sidekiq 6.4.1 → 6.5.6

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

data/lib/sidekiq/metrics/query.rb

@@ -0,0 +1,153 @@
+require "sidekiq"
+require "date"
+require "set"
+
+require "sidekiq/metrics/shared"
+
+module Sidekiq
+  module Metrics
+    # Allows caller to query for Sidekiq execution metrics within Redis.
+    # Caller sets a set of attributes to act as filters. {#fetch} will call
+    # Redis and return a Hash of results.
+    #
+    # NB: all metrics and times/dates are UTC only. We specifically do not
+    # support timezones.
+    class Query
+      def initialize(pool: Sidekiq.redis_pool, now: Time.now)
+        @time = now.utc
+        @pool = pool
+        @klass = nil
+      end
+
+      # Get metric data for all jobs from the last hour
+      def top_jobs(minutes: 60)
+        result = Result.new
+
+        time = @time
+        redis_results = @pool.with do |conn|
+          conn.pipelined do |pipe|
+            minutes.times do |idx|
+              key = "j|#{time.strftime("%Y%m%d")}|#{time.hour}:#{time.min}"
+              pipe.hgetall key
+              result.prepend_bucket time
+              time -= 60
+            end
+          end
+        end
+
+        time = @time
+        redis_results.each do |hash|
+          hash.each do |k, v|
+            kls, metric = k.split("|")
+            result.job_results[kls].add_metric metric, time, v.to_i
+          end
+          time -= 60
+        end
+
+        result.marks = fetch_marks(result.starts_at..result.ends_at)
+
+        result
+      end
+
+      def for_job(klass, minutes: 60)
+        result = Result.new
+
+        time = @time
+        redis_results = @pool.with do |conn|
+          conn.pipelined do |pipe|
+            minutes.times do |idx|
+              key = "j|#{time.strftime("%Y%m%d")}|#{time.hour}:#{time.min}"
+              pipe.hmget key, "#{klass}|ms", "#{klass}|p", "#{klass}|f"
+              result.prepend_bucket time
+              time -= 60
+            end
+          end
+        end
+
+        time = @time
+        @pool.with do |conn|
+          redis_results.each do |(ms, p, f)|
+            result.job_results[klass].add_metric "ms", time, ms.to_i if ms
+            result.job_results[klass].add_metric "p", time, p.to_i if p
+            result.job_results[klass].add_metric "f", time, f.to_i if f
+            result.job_results[klass].add_hist time, Histogram.new(klass).fetch(conn, time)
+            time -= 60
+          end
+        end
+
+        result.marks = fetch_marks(result.starts_at..result.ends_at)
+
+        result
+      end
+
+      class Result < Struct.new(:starts_at, :ends_at, :size, :buckets, :job_results, :marks)
+        def initialize
+          super
+          self.buckets = []
+          self.marks = []
+          self.job_results = Hash.new { |h, k| h[k] = JobResult.new }
+        end
+
+        def prepend_bucket(time)
+          buckets.unshift time.strftime("%H:%M")
+          self.ends_at ||= time
+          self.starts_at = time
+        end
+      end
+
+      class JobResult < Struct.new(:series, :hist, :totals)
+        def initialize
+          super
+          self.series = Hash.new { |h, k| h[k] = Hash.new(0) }
+          self.hist = Hash.new { |h, k| h[k] = [] }
+          self.totals = Hash.new(0)
+        end
+
+        def add_metric(metric, time, value)
+          totals[metric] += value
+          series[metric][time.strftime("%H:%M")] += value
+
+          # Include timing measurements in seconds for convenience
+          add_metric("s", time, value / 1000.0) if metric == "ms"
+        end
+
+        def add_hist(time, hist_result)
+          hist[time.strftime("%H:%M")] = hist_result
+        end
+
+        def total_avg(metric = "ms")
+          completed = totals["p"] - totals["f"]
+          totals[metric].to_f / completed
+        end
+
+        def series_avg(metric = "ms")
+          series[metric].each_with_object(Hash.new(0)) do |(bucket, value), result|
+            completed = series.dig("p", bucket) - series.dig("f", bucket)
+            result[bucket] = completed == 0 ? 0 : value.to_f / completed
+          end
+        end
+      end
+
+      class MarkResult < Struct.new(:time, :label)
+        def bucket
+          time.strftime("%H:%M")
+        end
+      end
+
+      private
+
+      def fetch_marks(time_range)
+        [].tap do |result|
+          marks = @pool.with { |c| c.hgetall("#{@time.strftime("%Y%m%d")}-marks") }
+
+          marks.each do |timestamp, label|
+            time = Time.parse(timestamp)
+            if time_range.cover? time
+              result << MarkResult.new(time, label)
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/lib/sidekiq/metrics/shared.rb

@@ -0,0 +1,94 @@
+require "concurrent"
+
+module Sidekiq
+  module Metrics
+    # TODO Support apps without concurrent-ruby
+    Counter = ::Concurrent::AtomicFixnum
+
+    # Implements space-efficient but statistically useful histogram storage.
+    # A precise time histogram stores every time. Instead we break times into a set of
+    # known buckets and increment counts of the associated time bucket. Even if we call
+    # the histogram a million times, we'll still only store 26 buckets.
+    # NB: needs to be thread-safe or resiliant to races.
+    #
+    # To store this data, we use Redis' BITFIELD command to store unsigned 16-bit counters
+    # per bucket per klass per minute. It's unlikely that most people will be executing more
+    # than 1000 job/sec for a full minute of a specific type.
+    class Histogram
+      include Enumerable
+
+      # This number represents the maximum milliseconds for this bucket.
+      # 20 means all job executions up to 20ms, e.g. if a job takes
+      # 280ms, it'll increment bucket[7]. Note we can track job executions
+      # up to about 5.5 minutes. After that, it's assumed you're probably
+      # not too concerned with its performance.
+      BUCKET_INTERVALS = [
+        20, 30, 45, 65, 100,
+        150, 225, 335, 500, 750,
+        1100, 1700, 2500, 3800, 5750,
+        8500, 13000, 20000, 30000, 45000,
+        65000, 100000, 150000, 225000, 335000,
+        Float::INFINITY # the "maybe your job is too long" bucket
+      ]
+      LABELS = [
+        "20ms", "30ms", "45ms", "65ms", "100ms",
+        "150ms", "225ms", "335ms", "500ms", "750ms",
+        "1.1s", "1.7s", "2.5s", "3.8s", "5.75s",
+        "8.5s", "13s", "20s", "30s", "45s",
+        "65s", "100s", "150s", "225s", "335s",
+        "Slow"
+      ]
+
+      FETCH = "GET u16 #0 GET u16 #1 GET u16 #2 GET u16 #3 \
+        GET u16 #4 GET u16 #5 GET u16 #6 GET u16 #7 \
+        GET u16 #8 GET u16 #9 GET u16 #10 GET u16 #11 \
+        GET u16 #12 GET u16 #13 GET u16 #14 GET u16 #15 \
+        GET u16 #16 GET u16 #17 GET u16 #18 GET u16 #19 \
+        GET u16 #20 GET u16 #21 GET u16 #22 GET u16 #23 \
+        GET u16 #24 GET u16 #25".split
+
+      def each
+        buckets.each { |counter| yield counter.value }
+      end
+
+      def label(idx)
+        LABELS[idx]
+      end
+
+      attr_reader :buckets
+      def initialize(klass)
+        @klass = klass
+        @buckets = Array.new(BUCKET_INTERVALS.size) { Counter.new }
+      end
+
+      def record_time(ms)
+        index_to_use = BUCKET_INTERVALS.each_index do |idx|
+          break idx if ms < BUCKET_INTERVALS[idx]
+        end
+
+        @buckets[index_to_use].increment
+      end
+
+      def fetch(conn, now = Time.now)
+        window = now.utc.strftime("%d-%H:%-M")
+        key = "#{@klass}-#{window}"
+        conn.bitfield(key, *FETCH)
+      end
+
+      def persist(conn, now = Time.now)
+        buckets, @buckets = @buckets, []
+        window = now.utc.strftime("%d-%H:%-M")
+        key = "#{@klass}-#{window}"
+        cmd = [key, "OVERFLOW", "SAT"]
+        buckets.each_with_index do |counter, idx|
+          val = counter.value
+          cmd << "INCRBY" << "u16" << "##{idx}" << val.to_s if val > 0
+        end
+
+        conn.bitfield(*cmd) if cmd.size > 3
+        conn.expire(key, 86400)
+        key
+      end
+    end
+  end
+end

data/lib/sidekiq/metrics/tracking.rb

@@ -0,0 +1,134 @@
+require "time"
+require "sidekiq"
+require "sidekiq/metrics/shared"
+
+# This file contains the components which track execution metrics within Sidekiq.
+module Sidekiq
+  module Metrics
+    class ExecutionTracker
+      include Sidekiq::Component
+
+      def initialize(config)
+        @config = config
+        @jobs = Hash.new(0)
+        @totals = Hash.new(0)
+        @grams = Hash.new { |hash, key| hash[key] = Histogram.new(key) }
+        @lock = Mutex.new
+      end
+
+      def track(queue, klass)
+        start = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC, :millisecond)
+        time_ms = 0
+        begin
+          begin
+            yield
+          ensure
+            finish = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC, :millisecond)
+            time_ms = finish - start
+          end
+          # We don't track time for failed jobs as they can have very unpredictable
+          # execution times. more important to know average time for successful jobs so we
+          # can better recognize when a perf regression is introduced.
+          @lock.synchronize {
+            @grams[klass].record_time(time_ms)
+            @jobs["#{klass}|ms"] += time_ms
+            @totals["ms"] += time_ms
+          }
+        rescue Exception
+          @lock.synchronize {
+            @jobs["#{klass}|f"] += 1
+            @totals["f"] += 1
+          }
+          raise
+        ensure
+          @lock.synchronize {
+            @jobs["#{klass}|p"] += 1
+            @totals["p"] += 1
+          }
+        end
+      end
+
+      LONG_TERM = 90 * 24 * 60 * 60
+      MID_TERM = 7 * 24 * 60 * 60
+      SHORT_TERM = 8 * 60 * 60
+
+      def flush(time = Time.now)
+        totals, jobs, grams = reset
+        procd = totals["p"]
+        fails = totals["f"]
+        return if procd == 0 && fails == 0
+
+        now = time.utc
+        nowdate = now.strftime("%Y%m%d")
+        nowhour = now.strftime("%Y%m%d|%-H")
+        nowmin = now.strftime("%Y%m%d|%-H:%-M")
+        count = 0
+
+        redis do |conn|
+          if grams.size > 0
+            conn.pipelined do |pipe|
+              grams.each do |_, gram|
+                gram.persist(pipe, now)
+              end
+            end
+          end
+
+          [
+            ["j", jobs, nowdate, LONG_TERM],
+            ["j", jobs, nowhour, MID_TERM],
+            ["j", jobs, nowmin, SHORT_TERM]
+          ].each do |prefix, data, bucket, ttl|
+            # Quietly seed the new 7.0 stats format so migration is painless.
+            conn.pipelined do |xa|
+              stats = "#{prefix}|#{bucket}"
+              # logger.debug "Flushing metrics #{stats}"
+              data.each_pair do |key, value|
+                xa.hincrby stats, key, value
+                count += 1
+              end
+              xa.expire(stats, ttl)
+            end
+          end
+          logger.info "Flushed #{count} metrics"
+          count
+        end
+      end
+
+      private
+
+      def reset
+        @lock.synchronize {
+          array = [@totals, @jobs, @grams]
+          @totals = Hash.new(0)
+          @jobs = Hash.new(0)
+          @grams = Hash.new { |hash, key| hash[key] = Histogram.new(key) }
+          array
+        }
+      end
+    end
+
+    class Middleware
+      include Sidekiq::ServerMiddleware
+
+      def initialize(options)
+        @exec = options
+      end
+
+      def call(_instance, hash, queue, &block)
+        @exec.track(queue, hash["wrapped"] || hash["class"], &block)
+      end
+    end
+  end
+end
+
+if ENV["SIDEKIQ_METRICS_BETA"] == "1"
+  Sidekiq.configure_server do |config|
+    exec = Sidekiq::Metrics::ExecutionTracker.new(config)
+    config.server_middleware do |chain|
+      chain.add Sidekiq::Metrics::Middleware, exec
+    end
+    config.on(:beat) do
+      exec.flush
+    end
+  end
+end

data/lib/sidekiq/middleware/chain.rb

@@ -1,82 +1,102 @@
 # frozen_string_literal: true
 
+require "sidekiq/middleware/modules"
+
 module Sidekiq
   # Middleware is code configured to run before/after
-  # a message is processed.  It is patterned after Rack
+  # a job is processed.  It is patterned after Rack
   # middleware. Middleware exists for the client side
   # (pushing jobs onto the queue) as well as the server
   # side (when jobs are actually processed).
   #
+  # Callers will register middleware Classes and Sidekiq will
+  # create new instances of the middleware for every job. This
+  # is important so that instance state is not shared accidentally
+  # between job executions.
+  #
   # To add middleware for the client:
   #
-  # Sidekiq.configure_client do |config|
-  #   config.client_middleware do |chain|
-  #     chain.add MyClientHook
+  #   Sidekiq.configure_client do |config|
+  #     config.client_middleware do |chain|
+  #       chain.add MyClientHook
+  #     end
   #   end
-  # end
   #
   # To modify middleware for the server, just call
   # with another block:
   #
-  # Sidekiq.configure_server do |config|
-  #   config.server_middleware do |chain|
-  #     chain.add MyServerHook
-  #     chain.remove ActiveRecord
+  #   Sidekiq.configure_server do |config|
+  #     config.server_middleware do |chain|
+  #       chain.add MyServerHook
+  #       chain.remove ActiveRecord
+  #     end
   #   end
-  # end
   #
   # To insert immediately preceding another entry:
   #
-  # Sidekiq.configure_client do |config|
-  #   config.client_middleware do |chain|
-  #     chain.insert_before ActiveRecord, MyClientHook
+  #   Sidekiq.configure_client do |config|
+  #     config.client_middleware do |chain|
+  #       chain.insert_before ActiveRecord, MyClientHook
+  #     end
   #   end
-  # end
   #
   # To insert immediately after another entry:
   #
-  # Sidekiq.configure_client do |config|
-  #   config.client_middleware do |chain|
-  #     chain.insert_after ActiveRecord, MyClientHook
+  #   Sidekiq.configure_client do |config|
+  #     config.client_middleware do |chain|
+  #       chain.insert_after ActiveRecord, MyClientHook
+  #     end
   #   end
-  # end
   #
   # This is an example of a minimal server middleware:
   #
-  # class MyServerHook
-  #   def call(worker_instance, msg, queue)
-  #     puts "Before work"
-  #     yield
-  #     puts "After work"
+  #   class MyServerHook
+  #     include Sidekiq::ServerMiddleware
+  #
+  #     def call(job_instance, msg, queue)
+  #       logger.info "Before job"
+  #       redis {|conn| conn.get("foo") } # do something in Redis
+  #       yield
+  #       logger.info "After job"
+  #     end
   #   end
-  # end
   #
   # This is an example of a minimal client middleware, note
   # the method must return the result or the job will not push
   # to Redis:
   #
-  # class MyClientHook
-  #   def call(worker_class, msg, queue, redis_pool)
-  #     puts "Before push"
-  #     result = yield
-  #     puts "After push"
-  #     result
+  #   class MyClientHook
+  #     include Sidekiq::ClientMiddleware
+  #
+  #     def call(job_class, msg, queue, redis_pool)
+  #       logger.info "Before push"
+  #       result = yield
+  #       logger.info "After push"
+  #       result
+  #     end
   #   end
-  # end
   #
   module Middleware
     class Chain
       include Enumerable
 
+      # A unique instance of the middleware chain is created for
+      # each job executed in order to be thread-safe.
+      # @param copy [Sidekiq::Middleware::Chain] New instance of Chain
+      # @returns nil
       def initialize_copy(copy)
         copy.instance_variable_set(:@entries, entries.dup)
+        nil
       end
 
+      # Iterate through each middleware in the chain
       def each(&block)
         entries.each(&block)
       end
 
-      def initialize
+      # @api private
+      def initialize(config = nil) # :nodoc:
+        @config = config
         @entries = nil
         yield self if block_given?
       end
@@ -85,38 +105,55 @@ module Sidekiq
         @entries ||= []
       end
 
+      # Remove all middleware matching the given Class
+      # @param klass [Class]
       def remove(klass)
         entries.delete_if { |entry| entry.klass == klass }
       end
 
+      # Add the given middleware to the end of the chain.
+      # Sidekiq will call `klass.new(*args)` to create a clean
+      # copy of your middleware for every job executed.
+      #
+      #   chain.add(Statsd::Metrics, { collector: "localhost:8125" })
+      #
+      # @param klass [Class] Your middleware class
+      # @param *args [Array<Object>] Set of arguments to pass to every instance of your middleware
       def add(klass, *args)
         remove(klass)
-        entries << Entry.new(klass, *args)
+        entries << Entry.new(@config, klass, *args)
       end
 
+      # Identical to {#add} except the middleware is added to the front of the chain.
       def prepend(klass, *args)
         remove(klass)
-        entries.insert(0, Entry.new(klass, *args))
+        entries.insert(0, Entry.new(@config, klass, *args))
       end
 
+      # Inserts +newklass+ before +oldklass+ in the chain.
+      # Useful if one middleware must run before another middleware.
       def insert_before(oldklass, newklass, *args)
         i = entries.index { |entry| entry.klass == newklass }
-        new_entry = i.nil? ? Entry.new(newklass, *args) : entries.delete_at(i)
+        new_entry = i.nil? ? Entry.new(@config, newklass, *args) : entries.delete_at(i)
         i = entries.index { |entry| entry.klass == oldklass } || 0
         entries.insert(i, new_entry)
       end
 
+      # Inserts +newklass+ after +oldklass+ in the chain.
+      # Useful if one middleware must run after another middleware.
       def insert_after(oldklass, newklass, *args)
         i = entries.index { |entry| entry.klass == newklass }
-        new_entry = i.nil? ? Entry.new(newklass, *args) : entries.delete_at(i)
+        new_entry = i.nil? ? Entry.new(@config, newklass, *args) : entries.delete_at(i)
         i = entries.index { |entry| entry.klass == oldklass } || entries.count - 1
         entries.insert(i + 1, new_entry)
       end
 
+      # @return [Boolean] if the given class is already in the chain
       def exists?(klass)
         any? { |entry| entry.klass == klass }
       end
 
+      # @return [Boolean] if the chain contains no middleware
       def empty?
         @entries.nil? || @entries.empty?
       end
@@ -129,6 +166,8 @@ module Sidekiq
         entries.clear
       end
 
+      # Used by Sidekiq to execute the middleware at runtime
+      # @api private
       def invoke(*args)
         return yield if empty?
 
@@ -146,16 +185,21 @@ module Sidekiq
 
     private
 
+    # Represents each link in the middleware chain
+    # @api private
     class Entry
       attr_reader :klass
 
-      def initialize(klass, *args)
+      def initialize(config, klass, *args)
+        @config = config
         @klass = klass
         @args = args
       end
 
       def make_new
-        @klass.new(*@args)
+        x = @klass.new(*@args)
+        x.config = @config if @config && x.respond_to?(:config=)
+        x
       end
     end
   end

data/lib/sidekiq/middleware/current_attributes.rb

@@ -15,22 +15,28 @@ module Sidekiq
   #
   module CurrentAttributes
     class Save
+      include Sidekiq::ClientMiddleware
+
       def initialize(cattr)
         @klass = cattr
       end
 
       def call(_, job, _, _)
         attrs = @klass.attributes
-        if job.has_key?("cattr")
-          job["cattr"].merge!(attrs)
-        else
-          job["cattr"] = attrs
+        if attrs.any?
+          if job.has_key?("cattr")
+            job["cattr"].merge!(attrs)
+          else
+            job["cattr"] = attrs
+          end
         end
         yield
       end
     end
 
     class Load
+      include Sidekiq::ServerMiddleware
+
       def initialize(cattr)
         @klass = cattr
       end