identity_cache 1.0.0 → 1.2.0

data/{railgun.yml → isogun.yml}

@@ -2,15 +2,10 @@
 name: identity-cache
 
 vm:
-  image:      /opt/dev/misc/railgun-images/default
   ip_address: 192.168.64.98
   memory:     1G
   cores:      2
 
-volumes:
-  root: 1G
-
 services:
   - mysql
-  - postgresql
   - memcached

data/lib/identity_cache/belongs_to_caching.rb

@@ -1,4 +1,5 @@
 # frozen_string_literal: true
+
 module IdentityCache
   module BelongsToCaching
     extend ActiveSupport::Concern

data/lib/identity_cache/cache_fetcher.rb

@@ -1,8 +1,53 @@
 # frozen_string_literal: true
+
+require "securerandom"
+
 module IdentityCache
   class CacheFetcher
     attr_accessor :cache_backend
 
+    EMPTY_HASH = {}.freeze
+
+    class FillLock
+      FILL_LOCKED = :fill_locked
+      FAILED_CLIENT_ID = "fill_failed"
+
+      class << self
+        def from_cache(marker, client_id, data_version)
+          raise ArgumentError unless marker == FILL_LOCKED
+
+          new(client_id: client_id, data_version: data_version)
+        end
+
+        def cache_value?(cache_value)
+          cache_value.is_a?(Array) && cache_value.length == 3 && cache_value.first == FILL_LOCKED
+        end
+      end
+
+      attr_reader :client_id, :data_version
+
+      def initialize(client_id:, data_version:)
+        @client_id = client_id
+        @data_version = data_version
+      end
+
+      def cache_value
+        [FILL_LOCKED, client_id, data_version]
+      end
+
+      def mark_failed
+        @client_id = FAILED_CLIENT_ID
+      end
+
+      def fill_failed?
+        @client_id == FAILED_CLIENT_ID
+      end
+
+      def ==(other)
+        self.class == other.class && client_id == other.client_id && data_version == other.data_version
+      end
+    end
+
     def initialize(cache_backend)
       @cache_backend = cache_backend
     end
@@ -25,27 +70,204 @@ module IdentityCache
       results
     end
 
-    def fetch(key)
-      result = nil
-      yielded = false
-      @cache_backend.cas(key) do |value|
-        yielded = true
-        unless IdentityCache::DELETED == value
-          result = value
-          break
+    def fetch(key, fill_lock_duration: nil, lock_wait_tries: 2, &block)
+      if fill_lock_duration && IdentityCache.should_fill_cache?
+        fetch_with_fill_lock(key, fill_lock_duration, lock_wait_tries, &block)
+      else
+        fetch_without_fill_lock(key, &block)
+      end
+    end
+
+    private
+
+    def fetch_without_fill_lock(key)
+      data = nil
+      upsert(key) do |value|
+        value = nil if value == IdentityCache::DELETED || FillLock.cache_value?(value)
+        unless value.nil?
+          return value
         end
-        result = yield
+
+        data = yield
         break unless IdentityCache.should_fill_cache?
-        result
+
+        data
+      end
+      data
+    end
+
+    def fetch_with_fill_lock(key, fill_lock_duration, lock_wait_tries, &block)
+      raise ArgumentError, "fill_lock_duration must be greater than 0.0" unless fill_lock_duration > 0.0
+      raise ArgumentError, "lock_wait_tries must be greater than 0" unless lock_wait_tries > 0
+
+      lock = nil
+      using_fallback_key = false
+      expiration_options = EMPTY_HASH
+      (lock_wait_tries + 2).times do # +2 is for first attempt and retry with fallback key
+        result = fetch_or_take_lock(key, old_lock: lock, **expiration_options)
+        case result
+        when FillLock
+          lock = result
+          if lock.client_id == client_id # have lock
+            data = begin
+              yield
+            rescue
+              mark_fill_failure_on_lock(key, expiration_options)
+              raise
+            end
+
+            if !fill_with_lock(key, data, lock, expiration_options) && !using_fallback_key
+              # fallback to storing data in the fallback key so it is available to clients waiting on the lock
+              expiration_options = fallback_key_expiration_options(fill_lock_duration)
+              @cache_backend.write(lock_fill_fallback_key(key, lock), data, expiration_options)
+            end
+            return data
+          else
+            raise LockWaitTimeout if lock_wait_tries <= 0
+
+            lock_wait_tries -= 1
+
+            # If fill failed in the other client, then it might be failing fast
+            # so avoid waiting the typical amount of time for a lock wait. The
+            # semian gem can be used to handle failing fast when the database is slow.
+            if lock.fill_failed?
+              return fetch_without_fill_lock(key, &block)
+            end
+
+            # lock wait
+            sleep(fill_lock_duration)
+            # loop around to retry fetch_or_take_lock
+          end
+        when IdentityCache::DELETED # interrupted by cache invalidation
+          if using_fallback_key
+            raise "unexpected cache invalidation of versioned fallback key"
+          elsif lock
+            # Cache invalidated during lock wait, use a versioned fallback key
+            # to avoid further cache invalidation interruptions.
+            using_fallback_key = true
+            key = lock_fill_fallback_key(key, lock)
+            expiration_options = fallback_key_expiration_options(fill_lock_duration)
+            # loop around to retry with fallback key
+          else
+            # Cache invalidation prevented lock from being taken or read, so we don't
+            # have a data version to use to build a shared fallback key. In the future
+            # we could add the data version to the cache invalidation value so a fallback
+            # key could be used here. For now, we assume that a cache invalidation occuring
+            # just after the cache wasn't filled is more likely a sign of a key that is
+            # written more than read (which this cache isn't a good fit for), rather than
+            # a thundering herd or reads.
+            return yield
+          end
+        when nil # Errors talking to memcached
+          return yield
+        else # hit
+          return result
+        end
+      end
+      raise "unexpected number of loop iterations"
+    end
+
+    def mark_fill_failure_on_lock(key, expiration_options)
+      @cache_backend.cas(key, expiration_options) do |value|
+        break unless FillLock.cache_value?(value)
+
+        lock = FillLock.from_cache(*value)
+        break if lock.client_id != client_id
+
+        lock.mark_failed
+        lock.cache_value
+      end
+    end
+
+    def upsert(key, expiration_options = EMPTY_HASH)
+      yielded = false
+      upserted = @cache_backend.cas(key, expiration_options) do |value|
+        yielded = true
+        yield value
       end
       unless yielded
-        result = yield
-        add(key, result)
+        data = yield nil
+        upserted = add(key, data, expiration_options)
       end
-      result
+      upserted
     end
 
-    private
+    def fetch_or_take_lock(key, old_lock:, **expiration_options)
+      new_lock = nil
+      upserted = upsert(key, expiration_options) do |value|
+        if value.nil? || value == IdentityCache::DELETED
+          if old_lock # cache invalidated
+            return value
+          else
+            new_lock = FillLock.new(client_id: client_id, data_version: SecureRandom.uuid)
+          end
+        elsif FillLock.cache_value?(value)
+          fetched_lock = FillLock.from_cache(*value)
+          if old_lock == fetched_lock
+            # preserve data version since there hasn't been any cache invalidations
+            new_lock = FillLock.new(client_id: client_id, data_version: old_lock.data_version)
+          elsif old_lock && fetched_lock.data_version != old_lock.data_version
+            # Cache was invalidated, then another lock was taken during a lock wait.
+            # Treat it as any other cache invalidation, where the caller will switch
+            # to the fallback key.
+            return IdentityCache::DELETED
+          else
+            return fetched_lock
+          end
+        else # hit
+          return value
+        end
+        new_lock.cache_value # take lock
+      end
+
+      return new_lock if upserted
+
+      value = @cache_backend.read(key)
+      if FillLock.cache_value?(value)
+        FillLock.from_cache(*value)
+      else
+        value
+      end
+    end
+
+    def fill_with_lock(key, data, my_lock, expiration_options)
+      upserted = upsert(key, expiration_options) do |value|
+        return false if value.nil? || value == IdentityCache::DELETED
+        return true unless FillLock.cache_value?(value) # already filled
+
+        current_lock = FillLock.from_cache(*value)
+        if current_lock.data_version != my_lock.data_version
+          return false # invalidated then relocked
+        end
+
+        data
+      end
+
+      upserted
+    end
+
+    def lock_fill_fallback_key(key, lock)
+      "lock_fill:#{lock.data_version}:#{key}"
+    end
+
+    def fallback_key_expiration_options(fill_lock_duration)
+      # Override the default TTL for the fallback key lock since it won't be used for very long.
+      expires_in = fill_lock_duration * 2
+
+      # memcached uses integer number of seconds for TTL so round up to avoid having
+      # the cache store round down with `to_i`
+      expires_in = expires_in.ceil
+
+      # memcached TTL only gets part of the first second (https://github.com/memcached/memcached/issues/307),
+      # so increase TTL by 1 to compensate
+      expires_in += 1
+
+      { expires_in: expires_in }
+    end
+
+    def client_id
+      @client_id ||= SecureRandom.uuid
+    end
 
     def cas_multi(keys)
       result = nil
@@ -70,6 +292,7 @@ module IdentityCache
 
         break if updates.empty?
         break unless IdentityCache.should_fill_cache?
+
         updates
       end
       result
@@ -81,8 +304,10 @@ module IdentityCache
       result.each { |k, v| add(k, v) }
     end
 
-    def add(key, value)
-      @cache_backend.write(key, value, unless_exist: true) if IdentityCache.should_fill_cache?
+    def add(key, value, expiration_options = EMPTY_HASH)
+      return false unless IdentityCache.should_fill_cache?
+
+      @cache_backend.write(key, value, { unless_exist: true, **expiration_options })
     end
   end
 end

data/lib/identity_cache/cache_hash.rb

@@ -1,9 +1,10 @@
 # frozen_string_literal: true
+
 # Use CityHash for fast hashing if it is available; use Digest::MD5 otherwise
 begin
-  require 'cityhash'
+  require "cityhash"
 rescue LoadError
-  unless RUBY_PLATFORM == 'java'
+  unless RUBY_PLATFORM == "java"
     warn(<<-NOTICE)
       ** Notice: CityHash was not loaded. **
 
@@ -15,20 +16,20 @@ rescue LoadError
     NOTICE
   end
 
-  require 'digest/md5'
+  require "digest/md5"
 end
 
 module IdentityCache
   module CacheHash
     if defined?(CityHash)
 
-      def memcache_hash(key) #:nodoc:
+      def memcache_hash(key) # :nodoc:
         CityHash.hash64(key)
       end
     else
 
-      def memcache_hash(key) #:nodoc:
-        a = Digest::MD5.digest(key).unpack('LL')
+      def memcache_hash(key) # :nodoc:
+        a = Digest::MD5.digest(key).unpack("LL")
         (a[0] << 32) | a[1]
       end
     end

data/lib/identity_cache/cache_invalidation.rb

@@ -1,4 +1,5 @@
 # frozen_string_literal: true
+
 module IdentityCache
   module CacheInvalidation
     CACHE_KEY_NAMES = [:ids_variable_name, :id_variable_name, :records_variable_name]
@@ -11,7 +12,7 @@ module IdentityCache
     private
 
     def clear_cached_associations
-      self.class.send(:all_cached_associations).each_value do |association|
+      self.class.all_cached_associations.each_value do |association|
         association.clear(self)
       end
     end

data/lib/identity_cache/cache_key_generation.rb

@@ -1,33 +1,36 @@
 # frozen_string_literal: true
+
 module IdentityCache
   module CacheKeyGeneration
     extend ActiveSupport::Concern
     DEFAULT_NAMESPACE = "IDC:#{CACHE_VERSION}:"
 
-    def self.schema_to_string(columns)
-      columns.sort_by(&:name).map { |c| "#{c.name}:#{c.type}" }.join(',')
-    end
+    class << self
+      def schema_to_string(columns)
+        columns.sort_by(&:name).map { |c| "#{c.name}:#{c.type}" }.join(",")
+      end
 
-    def self.denormalized_schema_string(klass)
-      schema_to_string(klass.columns).tap do |schema_string|
-        klass.send(:all_cached_associations).sort.each do |name, association|
-          klass.send(:check_association_scope, name)
-          association.validate if association.embedded?
-          case association
-          when Cached::Recursive::Association
-            schema_string << ",#{name}:(#{denormalized_schema_hash(association.reflection.klass)})"
-          when Cached::Reference::HasMany
-            schema_string << ",#{name}:ids"
-          when Cached::Reference::HasOne
-            schema_string << ",#{name}:id"
+      def denormalized_schema_string(klass)
+        schema_to_string(klass.columns).tap do |schema_string|
+          klass.all_cached_associations.sort.each do |name, association|
+            klass.send(:check_association_scope, name)
+            association.validate if association.embedded?
+            case association
+            when Cached::Recursive::Association
+              schema_string << ",#{name}:(#{denormalized_schema_hash(association.reflection.klass)})"
+            when Cached::Reference::HasMany
+              schema_string << ",#{name}:ids"
+            when Cached::Reference::HasOne
+              schema_string << ",#{name}:id"
+            end
           end
         end
       end
-    end
 
-    def self.denormalized_schema_hash(klass)
-      schema_string = denormalized_schema_string(klass)
-      IdentityCache.memcache_hash(schema_string)
+      def denormalized_schema_hash(klass)
+        schema_string = denormalized_schema_string(klass)
+        IdentityCache.memcache_hash(schema_string)
+      end
     end
 
     module ClassMethods

data/lib/identity_cache/cache_key_loader.rb

@@ -25,12 +25,12 @@ module IdentityCache
       # @param cache_fetcher [_CacheFetcher]
       # @param db_key Reference to what to load from the database.
       # @return The database value corresponding to the database key.
-      def load(cache_fetcher, db_key)
+      def load(cache_fetcher, db_key, cache_fetcher_options = {})
         cache_key = cache_fetcher.cache_key(db_key)
 
         db_value = nil
 
-        cache_value = IdentityCache.fetch(cache_key) do
+        cache_value = IdentityCache.fetch(cache_key, cache_fetcher_options) do
           db_value = cache_fetcher.load_one_from_db(db_key)
           cache_fetcher.cache_encode(db_value)
         end

data/lib/identity_cache/cached/association.rb

@@ -1,4 +1,5 @@
 # frozen_string_literal: true
+
 module IdentityCache
   module Cached
     class Association # :nodoc:
@@ -50,10 +51,7 @@ module IdentityCache
       end
 
       def inverse_name
-        @inverse_name ||= begin
-          reflection.inverse_of&.name ||
-          reflection.active_record.name.underscore
-        end
+        @inverse_name ||= reflection.inverse_of&.name || reflection.active_record.name.underscore
       end
 
       def validate

data/lib/identity_cache/cached/attribute.rb

@@ -87,11 +87,11 @@ module IdentityCache
 
       def cache_key_prefix
         @cache_key_prefix ||= begin
-          unique_indicator = unique ? '' : 's'
+          unique_indicator = unique ? "" : "s"
           "attr#{unique_indicator}" \
             ":#{model.base_class.name}" \
             ":#{attribute}" \
-            ":#{key_fields.join('/')}:"
+            ":#{key_fields.join("/")}:"
         end
       end
 
@@ -116,7 +116,7 @@ module IdentityCache
       end
 
       def fetch_method_suffix
-        "#{alias_name}_by_#{key_fields.join('_and_')}"
+        "#{alias_name}_by_#{key_fields.join("_and_")}"
       end
     end
   end

data/lib/identity_cache/cached/attribute_by_multi.rb

@@ -24,7 +24,7 @@ module IdentityCache
       end
 
       def unhashed_values_cache_key_string(key_values)
-        key_values.map { |v| v.try!(:to_s).inspect }.join('/')
+        key_values.map { |v| v.try!(:to_s).inspect }.join("/")
       end
 
       def load_from_db_where_conditions(key_values)

data/lib/identity_cache/cached/belongs_to.rb

@@ -1,4 +1,5 @@
 # frozen_string_literal: true
+
 module IdentityCache
   module Cached
     class BelongsTo < Association # :nodoc:
@@ -27,34 +28,43 @@ module IdentityCache
         end
       end
 
+      def write(owner_record, associated_record)
+        owner_record.instance_variable_set(records_variable_name, associated_record)
+      end
+
       def fetch(records)
         fetch_async(LoadStrategy::Eager, records) { |associated_records| associated_records }
       end
 
       def fetch_async(load_strategy, records)
         if reflection.polymorphic?
-          cache_keys_to_associated_ids = {}
+          type_fetcher_to_db_ids_hash = {}
 
           records.each do |owner_record|
             associated_id = owner_record.send(reflection.foreign_key)
             next unless associated_id && !owner_record.instance_variable_defined?(records_variable_name)
-            associated_cache_key = Object.const_get(
+
+            foreign_type_fetcher = Object.const_get(
               owner_record.send(reflection.foreign_type)
             ).cached_model.cached_primary_index
-            unless cache_keys_to_associated_ids[associated_cache_key]
-              cache_keys_to_associated_ids[associated_cache_key] = {}
-            end
-            cache_keys_to_associated_ids[associated_cache_key][associated_id] = owner_record
+            db_ids = type_fetcher_to_db_ids_hash[foreign_type_fetcher] ||= []
+            db_ids << associated_id
           end
 
-          load_strategy.load_batch(cache_keys_to_associated_ids) do |associated_records_by_cache_key|
+          load_strategy.load_batch(type_fetcher_to_db_ids_hash) do |batch_load_result|
             batch_records = []
-            associated_records_by_cache_key.each do |cache_key, associated_records|
-              associated_records.keys.each do |id, associated_record|
-                owner_record = cache_keys_to_associated_ids.fetch(cache_key).fetch(id)
-                batch_records << owner_record
-                owner_record.instance_variable_set(records_variable_name, associated_record)
-              end
+
+            records.each do |owner_record|
+              associated_id = owner_record.send(reflection.foreign_key)
+              next unless associated_id && !owner_record.instance_variable_defined?(records_variable_name)
+
+              foreign_type_fetcher = Object.const_get(
+                owner_record.send(reflection.foreign_type)
+              ).cached_model.cached_primary_index
+
+              associated_record = batch_load_result.fetch(foreign_type_fetcher).fetch(associated_id)
+              batch_records << owner_record
+              write(owner_record, associated_record)
             end
 
             yield batch_records
@@ -73,7 +83,7 @@ module IdentityCache
           ) do |associated_records_by_id|
             associated_records_by_id.each do |id, associated_record|
               owner_record = ids_to_owner_record.fetch(id)
-              owner_record.instance_variable_set(records_variable_name, associated_record)
+              write(owner_record, associated_record)
             end
 
             yield associated_records_by_id.values.compact

data/lib/identity_cache/cached/embedded_fetching.rb

@@ -1,4 +1,5 @@
 # frozen_string_literal: true
+
 module IdentityCache
   module Cached
     module EmbeddedFetching
@@ -22,6 +23,7 @@ module IdentityCache
           cached_associations.each_value do |cached_association|
             records.each do |record|
               next unless (cached_record = cached_records_by_id[record.id])
+
               cached_value = cached_association.read(cached_record)
               cached_association.write(record, cached_value)
             end

data/lib/identity_cache/cached/prefetcher.rb

@@ -37,14 +37,24 @@ module IdentityCache
         private
 
         def fetch_association(load_strategy, klass, association, records, &block)
-          unless records.first.class.should_use_cache?
-            ActiveRecord::Associations::Preloader.new.preload(records, association)
+          unless klass.should_use_cache?
+            preload_records(records, association)
             return yield
           end
 
           cached_association = klass.cached_association(association)
           cached_association.fetch_async(load_strategy, records, &block)
         end
+
+        if ActiveRecord.gem_version < Gem::Version.new("6.2.0.alpha")
+          def preload_records(records, association)
+            ActiveRecord::Associations::Preloader.new.preload(records, association)
+          end
+        else
+          def preload_records(records, association)
+            ActiveRecord::Associations::Preloader.new(records: records, associations: association).call
+          end
+        end
       end
     end
   end

data/lib/identity_cache/cached/primary_index.rb

@@ -9,11 +9,12 @@ module IdentityCache
         @model = model
       end
 
-      def fetch(id)
+      def fetch(id, cache_fetcher_options)
         id = cast_id(id)
         return unless id
+
         record = if model.should_use_cache?
-          object = CacheKeyLoader.load(self, id)
+          object = CacheKeyLoader.load(self, id, cache_fetcher_options)
           if object && object.id != id
             IdentityCache.logger.error(
               <<~MSG.squish
@@ -60,7 +61,6 @@ module IdentityCache
       def load_multi_from_db(ids)
         return {} if ids.empty?
 
-        ids = ids.map { |id| model.connection.type_cast(id, id_column) }
         records = build_query(ids).to_a
         model.send(:setup_embedded_associations_on_miss, records)
         records.index_by(&:id)