETS)
         # A job's "score" in Redis is the time at which it should be processed.
         # Just check Redis for the set of jobs with a timestamp before now.
-        Sidekiq.redis do |conn|
+        redis do |conn|
           sorted_sets.each do |sorted_set|
-            # Get next items in the queue with scores (time to execute) <= now.
-            until (jobs = conn.zrangebyscore(sorted_set, "-inf", now, limit: [0, 100])).empty?
-              # We need to go through the list one at a time to reduce the risk of something
-              # going wrong between the time jobs are popped from the scheduled queue and when
-              # they are pushed onto a work queue and losing the jobs.
-              jobs.each do |job|
-                # Pop item off the queue and add it to the work queue. If the job can't be popped from
-                # the queue, it's because another process already popped it so we can move on to the
-                # next one.
-                if conn.zrem(sorted_set, job)
-                  Sidekiq::Client.push(Sidekiq.load_json(job))
-                  Sidekiq.logger.debug { "enqueued #{sorted_set}: #{job}" }
-                end
-              end
+            # Get next item in the queue with score (time to execute) <= now.
+            # We need to go through the list one at a time to reduce the risk of something
+            # going wrong between the time jobs are popped from the scheduled queue and when
+            # they are pushed onto a work queue and losing the jobs.
+            while !@done && (job = zpopbyscore(conn, keys: [sorted_set], argv: [Time.now.to_f.to_s]))
+              @client.push(Sidekiq.load_json(job))
+              logger.debug { "enqueued #{sorted_set}: #{job}" }
             end
           end
         end
       end
+
+      def terminate
+        @done = true
+      end
+
+      private
+
+      def zpopbyscore(conn, keys: nil, argv: nil)
+        if @lua_zpopbyscore_sha.nil?
+          @lua_zpopbyscore_sha = conn.script(:load, LUA_ZPOPBYSCORE)
+        end
+
+        conn.call("EVALSHA", @lua_zpopbyscore_sha, keys.size, *keys, *argv)
+      rescue RedisClient::CommandError => e
+        raise unless e.message.start_with?("NOSCRIPT")
+
+        @lua_zpopbyscore_sha = nil
+        retry
+      end
     end
 
     ##
@@ -40,27 +69,28 @@ module Sidekiq
     # just pops the job back onto its original queue so the
     # workers can pick it up like any other job.
     class Poller
-      include Util
+      include Sidekiq::Component
 
       INITIAL_WAIT = 10
+      attr_accessor :rnd
 
-      def initialize
-        @enq = (Sidekiq.options[:scheduled_enq] || Sidekiq::Scheduled::Enq).new
+      def initialize(config)
+        @config = config
+        @enq = (config[:scheduled_enq] || Sidekiq::Scheduled::Enq).new(config)
         @sleeper = ConnectionPool::TimedStack.new
         @done = false
         @thread = nil
         @count_calls = 0
+        @rnd = Random.new
       end
 
       # Shut down this instance, will pause until the thread is dead.
       def terminate
         @done = true
-        if @thread
-          t = @thread
-          @thread = nil
-          @sleeper << 0
-          t.value
-        end
+        @enq.terminate
+
+        @sleeper << 0
+        @thread&.value
       end
 
       def start
@@ -71,7 +101,7 @@ module Sidekiq
             enqueue
             wait
           end
-          Sidekiq.logger.info("Scheduler exiting...")
+          logger.info("Scheduler exiting...")
         }
       end
 
@@ -87,9 +117,7 @@ module Sidekiq
       private
 
       def wait
-        @sleeper.pop(random_poll_interval)
-      rescue Timeout::Error
-        # expected
+        @sleeper.pop(timeout: random_poll_interval, exception: false)
       rescue => ex
         # if poll_interval_average hasn't been calculated yet, we can
         # raise an error trying to reach Redis.
@@ -116,15 +144,18 @@ module Sidekiq
         # In the example above, each process should schedule every 10 seconds on average. We special
         # case smaller clusters to add 50% so they would sleep somewhere between 5 and 15 seconds.
         # As we run more processes, the scheduling interval average will approach an even spread
-        # between 0 and poll interval so we don't need this artifical boost.
+        # between 0 and poll interval so we don't need this artificial boost.
         #
-        if process_count < 10
+        count = process_count
+        interval = poll_interval_average(count)
+
+        if count < 10
           # For small clusters, calculate a random interval that is ±50% the desired average.
-          poll_interval_average * rand + poll_interval_average.to_f / 2
+          interval * @rnd.rand + interval.to_f / 2
         else
           # With 10+ processes, we should have enough randomness to get decent polling
           # across the entire timespan
-          poll_interval_average * rand
+          interval * @rnd.rand * 2
         end
       end
 
@@ -141,38 +172,63 @@ module Sidekiq
       # the same time: the thundering herd problem.
       #
       # We only do this if poll_interval_average is unset (the default).
-      def poll_interval_average
-        Sidekiq.options[:poll_interval_average] ||= scaled_poll_interval
+      def poll_interval_average(count)
+        @config[:poll_interval_average] || scaled_poll_interval(count)
       end
 
       # Calculates an average poll interval based on the number of known Sidekiq processes.
       # This minimizes a single point of failure by dispersing check-ins but without taxing
       # Redis if you run many Sidekiq processes.
-      def scaled_poll_interval
-        process_count * Sidekiq.options[:average_scheduled_poll_interval]
+      def scaled_poll_interval(process_count)
+        process_count * @config[:average_scheduled_poll_interval]
       end
 
       def process_count
-        # The work buried within Sidekiq::ProcessSet#cleanup can be
-        # expensive at scale. Cut it down by 90% with this counter.
-        # NB: This method is only called by the scheduler thread so we
-        # don't need to worry about the thread safety of +=.
-        pcount = Sidekiq::ProcessSet.new(@count_calls % 10 == 0).size
+        pcount = Sidekiq.redis { |conn| conn.scard("processes") }
         pcount = 1 if pcount == 0
-        @count_calls += 1
         pcount
       end
 
+      # A copy of Sidekiq::ProcessSet#cleanup because server
+      # should never depend on sidekiq/api.
+      def cleanup
+        # dont run cleanup more than once per minute
+        return 0 unless redis { |conn| conn.set("process_cleanup", "1", "NX", "EX", "60") }
+
+        count = 0
+        redis do |conn|
+          procs = conn.sscan("processes").to_a
+          heartbeats = conn.pipelined { |pipeline|
+            procs.each do |key|
+              pipeline.hget(key, "info")
+            end
+          }
+
+          # the hash named key has an expiry of 60 seconds.
+          # if it's not found, that means the process has not reported
+          # in to Redis and probably died.
+          to_prune = procs.select.with_index { |proc, i|
+            heartbeats[i].nil?
+          }
+          count = conn.srem("processes", to_prune) unless to_prune.empty?
+        end
+        count
+      end
+
       def initial_wait
-        # Have all processes sleep between 5-15 seconds.  10 seconds
-        # to give time for the heartbeat to register (if the poll interval is going to be calculated by the number
+        # Have all processes sleep between 5-15 seconds. 10 seconds to give time for
+        # the heartbeat to register (if the poll interval is going to be calculated by the number
         # of workers), and 5 random seconds to ensure they don't all hit Redis at the same time.
         total = 0
-        total += INITIAL_WAIT unless Sidekiq.options[:poll_interval_average]
+        total += INITIAL_WAIT unless @config[:poll_interval_average]
         total += (5 * rand)
 
-        @sleeper.pop(total)
-      rescue Timeout::Error
+        @sleeper.pop(timeout: total, exception: false)
+      ensure
+        # periodically clean out the `processes` set in Redis which can collect
+        # references to dead processes over time. The process count affects how
+        # often we scan for scheduled jobs.
+        cleanup
       end
     end
   end

data/lib/sidekiq/systemd.rb

@@ -1,3 +1,5 @@
+# frozen_string_literal: true
+
 #
 # Sidekiq's systemd integration allows Sidekiq to inform systemd:
 #  1. when it has successfully started