graphql 2.0.31 → 2.5.18

data/lib/graphql/dataloader/null_dataloader.rb

@@ -2,23 +2,64 @@
 
 module GraphQL
   class Dataloader
-    # The default implementation of dataloading -- all no-ops.
+    # GraphQL-Ruby uses this when Dataloader isn't enabled.
     #
-    # The Dataloader interface isn't public, but it enables
-    # simple internal code while adding the option to add Dataloader.
+    # It runs execution code inline and gathers lazy objects (eg. Promises)
+    # and resolves them during {#run}.
     class NullDataloader < Dataloader
-      # These are all no-ops because code was
-      # executed sychronously.
-      def run; end
-      def run_isolated; yield; end
-      def yield
+      def initialize(*)
+        @lazies_at_depth = Hash.new { |h,k| h[k] = [] }
+      end
+
+      def freeze
+        @lazies_at_depth.default_proc = nil
+        @lazies_at_depth.freeze
+        super
+      end
+
+      def run(trace_query_lazy: nil)
+        with_trace_query_lazy(trace_query_lazy) do
+          while !@lazies_at_depth.empty?
+            smallest_depth = nil
+            @lazies_at_depth.each_key do |depth_key|
+              smallest_depth ||= depth_key
+              if depth_key < smallest_depth
+                smallest_depth = depth_key
+              end
+            end
+
+            if smallest_depth
+              lazies = @lazies_at_depth.delete(smallest_depth)
+              lazies.each(&:value) # resolve these Lazy instances
+            end
+          end
+        end
+      end
+
+      def run_isolated
+        new_dl = self.class.new
+        res = nil
+        new_dl.append_job {
+          res = yield
+        }
+        new_dl.run
+        res
+      end
+
+      def clear_cache; end
+
+      def yield(_source)
         raise GraphQL::Error, "GraphQL::Dataloader is not running -- add `use GraphQL::Dataloader` to your schema to use Dataloader sources."
       end
 
-      def append_job
-        yield
+      def append_job(callable = nil)
+        callable ? callable.call : yield
         nil
       end
+
+      def with(*)
+        raise GraphQL::Error, "GraphQL::Dataloader is not running -- add `use GraphQL::Dataloader` to your schema to use Dataloader sources."
+      end
     end
   end
 end

data/lib/graphql/dataloader/request.rb

@@ -14,6 +14,11 @@ module GraphQL
       def load
         @source.load(@key)
       end
+
+      def load_with_deprecation_warning
+        warn("Returning `.request(...)` from GraphQL::Dataloader is deprecated, use `.load(...)` instead. (See usage of #{@source} with #{@key.inspect}).")
+        load
+      end
     end
   end
 end

data/lib/graphql/dataloader/source.rb

@@ -21,7 +21,7 @@ module GraphQL
       def request(value)
         res_key = result_key_for(value)
         if !@results.key?(res_key)
-          @pending[res_key] ||= value
+          @pending[res_key] ||= normalize_fetch_key(value)
         end
         Dataloader::Request.new(self, value)
       end
@@ -35,12 +35,24 @@ module GraphQL
         value
       end
 
+      # Implement this method if varying values given to {load} (etc) should be consolidated
+      # or normalized before being handed off to your {fetch} implementation.
+      #
+      # This is different than {result_key_for} because _that_ method handles unification inside Dataloader's cache,
+      # but this method changes the value passed into {fetch}.
+      #
+      # @param value [Object] The value passed to {load}, {load_all}, {request}, or {request_all}
+      # @return [Object] The value given to {fetch}
+      def normalize_fetch_key(value)
+        value
+      end
+
       # @return [Dataloader::Request] a pending request for a values from `keys`. Call `.load` on that object to wait for the results.
       def request_all(values)
         values.each do |v|
           res_key = result_key_for(v)
           if !@results.key?(res_key)
-            @pending[res_key] ||= v
+            @pending[res_key] ||= normalize_fetch_key(v)
           end
         end
         Dataloader::RequestAll.new(self, values)
@@ -53,7 +65,7 @@ module GraphQL
         if @results.key?(result_key)
           result_for(result_key)
         else
-          @pending[result_key] ||= value
+          @pending[result_key] ||= normalize_fetch_key(value)
           sync([result_key])
           result_for(result_key)
         end
@@ -68,12 +80,12 @@ module GraphQL
           k = result_key_for(v)
           result_keys << k
           if !@results.key?(k)
-            @pending[k] ||= v
+            @pending[k] ||= normalize_fetch_key(v)
             pending_keys << k
           end
         }
 
-        if pending_keys.any?
+        if !pending_keys.empty?
           sync(pending_keys)
         end
 
@@ -88,18 +100,19 @@ module GraphQL
         raise "Implement `#{self.class}#fetch(#{keys.inspect}) to return a record for each of the keys"
       end
 
+      MAX_ITERATIONS = 1000
       # Wait for a batch, if there's anything to batch.
       # Then run the batch and update the cache.
       # @return [void]
       def sync(pending_result_keys)
-        @dataloader.yield
+        @dataloader.yield(self)
         iterations = 0
         while pending_result_keys.any? { |key| !@results.key?(key) }
           iterations += 1
-          if iterations > 1000
-            raise "#{self.class}#sync tried 1000 times to load pending keys (#{pending_result_keys}), but they still weren't loaded. There is likely a circular dependency."
+          if iterations > MAX_ITERATIONS
+            raise "#{self.class}#sync tried #{MAX_ITERATIONS} times to load pending keys (#{pending_result_keys}), but they still weren't loaded. There is likely a circular dependency#{@dataloader.fiber_limit ? " or `fiber_limit: #{@dataloader.fiber_limit}` is set too low" : ""}."
           end
-          @dataloader.yield
+          @dataloader.yield(self)
         end
         nil
       end
@@ -161,7 +174,14 @@ module GraphQL
         [*batch_args, **batch_kwargs]
       end
 
-      attr_reader :pending
+      # Clear any already-loaded objects for this source
+      # @return [void]
+      def clear_cache
+        @results.clear
+        nil
+      end
+
+      attr_reader :pending, :results
 
       private
 
@@ -178,8 +198,11 @@ This key should have been loaded already. This is a bug in GraphQL::Dataloader,
 ERR
         end
         result = @results[key]
-
-        raise result if result.class <= StandardError
+        if result.is_a?(StandardError)
+          # Dup it because the rescuer may modify it.
+          # (This happens for GraphQL::ExecutionErrors, at least)
+          raise result.dup
+        end
 
         result
       end

data/lib/graphql/dataloader.rb

@@ -4,6 +4,8 @@ require "graphql/dataloader/null_dataloader"
 require "graphql/dataloader/request"
 require "graphql/dataloader/request_all"
 require "graphql/dataloader/source"
+require "graphql/dataloader/active_record_association_source"
+require "graphql/dataloader/active_record_source"
 
 module GraphQL
   # This plugin supports Fiber-based concurrency, along with {GraphQL::Dataloader::Source}.
@@ -24,17 +26,23 @@ module GraphQL
   #
   class Dataloader
     class << self
-      attr_accessor :default_nonblocking
+      attr_accessor :default_nonblocking, :default_fiber_limit
     end
 
-    AsyncDataloader = Class.new(self) { self.default_nonblocking = true }
-
-    def self.use(schema, nonblocking: nil)
-      schema.dataloader_class = if nonblocking
-        AsyncDataloader
+    def self.use(schema, nonblocking: nil, fiber_limit: nil)
+      dataloader_class = if nonblocking
+        warn("`nonblocking: true` is deprecated from `GraphQL::Dataloader`, please use `GraphQL::Dataloader::AsyncDataloader` instead. Docs: https://graphql-ruby.org/dataloader/async_dataloader.")
+        Class.new(self) { self.default_nonblocking = true }
       else
         self
       end
+
+      if fiber_limit
+        dataloader_class = Class.new(dataloader_class)
+        dataloader_class.default_fiber_limit = fiber_limit
+      end
+
+      schema.dataloader_class = dataloader_class
     end
 
     # Call the block with a Dataloader instance,
@@ -49,18 +57,51 @@ module GraphQL
       result
     end
 
-    def initialize(nonblocking: self.class.default_nonblocking)
+    def initialize(nonblocking: self.class.default_nonblocking, fiber_limit: self.class.default_fiber_limit)
       @source_cache = Hash.new { |h, k| h[k] = {} }
       @pending_jobs = []
       if !nonblocking.nil?
         @nonblocking = nonblocking
       end
+      @fiber_limit = fiber_limit
+      @lazies_at_depth = Hash.new { |h, k| h[k] = [] }
     end
 
+    # @return [Integer, nil]
+    attr_reader :fiber_limit
+
     def nonblocking?
       @nonblocking
     end
 
+    # This is called before the fiber is spawned, from the parent context (i.e. from
+    # the thread or fiber that it is scheduled from).
+    #
+    # @return [Hash<Symbol, Object>] Current fiber-local variables
+    def get_fiber_variables
+      fiber_vars = {}
+      Thread.current.keys.each do |fiber_var_key|
+        fiber_vars[fiber_var_key] = Thread.current[fiber_var_key]
+      end
+      fiber_vars
+    end
+
+    # Set up the fiber variables in a new fiber.
+    #
+    # This is called within the fiber, right after it is spawned.
+    #
+    # @param vars [Hash<Symbol, Object>] Fiber-local variables from {get_fiber_variables}
+    # @return [void]
+    def set_fiber_variables(vars)
+      vars.each { |k, v| Thread.current[k] = v }
+      nil
+    end
+
+    # This method is called when Dataloader is finished using a fiber.
+    # Use it to perform any cleanup, such as releasing database connections (if required manually)
+    def cleanup_fiber
+    end
+
     # Get a Source instance from this dataloader, for calling `.load(...)` or `.request(...)` on.
     #
     # @param source_class [Class<GraphQL::Dataloader::Source]
@@ -91,16 +132,28 @@ module GraphQL
     # Dataloader will resume the fiber after the requested data has been loaded (by another Fiber).
     #
     # @return [void]
-    def yield
+    def yield(source = Fiber[:__graphql_current_dataloader_source])
+      trace = Fiber[:__graphql_current_multiplex]&.current_trace
+      trace&.dataloader_fiber_yield(source)
       Fiber.yield
+      trace&.dataloader_fiber_resume(source)
       nil
     end
 
     # @api private Nothing to see here
-    def append_job(&job)
+    def append_job(callable = nil, &job)
       # Given a block, queue it up to be worked through when `#run` is called.
-      # (If the dataloader is already running, than a Fiber will pick this up later.)
-      @pending_jobs.push(job)
+      # (If the dataloader is already running, then a Fiber will pick this up later.)
+      @pending_jobs.push(callable || job)
+      nil
+    end
+
+    # Clear any already-loaded objects from {Source} caches
+    # @return [void]
+    def clear_cache
+      @source_cache.each do |_source_class, batched_sources|
+        batched_sources.each_value(&:clear_cache)
+      end
       nil
     end
 
@@ -108,6 +161,10 @@ module GraphQL
     def run_isolated
       prev_queue = @pending_jobs
       prev_pending_keys = {}
+      prev_lazies_at_depth = @lazies_at_depth
+      @lazies_at_depth = @lazies_at_depth.dup.clear
+      # Clear pending loads but keep already-cached records
+      # in case they are useful to the given block.
       @source_cache.each do |source_class, batched_sources|
         batched_sources.each do |batch_args, batched_source_instance|
           if batched_source_instance.pending?
@@ -127,185 +184,203 @@ module GraphQL
       res
     ensure
       @pending_jobs = prev_queue
+      @lazies_at_depth = prev_lazies_at_depth
       prev_pending_keys.each do |source_instance, pending|
-        source_instance.pending.merge!(pending)
+        pending.each do |key, value|
+          if !source_instance.results.key?(key)
+            source_instance.pending[key] = value
+          end
+        end
       end
     end
 
-    # @api private Move along, move along
-    def run
-      if @nonblocking && !Fiber.scheduler
-        raise "`nonblocking: true` requires `Fiber.scheduler`, assign one with `Fiber.set_scheduler(...)` before executing GraphQL."
-      end
-      # At a high level, the algorithm is:
-      #
-      #  A) Inside Fibers, run jobs from the queue one-by-one
-      #    - When one of the jobs yields to the dataloader (`Fiber.yield`), then that fiber will pause
-      #    - In that case, if there are still pending jobs, a new Fiber will be created to run jobs
-      #    - Continue until all jobs have been _started_ by a Fiber. (Any number of those Fibers may be waiting to be resumed, after their data is loaded)
-      #  B) Once all known jobs have been run until they are complete or paused for data, run all pending data sources.
-      #    - Similarly, create a Fiber to consume pending sources and tell them to load their data.
-      #    - If one of those Fibers pauses, then create a new Fiber to continue working through remaining pending sources.
-      #    - When a source causes another source to become pending, run the newly-pending source _first_, since it's a dependency of the previous one.
-      #  C) After all pending sources have been completely loaded (there are no more pending sources), resume any Fibers that were waiting for data.
-      #    - Those Fibers assume that source caches will have been populated with the data they were waiting for.
-      #    - Those Fibers may request data from a source again, in which case they will yeilded and be added to a new pending fiber list.
-      #  D) Once all pending fibers have been resumed once, return to `A` above.
-      #
-      # For whatever reason, the best implementation I could find was to order the steps `[D, A, B, C]`, with a special case for skipping `D`
-      # on the first pass. I just couldn't find a better way to write the loops in a way that was DRY and easy to read.
-      #
-      pending_fibers = []
-      next_fibers = []
-      pending_source_fibers = []
+    # @param trace_query_lazy [nil, Execution::Multiplex]
+    def run(trace_query_lazy: nil)
+      trace = Fiber[:__graphql_current_multiplex]&.current_trace
+      jobs_fiber_limit, total_fiber_limit = calculate_fiber_limit
+      job_fibers = []
+      next_job_fibers = []
+      source_fibers = []
       next_source_fibers = []
       first_pass = true
-
-      while first_pass || (f = pending_fibers.shift)
-        if first_pass
+      manager = spawn_fiber do
+        trace&.begin_dataloader(self)
+        while first_pass || !job_fibers.empty?
           first_pass = false
-        else
-          # These fibers were previously waiting for sources to load data,
-          # resume them. (They might wait again, in which case, re-enqueue them.)
-          resume(f)
-          if f.alive?
-            next_fibers << f
-          end
-        end
 
-        while @pending_jobs.any?
-          # Create a Fiber to consume jobs until one of the jobs yields
-          # or jobs run out
-          f = spawn_fiber {
-            while (job = @pending_jobs.shift)
-              job.call
+          run_pending_steps(trace, job_fibers, next_job_fibers, jobs_fiber_limit, source_fibers, next_source_fibers, total_fiber_limit)
+
+          if !@lazies_at_depth.empty?
+            with_trace_query_lazy(trace_query_lazy) do
+              run_next_pending_lazies(job_fibers, trace)
+              run_pending_steps(trace, job_fibers, next_job_fibers, jobs_fiber_limit, source_fibers, next_source_fibers, total_fiber_limit)
             end
-          }
-          resume(f)
-          # In this case, the job yielded. Queue it up to run again after
-          # we load whatever it's waiting for.
-          if f.alive?
-            next_fibers << f
           end
         end
 
-        if pending_fibers.empty?
-          # Now, run all Sources which have become pending _before_ resuming GraphQL execution.
-          # Sources might queue up other Sources, which is fine -- those will also run before resuming execution.
-          #
-          # This is where an evented approach would be even better -- can we tell which
-          # fibers are ready to continue, and continue execution there?
-          #
-          if (first_source_fiber = create_source_fiber)
-            pending_source_fibers << first_source_fiber
-          end
+        trace&.end_dataloader(self)
+      end
 
-          while pending_source_fibers.any?
-            while (outer_source_fiber = pending_source_fibers.pop)
-              resume(outer_source_fiber)
-              if outer_source_fiber.alive?
-                next_source_fibers << outer_source_fiber
-              end
-              if (next_source_fiber = create_source_fiber)
-                pending_source_fibers << next_source_fiber
-              end
-            end
-            join_queues(pending_source_fibers, next_source_fibers)
-            next_source_fibers.clear
-          end
-          # Move newly-enqueued Fibers on to the list to be resumed.
-          # Clear out the list of next-round Fibers, so that
-          # any Fibers that pause can be put on it.
-          join_queues(pending_fibers, next_fibers)
-          next_fibers.clear
-        end
+      run_fiber(manager)
+
+      if manager.alive?
+        raise "Invariant: Manager fiber didn't terminate properly."
       end
 
-      if @pending_jobs.any?
-        raise "Invariant: #{@pending_jobs.size} pending jobs"
-      elsif pending_fibers.any?
-        raise "Invariant: #{pending_fibers.size} pending fibers"
-      elsif next_fibers.any?
-        raise "Invariant: #{next_fibers.size} next fibers"
+      if !job_fibers.empty?
+        raise "Invariant: job fibers should have exited but #{job_fibers.size} remained"
       end
-      nil
+      if !source_fibers.empty?
+        raise "Invariant: source fibers should have exited but #{source_fibers.size} remained"
+      end
+
+    rescue UncaughtThrowError => e
+      throw e.tag, e.value
     end
 
-    def join_queues(previous_queue, next_queue)
-      if @nonblocking
-        Fiber.scheduler.run
-        next_queue.select!(&:alive?)
+    def run_fiber(f)
+      f.resume
+    end
+
+    # @api private
+    def lazy_at_depth(depth, lazy)
+      @lazies_at_depth[depth] << lazy
+    end
+
+    def spawn_fiber
+      fiber_vars = get_fiber_variables
+      Fiber.new(blocking: !@nonblocking) {
+        set_fiber_variables(fiber_vars)
+        yield
+        cleanup_fiber
+      }
+    end
+
+    # Pre-warm the Dataloader cache with ActiveRecord objects which were loaded elsewhere.
+    # These will be used by {Dataloader::ActiveRecordSource}, {Dataloader::ActiveRecordAssociationSource} and their helper
+    # methods, `dataload_record` and `dataload_association`.
+    # @param records [Array<ActiveRecord::Base>] Already-loaded records to warm the cache with
+    # @param index_by [Symbol] The attribute to use as the cache key. (Should match `find_by:` when using {ActiveRecordSource})
+    # @return [void]
+    def merge_records(records, index_by: :id)
+      records_by_class = Hash.new { |h, k| h[k] = {} }
+      records.each do |r|
+        records_by_class[r.class][r.public_send(index_by)] = r
+      end
+      records_by_class.each do |r_class, records|
+        with(ActiveRecordSource, r_class).merge(records)
       end
-      previous_queue.concat(next_queue)
     end
 
     private
 
-    # If there are pending sources, return a fiber for running them.
-    # Otherwise, return `nil`.
-    #
-    # @return [Fiber, nil]
-    def create_source_fiber
-      pending_sources = nil
-      @source_cache.each_value do |source_by_batch_params|
-        source_by_batch_params.each_value do |source|
-          if source.pending?
-            pending_sources ||= []
-            pending_sources << source
+    def run_next_pending_lazies(job_fibers, trace)
+      smallest_depth = nil
+      @lazies_at_depth.each_key do |depth_key|
+        smallest_depth ||= depth_key
+        if depth_key < smallest_depth
+          smallest_depth = depth_key
+        end
+      end
+
+      if smallest_depth
+        lazies = @lazies_at_depth.delete(smallest_depth)
+        if !lazies.empty?
+          lazies.each_with_index do |l, idx|
+            append_job { l.value }
           end
+          job_fibers.unshift(spawn_job_fiber(trace))
         end
       end
+    end
 
-      if pending_sources
-        # By passing the whole array into this Fiber, it's possible that we set ourselves up for a bunch of no-ops.
-        # For example, if you have sources `[a, b, c]`, and `a` is loaded, then `b` yields to wait for `d`, then
-        # the next fiber would be dispatched with `[c, d]`. It would fulfill `c`, then `d`, then eventually
-        # the previous fiber would start up again. `c` would no longer be pending, but it would still receive `.run_pending_keys`.
-        # That method is short-circuited since it isn't pending any more, but it's still a waste.
-        #
-        # This design could probably be improved by maintaining a `@pending_sources` queue which is shared by the fibers,
-        # similar to `@pending_jobs`. That way, when a fiber is resumed, it would never pick up work that was finished by a different fiber.
-        source_fiber = spawn_fiber do
-          pending_sources.each(&:run_pending_keys)
+    def run_pending_steps(trace, job_fibers, next_job_fibers, jobs_fiber_limit, source_fibers, next_source_fibers, total_fiber_limit)
+      while (f = (job_fibers.shift || (((next_job_fibers.size + job_fibers.size) < jobs_fiber_limit) && spawn_job_fiber(trace))))
+        if f.alive?
+          finished = run_fiber(f)
+          if !finished
+            next_job_fibers << f
+          end
         end
       end
+      join_queues(job_fibers, next_job_fibers)
 
-      source_fiber
+      while (!source_fibers.empty? || @source_cache.each_value.any? { |group_sources| group_sources.each_value.any?(&:pending?) })
+        while (f = source_fibers.shift || (((job_fibers.size + source_fibers.size + next_source_fibers.size + next_job_fibers.size) < total_fiber_limit) && spawn_source_fiber(trace)))
+          if f.alive?
+            finished = run_fiber(f)
+            if !finished
+              next_source_fibers << f
+            end
+          end
+        end
+        join_queues(source_fibers, next_source_fibers)
+      end
     end
 
-    def resume(fiber)
-      fiber.resume
-    rescue UncaughtThrowError => e
-      throw e.tag, e.value
+    def with_trace_query_lazy(multiplex_or_nil, &block)
+      if (multiplex = multiplex_or_nil)
+        query = multiplex.queries.length == 1 ? multiplex.queries[0] : nil
+        multiplex.current_trace.execute_query_lazy(query: query, multiplex: multiplex, &block)
+      else
+        yield
+      end
     end
 
-    # Copies the thread local vars into the fiber thread local vars. Many
-    # gems (such as RequestStore, MiniRacer, etc.) rely on thread local vars
-    # to keep track of execution context, and without this they do not
-    # behave as expected.
-    #
-    # @see https://github.com/rmosolgo/graphql-ruby/issues/3449
-    def spawn_fiber
-      fiber_locals = {}
+    def calculate_fiber_limit
+      total_fiber_limit = @fiber_limit || Float::INFINITY
+      if total_fiber_limit < 4
+        raise ArgumentError, "Dataloader fiber limit is too low (#{total_fiber_limit}), it must be at least 4"
+      end
+      total_fiber_limit -= 1 # deduct one fiber for `manager`
+      # Deduct at least one fiber for sources
+      jobs_fiber_limit = total_fiber_limit - 2
+      return jobs_fiber_limit, total_fiber_limit
+    end
 
-      Thread.current.keys.each do |fiber_var_key|
-        # This variable should be fresh in each new fiber
-        if fiber_var_key != :__graphql_runtime_info
-          fiber_locals[fiber_var_key] = Thread.current[fiber_var_key]
+    def join_queues(prev_queue, new_queue)
+      @nonblocking && Fiber.scheduler.run
+      prev_queue.concat(new_queue)
+      new_queue.clear
+    end
+
+    def spawn_job_fiber(trace)
+      if !@pending_jobs.empty?
+        spawn_fiber do
+          trace&.dataloader_spawn_execution_fiber(@pending_jobs)
+          while job = @pending_jobs.shift
+            job.call
+          end
+          trace&.dataloader_fiber_exit
         end
       end
+    end
 
-      if @nonblocking
-        Fiber.new(blocking: false) do
-          fiber_locals.each { |k, v| Thread.current[k] = v }
-          yield
+    def spawn_source_fiber(trace)
+      pending_sources = nil
+      @source_cache.each_value do |source_by_batch_params|
+        source_by_batch_params.each_value do |source|
+          if source.pending?
+            pending_sources ||= []
+            pending_sources << source
+          end
         end
-      else
-        Fiber.new do
-          fiber_locals.each { |k, v| Thread.current[k] = v }
-          yield
+      end
+
+      if pending_sources
+        spawn_fiber do
+          trace&.dataloader_spawn_source_fiber(pending_sources)
+          pending_sources.each do |source|
+            Fiber[:__graphql_current_dataloader_source] = source
+            trace&.begin_dataloader_source(source)
+            source.run_pending_keys
+            trace&.end_dataloader_source(source)
+          end
+          trace&.dataloader_fiber_exit
         end
       end
     end
   end
 end
+
+require "graphql/dataloader/async_dataloader"