temporalio 0.2.0 → 0.3.0

data/lib/temporalio/internal/worker/workflow_instance/outbound_implementation.rb

@@ -0,0 +1,415 @@
+# frozen_string_literal: true
+
+require 'temporalio/activity/definition'
+require 'temporalio/cancellation'
+require 'temporalio/error'
+require 'temporalio/internal/bridge/api'
+require 'temporalio/internal/proto_utils'
+require 'temporalio/internal/worker/workflow_instance'
+require 'temporalio/worker/interceptor'
+require 'temporalio/workflow'
+require 'temporalio/workflow/child_workflow_handle'
+
+module Temporalio
+  module Internal
+    module Worker
+      class WorkflowInstance
+        # Root implementation of the outbound interceptor.
+        class OutboundImplementation < Temporalio::Worker::Interceptor::Workflow::Outbound
+          def initialize(instance)
+            super(nil) # steep:ignore
+            @instance = instance
+            @activity_counter = 0
+            @timer_counter = 0
+            @child_counter = 0
+            @external_signal_counter = 0
+            @external_cancel_counter = 0
+          end
+
+          def cancel_external_workflow(input)
+            # Add command
+            seq = (@external_cancel_counter += 1)
+            cmd = Bridge::Api::WorkflowCommands::RequestCancelExternalWorkflowExecution.new(
+              seq:,
+              workflow_execution: Bridge::Api::Common::NamespacedWorkflowExecution.new(
+                namespace: @instance.info.namespace,
+                workflow_id: input.id,
+                run_id: input.run_id
+              )
+            )
+            @instance.add_command(
+              Bridge::Api::WorkflowCommands::WorkflowCommand.new(request_cancel_external_workflow_execution: cmd)
+            )
+            @instance.pending_external_cancels[seq] = Fiber.current
+
+            # Wait
+            resolution = Fiber.yield
+
+            # Raise if resolution has failure
+            return unless resolution.failure
+
+            raise @instance.failure_converter.from_failure(resolution.failure, @instance.payload_converter)
+          end
+
+          def execute_activity(input)
+            if input.schedule_to_close_timeout.nil? && input.start_to_close_timeout.nil?
+              raise ArgumentError, 'Activity must have schedule_to_close_timeout or start_to_close_timeout'
+            end
+
+            activity_type = case input.activity
+                            when Class
+                              Activity::Definition::Info.from_activity(input.activity).name
+                            when Symbol, String
+                              input.activity.to_s
+                            else
+                              raise ArgumentError, 'Activity must be a definition class, or a symbol/string'
+                            end
+            raise 'Cannot invoke dynamic activities' unless activity_type
+
+            execute_activity_with_local_backoffs(local: false, cancellation: input.cancellation) do
+              seq = (@activity_counter += 1)
+              @instance.add_command(
+                Bridge::Api::WorkflowCommands::WorkflowCommand.new(
+                  schedule_activity: Bridge::Api::WorkflowCommands::ScheduleActivity.new(
+                    seq:,
+                    activity_id: input.activity_id || seq.to_s,
+                    activity_type:,
+                    task_queue: input.task_queue,
+                    headers: ProtoUtils.headers_to_proto_hash(input.headers, @instance.payload_converter),
+                    arguments: ProtoUtils.convert_to_payload_array(@instance.payload_converter, input.args),
+                    schedule_to_close_timeout: ProtoUtils.seconds_to_duration(input.schedule_to_close_timeout),
+                    schedule_to_start_timeout: ProtoUtils.seconds_to_duration(input.schedule_to_start_timeout),
+                    start_to_close_timeout: ProtoUtils.seconds_to_duration(input.start_to_close_timeout),
+                    heartbeat_timeout: ProtoUtils.seconds_to_duration(input.heartbeat_timeout),
+                    retry_policy: input.retry_policy&._to_proto,
+                    cancellation_type: input.cancellation_type,
+                    do_not_eagerly_execute: input.disable_eager_execution
+                  )
+                )
+              )
+              seq
+            end
+          end
+
+          def execute_local_activity(input)
+            if input.schedule_to_close_timeout.nil? && input.start_to_close_timeout.nil?
+              raise ArgumentError, 'Activity must have schedule_to_close_timeout or start_to_close_timeout'
+            end
+
+            activity_type = case input.activity
+                            when Class
+                              Activity::Definition::Info.from_activity(input.activity).name
+                            when Symbol, String
+                              input.activity.to_s
+                            else
+                              raise ArgumentError, 'Activity must be a definition class, or a symbol/string'
+                            end
+            raise 'Cannot invoke dynamic activities' unless activity_type
+
+            execute_activity_with_local_backoffs(local: true, cancellation: input.cancellation) do |do_backoff|
+              seq = (@activity_counter += 1)
+              @instance.add_command(
+                Bridge::Api::WorkflowCommands::WorkflowCommand.new(
+                  schedule_local_activity: Bridge::Api::WorkflowCommands::ScheduleLocalActivity.new(
+                    seq:,
+                    activity_id: input.activity_id || seq.to_s,
+                    activity_type:,
+                    headers: ProtoUtils.headers_to_proto_hash(input.headers, @instance.payload_converter),
+                    arguments: ProtoUtils.convert_to_payload_array(@instance.payload_converter, input.args),
+                    schedule_to_close_timeout: ProtoUtils.seconds_to_duration(input.schedule_to_close_timeout),
+                    schedule_to_start_timeout: ProtoUtils.seconds_to_duration(input.schedule_to_start_timeout),
+                    start_to_close_timeout: ProtoUtils.seconds_to_duration(input.start_to_close_timeout),
+                    retry_policy: input.retry_policy&._to_proto,
+                    cancellation_type: input.cancellation_type,
+                    local_retry_threshold: ProtoUtils.seconds_to_duration(input.local_retry_threshold),
+                    attempt: do_backoff&.attempt || 0,
+                    original_schedule_time: do_backoff&.original_schedule_time
+                  )
+                )
+              )
+              seq
+            end
+          end
+
+          def execute_activity_with_local_backoffs(local:, cancellation:, &)
+            # We do not even want to schedule if the cancellation is already cancelled. We choose to use canceled
+            # failure instead of wrapping in activity failure which is similar to what other SDKs do, with the accepted
+            # tradeoff that it makes rescue more difficult (hence the presence of Error.canceled? helper).
+            raise Error::CanceledError, 'Activity canceled before scheduled' if cancellation.canceled?
+
+            # This has to be done in a loop for local activity backoff
+            last_local_backoff = nil
+            loop do
+              result = execute_activity_once(local:, cancellation:, last_local_backoff:, &)
+              return result unless result.is_a?(Bridge::Api::ActivityResult::DoBackoff)
+
+              # @type var result: untyped
+              last_local_backoff = result
+              # Have to sleep the amount of the backoff, which can be canceled with the same cancellation
+              # TODO(cretz): What should this cancellation raise?
+              Workflow.sleep(ProtoUtils.duration_to_seconds(result.backoff_duration), cancellation:)
+            end
+          end
+
+          # If this doesn't raise, it returns success | DoBackoff
+          def execute_activity_once(local:, cancellation:, last_local_backoff:, &)
+            # Add to pending activities (removed by the resolver)
+            seq = yield last_local_backoff
+            @instance.pending_activities[seq] = Fiber.current
+
+            # Add cancellation hook
+            cancel_callback_key = cancellation.add_cancel_callback do
+              # Only if the activity is present still
+              if @instance.pending_activities.include?(seq)
+                if local
+                  @instance.add_command(
+                    Bridge::Api::WorkflowCommands::WorkflowCommand.new(
+                      request_cancel_local_activity: Bridge::Api::WorkflowCommands::RequestCancelLocalActivity.new(seq:)
+                    )
+                  )
+                else
+                  @instance.add_command(
+                    Bridge::Api::WorkflowCommands::WorkflowCommand.new(
+                      request_cancel_activity: Bridge::Api::WorkflowCommands::RequestCancelActivity.new(seq:)
+                    )
+                  )
+                end
+              end
+            end
+
+            # Wait
+            resolution = Fiber.yield
+
+            # Remove cancellation callback
+            cancellation.remove_cancel_callback(cancel_callback_key)
+
+            case resolution.status
+            when :completed
+              @instance.payload_converter.from_payload(resolution.completed.result)
+            when :failed
+              raise @instance.failure_converter.from_failure(resolution.failed.failure, @instance.payload_converter)
+            when :cancelled
+              raise @instance.failure_converter.from_failure(resolution.cancelled.failure, @instance.payload_converter)
+            when :backoff
+              resolution.backoff
+            else
+              raise "Unrecognized resolution status: #{resolution.status}"
+            end
+          end
+
+          def initialize_continue_as_new_error(input)
+            # Do nothing
+          end
+
+          def signal_child_workflow(input)
+            _signal_external_workflow(
+              id: input.id,
+              run_id: nil,
+              child: true,
+              signal: input.signal,
+              args: input.args,
+              cancellation: input.cancellation,
+              headers: input.headers
+            )
+          end
+
+          def signal_external_workflow(input)
+            _signal_external_workflow(
+              id: input.id,
+              run_id: input.run_id,
+              child: false,
+              signal: input.signal,
+              args: input.args,
+              cancellation: input.cancellation,
+              headers: input.headers
+            )
+          end
+
+          def _signal_external_workflow(id:, run_id:, child:, signal:, args:, cancellation:, headers:)
+            raise Error::CanceledError, 'Signal canceled before scheduled' if cancellation.canceled?
+
+            # Add command
+            seq = (@external_signal_counter += 1)
+            cmd = Bridge::Api::WorkflowCommands::SignalExternalWorkflowExecution.new(
+              seq:,
+              signal_name: Workflow::Definition::Signal._name_from_parameter(signal),
+              args: ProtoUtils.convert_to_payload_array(@instance.payload_converter, args),
+              headers: ProtoUtils.headers_to_proto_hash(headers, @instance.payload_converter)
+            )
+            if child
+              cmd.child_workflow_id = id
+            else
+              cmd.workflow_execution = Bridge::Api::Common::NamespacedWorkflowExecution.new(
+                namespace: @instance.info.namespace,
+                workflow_id: id,
+                run_id:
+              )
+            end
+            @instance.add_command(
+              Bridge::Api::WorkflowCommands::WorkflowCommand.new(signal_external_workflow_execution: cmd)
+            )
+            @instance.pending_external_signals[seq] = Fiber.current
+
+            # Add a cancellation callback
+            cancel_callback_key = cancellation.add_cancel_callback do
+              # Add the command but do not raise, we will let resolution do that
+              @instance.add_command(
+                Bridge::Api::WorkflowCommands::WorkflowCommand.new(
+                  cancel_signal_workflow: Bridge::Api::WorkflowCommands::CancelSignalWorkflow.new(seq:)
+                )
+              )
+            end
+
+            # Wait
+            resolution = Fiber.yield
+
+            # Remove cancellation callback
+            cancellation.remove_cancel_callback(cancel_callback_key)
+
+            # Raise if resolution has failure
+            return unless resolution.failure
+
+            raise @instance.failure_converter.from_failure(resolution.failure, @instance.payload_converter)
+          end
+
+          def sleep(input)
+            # If already cancelled, raise as such
+            if input.cancellation.canceled?
+              raise Error::CanceledError,
+                    input.cancellation.canceled_reason || 'Timer canceled before started'
+            end
+
+            # Disallow negative durations
+            raise ArgumentError, 'Sleep duration cannot be less than 0' if input.duration&.negative?
+
+            # If the duration is infinite, just wait for cancellation
+            if input.duration.nil?
+              input.cancellation.wait
+              raise Error::CanceledError, input.cancellation.canceled_reason || 'Timer canceled'
+            end
+
+            # If duration is zero, we make it one millisecond. It was decided a 0 duration still makes a timer to ensure
+            # determinism if a timer's duration is altered from non-zero to zero or vice versa.
+            duration = input.duration
+            duration = 0.001 if duration.zero?
+
+            # Add command
+            seq = (@timer_counter += 1)
+            @instance.add_command(
+              Bridge::Api::WorkflowCommands::WorkflowCommand.new(
+                start_timer: Bridge::Api::WorkflowCommands::StartTimer.new(
+                  seq:,
+                  start_to_fire_timeout: ProtoUtils.seconds_to_duration(duration)
+                )
+              )
+            )
+            @instance.pending_timers[seq] = Fiber.current
+
+            # Add a cancellation callback
+            cancel_callback_key = input.cancellation.add_cancel_callback do
+              # Only if the timer is still present
+              fiber = @instance.pending_timers.delete(seq)
+              if fiber
+                # Add the command for cancel then raise
+                @instance.add_command(
+                  Bridge::Api::WorkflowCommands::WorkflowCommand.new(
+                    cancel_timer: Bridge::Api::WorkflowCommands::CancelTimer.new(seq:)
+                  )
+                )
+                if fiber.alive?
+                  fiber.raise(Error::CanceledError.new(input.cancellation.canceled_reason || 'Timer canceled'))
+                end
+              end
+            end
+
+            # Wait
+            Fiber.yield
+
+            # Remove cancellation callback (only needed on success)
+            input.cancellation.remove_cancel_callback(cancel_callback_key)
+          end
+
+          def start_child_workflow(input)
+            raise Error::CanceledError, 'Child canceled before scheduled' if input.cancellation.canceled?
+
+            # Add the command
+            seq = (@child_counter += 1)
+            @instance.add_command(
+              Bridge::Api::WorkflowCommands::WorkflowCommand.new(
+                start_child_workflow_execution: Bridge::Api::WorkflowCommands::StartChildWorkflowExecution.new(
+                  seq:,
+                  namespace: @instance.info.namespace,
+                  workflow_id: input.id,
+                  workflow_type: Workflow::Definition._workflow_type_from_workflow_parameter(input.workflow),
+                  task_queue: input.task_queue,
+                  input: ProtoUtils.convert_to_payload_array(@instance.payload_converter, input.args),
+                  workflow_execution_timeout: ProtoUtils.seconds_to_duration(input.execution_timeout),
+                  workflow_run_timeout: ProtoUtils.seconds_to_duration(input.run_timeout),
+                  workflow_task_timeout: ProtoUtils.seconds_to_duration(input.task_timeout),
+                  parent_close_policy: input.parent_close_policy,
+                  workflow_id_reuse_policy: input.id_reuse_policy,
+                  retry_policy: input.retry_policy&._to_proto,
+                  cron_schedule: input.cron_schedule,
+                  headers: ProtoUtils.headers_to_proto_hash(input.headers, @instance.payload_converter),
+                  memo: ProtoUtils.memo_to_proto_hash(input.memo, @instance.payload_converter),
+                  search_attributes: input.search_attributes&._to_proto_hash,
+                  cancellation_type: input.cancellation_type
+                )
+              )
+            )
+
+            # Set as pending start and register cancel callback
+            @instance.pending_child_workflow_starts[seq] = Fiber.current
+            cancel_callback_key = input.cancellation.add_cancel_callback do
+              # Send cancel if in start or pending
+              if @instance.pending_child_workflow_starts.include?(seq) ||
+                 @instance.pending_child_workflows.include?(seq)
+                @instance.add_command(
+                  Bridge::Api::WorkflowCommands::WorkflowCommand.new(
+                    cancel_child_workflow_execution: Bridge::Api::WorkflowCommands::CancelChildWorkflowExecution.new(
+                      child_workflow_seq: seq
+                    )
+                  )
+                )
+              end
+            end
+
+            # Wait for start
+            resolution = Fiber.yield
+
+            case resolution.status
+            when :succeeded
+              # Create handle, passing along the cancel callback key, and set it as pending
+              handle = ChildWorkflowHandle.new(
+                id: input.id,
+                first_execution_run_id: resolution.succeeded.run_id,
+                instance: @instance,
+                cancellation: input.cancellation,
+                cancel_callback_key:
+              )
+              @instance.pending_child_workflows[seq] = handle
+              handle
+            when :failed
+              # Remove cancel callback and handle failure
+              input.cancellation.remove_cancel_callback(cancel_callback_key)
+              if resolution.failed.cause == :START_CHILD_WORKFLOW_EXECUTION_FAILED_CAUSE_WORKFLOW_ALREADY_EXISTS
+                raise Error::WorkflowAlreadyStartedError.new(
+                  workflow_id: resolution.failed.workflow_id,
+                  workflow_type: resolution.failed.workflow_type,
+                  run_id: nil
+                )
+              end
+              raise "Unknown child start fail cause: #{resolution.failed.cause}"
+            when :cancelled
+              # Remove cancel callback and handle cancel
+              input.cancellation.remove_cancel_callback(cancel_callback_key)
+              raise @instance.failure_converter.from_failure(resolution.cancelled.failure, @instance.payload_converter)
+            else
+              raise "Unknown resolution status: #{resolution.status}"
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/lib/temporalio/internal/worker/workflow_instance/replay_safe_logger.rb

@@ -0,0 +1,37 @@
+# frozen_string_literal: true
+
+require 'temporalio/scoped_logger'
+require 'temporalio/workflow'
+
+module Temporalio
+  module Internal
+    module Worker
+      class WorkflowInstance
+        # Wrapper for a scoped logger that does not log on replay.
+        class ReplaySafeLogger < ScopedLogger
+          def initialize(logger:, instance:)
+            @instance = instance
+            @replay_safety_disabled = false
+            super(logger)
+          end
+
+          def replay_safety_disabled(&)
+            @replay_safety_disabled = true
+            yield
+          ensure
+            @replay_safety_disabled = false
+          end
+
+          def add(...)
+            if !@replay_safety_disabled && Temporalio::Workflow.in_workflow? && Temporalio::Workflow::Unsafe.replaying?
+              return true
+            end
+
+            # Disable illegal call tracing for the log call
+            @instance.illegal_call_tracing_disabled { super }
+          end
+        end
+      end
+    end
+  end
+end

data/lib/temporalio/internal/worker/workflow_instance/replay_safe_metric.rb

@@ -0,0 +1,40 @@
+# frozen_string_literal: true
+
+require 'temporalio/scoped_logger'
+
+module Temporalio
+  module Internal
+    module Worker
+      class WorkflowInstance
+        # Wrapper for a metric that does not log on replay.
+        class ReplaySafeMetric < SimpleDelegator
+          def record(value, additional_attributes: nil)
+            return if Temporalio::Workflow.in_workflow? && Temporalio::Workflow::Unsafe.replaying?
+
+            super
+          end
+
+          def with_additional_attributes(additional_attributes)
+            ReplaySafeMetric.new(super)
+          end
+
+          class Meter < SimpleDelegator
+            def create_metric(
+              metric_type,
+              name,
+              description: nil,
+              unit: nil,
+              value_type: :integer
+            )
+              ReplaySafeMetric.new(super)
+            end
+
+            def with_additional_attributes(additional_attributes)
+              Meter.new(super)
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/lib/temporalio/internal/worker/workflow_instance/scheduler.rb

@@ -0,0 +1,163 @@
+# frozen_string_literal: true
+
+require 'temporalio'
+require 'temporalio/cancellation'
+require 'temporalio/error'
+require 'temporalio/internal/worker/workflow_instance'
+require 'temporalio/workflow'
+require 'timeout'
+
+module Temporalio
+  module Internal
+    module Worker
+      class WorkflowInstance
+        # Deterministic {::Fiber::Scheduler} implementation.
+        class Scheduler
+          def initialize(instance)
+            @instance = instance
+            @fibers = []
+            @ready = []
+            @wait_conditions = {}
+            @wait_condition_counter = 0
+          end
+
+          def context
+            @instance.context
+          end
+
+          def run_until_all_yielded
+            loop do
+              # Run all fibers until all yielded
+              while (fiber = @ready.shift)
+                fiber.resume
+              end
+
+              # Find the _first_ resolvable wait condition and if there, resolve it, and loop again, otherwise return.
+              # It is important that we both let fibers get all settled _before_ this and only allow a _single_ wait
+              # condition to be satisfied before looping. This allows wait condition users to trust that the line of
+              # code after the wait condition still has the condition satisfied.
+              # @type var cond_fiber: Fiber?
+              cond_fiber = nil
+              cond_result = nil
+              @wait_conditions.each do |seq, cond|
+                next unless (cond_result = cond.first.call)
+
+                cond_fiber = cond[1]
+                @wait_conditions.delete(seq)
+                break
+              end
+              return unless cond_fiber
+
+              cond_fiber.resume(cond_result)
+            end
+          end
+
+          def wait_condition(cancellation:, &block)
+            raise Workflow::InvalidWorkflowStateError, 'Cannot wait in this context' if @instance.context_frozen
+
+            if cancellation&.canceled?
+              raise Error::CanceledError,
+                    cancellation.canceled_reason || 'Wait condition canceled before started'
+            end
+
+            seq = (@wait_condition_counter += 1)
+            @wait_conditions[seq] = [block, Fiber.current]
+
+            # Add a cancellation callback
+            cancel_callback_key = cancellation&.add_cancel_callback do
+              # Only if the condition is still present
+              cond = @wait_conditions.delete(seq)
+              if cond&.last&.alive?
+                cond&.last&.raise(Error::CanceledError.new(cancellation&.canceled_reason || 'Wait condition canceled'))
+              end
+            end
+
+            # This blocks until a resume is called on this fiber
+            result = Fiber.yield
+
+            # Remove cancellation callback (only needed on success)
+            cancellation&.remove_cancel_callback(cancel_callback_key) if cancel_callback_key
+
+            result
+          end
+
+          def stack_trace
+            # Collect backtraces of known fibers, separating with a blank line. We make sure to remove any lines that
+            # reference Temporal paths, and we remove any empty backtraces.
+            dir_path = @instance.illegal_call_tracing_disabled { File.dirname(Temporalio._root_file_path) }
+            @fibers.map do |fiber|
+              fiber.backtrace.reject { |s| s.start_with?(dir_path) }.join("\n")
+            end.reject(&:empty?).join("\n\n")
+          end
+
+          ###
+          # Fiber::Scheduler methods
+          #
+          # Note, we do not implement many methods here such as io_read and
+          # such. While it might seem to make sense to implement them and
+          # raise, we actually want to default to the blocking behavior of them
+          # not being present. This is so advanced things like logging still
+          # work inside of workflows. So we only implement the bare minimum.
+          ###
+
+          def block(_blocker, timeout = nil)
+            # TODO(cretz): Make the blocker visible in the stack trace?
+
+            # We just yield because unblock will resume this. We will just wrap in timeout if needed.
+            if timeout
+              begin
+                Timeout.timeout(timeout) { Fiber.yield }
+                true
+              rescue Timeout::Error
+                false
+              end
+            else
+              Fiber.yield
+              true
+            end
+          end
+
+          def close
+            # Nothing to do here, lifetime of scheduler is controlled by the instance
+          end
+
+          def fiber(&block)
+            if @instance.context_frozen
+              raise Workflow::InvalidWorkflowStateError, 'Cannot schedule fibers in this context'
+            end
+
+            fiber = Fiber.new do
+              block.call # steep:ignore
+            ensure
+              @fibers.delete(Fiber.current)
+            end
+            @fibers << fiber
+            @ready << fiber
+            fiber
+          end
+
+          def io_wait(io, events, timeout)
+            # TODO(cretz): This in a blocking fashion?
+            raise NotImplementedError, 'TODO'
+          end
+
+          def kernel_sleep(duration = nil)
+            Workflow.sleep(duration)
+          end
+
+          def process_wait(pid, flags)
+            raise NotImplementedError, 'Cannot wait on other processes in workflows'
+          end
+
+          def timeout_after(duration, exception_class, *exception_arguments, &)
+            context.timeout(duration, exception_class, *exception_arguments, summary: 'Timeout timer', &)
+          end
+
+          def unblock(_blocker, fiber)
+            @ready << fiber
+          end
+        end
+      end
+    end
+  end
+end