Class Executor

Inheritance Relationships

Derived Types

Class Documentation

class Executor

Coordinate the order and timing of available communication tasks.

Executor provides spin functions (including spin_node_once and spin_some). It coordinates the nodes and callback groups by looking for available work and completing it, based on the threading or concurrency scheme provided by the subclass implementation. An example of available work is executing a subscription callback, or a timer callback. The executor structure allows for a decoupling of the communication graph and the execution model. See SingleThreadedExecutor and MultiThreadedExecutor for examples of execution paradigms.

Subclassed by rclcpp::executors::MultiThreadedExecutor, rclcpp::executors::SingleThreadedExecutor, rclcpp::executors::StaticSingleThreadedExecutor, rclcpp::experimental::executors::EventsExecutor

Public Functions

explicit Executor(const rclcpp::ExecutorOptions &options = rclcpp::ExecutorOptions())

Default constructor.

Parameters:

options[in] Options used to configure the executor.

virtual ~Executor()

Default destructor.

virtual void spin() = 0

Do work periodically as it becomes available to us. Blocking call, may block indefinitely.

virtual void add_callback_group(rclcpp::CallbackGroup::SharedPtr group_ptr, rclcpp::node_interfaces::NodeBaseInterface::SharedPtr node_ptr, bool notify = true)

Add a callback group to an executor.

An executor can have zero or more callback groups which provide work during spin functions. When an executor attempts to add a callback group, the executor checks to see if it is already associated with another executor, and if it has been, then an exception is thrown. Otherwise, the callback group is added to the executor.

Adding a callback group with this method does not associate its node with this executor in any way

Parameters:
  • group_ptr[in] a shared ptr that points to a callback group

  • node_ptr[in] a shared pointer that points to a node base interface

  • notify[in] True to trigger the interrupt guard condition during this function. If the executor is blocked at the rmw layer while waiting for work and it is notified that a new callback group was added, it will wake up.

Throws:

std::runtime_error – if the callback group is associated to an executor

virtual std::vector<rclcpp::CallbackGroup::WeakPtr> get_all_callback_groups()

Get callback groups that belong to executor.

This function returns a vector of weak pointers that point to callback groups that were associated with the executor. The callback groups associated with this executor may have been added with add_callback_group, or added when a node was added to the executor with add_node, or automatically added when it created by a node already associated with this executor and the automatically_add_to_executor_with_node parameter was true.

Returns:

a vector of weak pointers that point to callback groups that are associated with the executor

virtual std::vector<rclcpp::CallbackGroup::WeakPtr> get_manually_added_callback_groups()

Get callback groups that belong to executor.

This function returns a vector of weak pointers that point to callback groups that were associated with the executor. The callback groups associated with this executor have been added with add_callback_group.

Returns:

a vector of weak pointers that point to callback groups that are associated with the executor

virtual std::vector<rclcpp::CallbackGroup::WeakPtr> get_automatically_added_callback_groups_from_nodes()

Get callback groups that belong to executor.

This function returns a vector of weak pointers that point to callback groups that were added from a node that is associated with the executor. The callback groups are added when a node is added to the executor with add_node, or automatically if they are created in the future by that node and have the automatically_add_to_executor_with_node argument set to true.

Returns:

a vector of weak pointers that point to callback groups from a node associated with the executor

virtual void remove_callback_group(rclcpp::CallbackGroup::SharedPtr group_ptr, bool notify = true)

Remove a callback group from the executor.

The callback group is removed from and disassociated with the executor. If the callback group removed was the last callback group from the node that is associated with the executor, the interrupt guard condition is triggered and node’s guard condition is removed from the executor.

This function only removes a callback group that was manually added with rclcpp::Executor::add_callback_group. To remove callback groups that were added from a node using rclcpp::Executor::add_node, use rclcpp::Executor::remove_node instead.

Parameters:
  • group_ptr[in] Shared pointer to the callback group to be added.

  • notify[in] True to trigger the interrupt guard condition during this function. If the executor is blocked at the rmw layer while waiting for work and it is notified that a callback group was removed, it will wake up.

Throws:
  • std::runtime_error – if node is deleted before callback group

  • std::runtime_error – if the callback group is not associated with the executor

virtual void add_node(rclcpp::node_interfaces::NodeBaseInterface::SharedPtr node_ptr, bool notify = true)

Add a node to the executor.

Nodes have associated callback groups, and this method adds any of those callback groups to this executor which have their automatically_add_to_executor_with_node parameter true. The node is also associated with the executor so that future callback groups which are created on the node with the automatically_add_to_executor_with_node parameter set to true are also automatically associated with this executor.

Callback groups with the automatically_add_to_executor_with_node parameter set to false must be manually added to an executor using the rclcpp::Executor::add_callback_group method.

If a node is already associated with an executor, this method throws an exception.

Parameters:
  • node_ptr[in] Shared pointer to the node to be added.

  • notify[in] True to trigger the interrupt guard condition during this function. If the executor is blocked at the rmw layer while waiting for work and it is notified that a new node was added, it will wake up.

Throws:

std::runtime_error – if a node is already associated to an executor

virtual void add_node(std::shared_ptr<rclcpp::Node> node_ptr, bool notify = true)

Convenience function which takes Node and forwards NodeBaseInterface.

virtual void remove_node(rclcpp::node_interfaces::NodeBaseInterface::SharedPtr node_ptr, bool notify = true)

Remove a node from the executor.

Any callback groups automatically added when this node was added with rclcpp::Executor::add_node are automatically removed, and the node is no longer associated with this executor.

This also means that future callback groups created by the given node are no longer automatically added to this executor.

Parameters:
  • node_ptr[in] Shared pointer to the node to remove.

  • notify[in] True to trigger the interrupt guard condition and wake up the executor. This is useful if the last node was removed from the executor while the executor was blocked waiting for work in another thread, because otherwise the executor would never be notified.

Throws:
  • std::runtime_error – if the node is not associated with an executor.

  • std::runtime_error – if the node is not associated with this executor.

virtual void remove_node(std::shared_ptr<rclcpp::Node> node_ptr, bool notify = true)

Convenience function which takes Node and forwards NodeBaseInterface.

template<typename RepT = int64_t, typename T = std::milli>
inline void spin_node_once(rclcpp::node_interfaces::NodeBaseInterface::SharedPtr node, std::chrono::duration<RepT, T> timeout = std::chrono::duration<RepT, T>(-1))

Add a node to executor, execute the next available unit of work, and remove the node.

Parameters:
  • node[in] Shared pointer to the node to add.

  • timeout[in] How long to wait for work to become available. Negative values cause spin_node_once to block indefinitely (the default behavior). A timeout of 0 causes this function to be non-blocking.

template<typename NodeT = rclcpp::Node, typename RepT = int64_t, typename T = std::milli>
inline void spin_node_once(std::shared_ptr<NodeT> node, std::chrono::duration<RepT, T> timeout = std::chrono::duration<RepT, T>(-1))

Convenience function which takes Node and forwards NodeBaseInterface.

void spin_node_some(rclcpp::node_interfaces::NodeBaseInterface::SharedPtr node)

Add a node, complete all immediately available work, and remove the node.

Parameters:

node[in] Shared pointer to the node to add.

void spin_node_some(std::shared_ptr<rclcpp::Node> node)

Convenience function which takes Node and forwards NodeBaseInterface.

virtual void spin_some(std::chrono::nanoseconds max_duration = std::chrono::nanoseconds(0))

Collect work once and execute all available work, optionally within a duration.

This function can be overridden. The default implementation is suitable for a single-threaded model of execution. Adding subscriptions, timers, services, etc. with blocking callbacks will cause this function to block (which may have unintended consequences).

Parameters:

max_duration[in] The maximum amount of time to spend executing work, or 0 for no limit. Note that spin_some() may take longer than this time as it only returns once max_duration has been exceeded.

virtual void spin_all(std::chrono::nanoseconds max_duration)

Collect and execute work repeatedly within a duration or until no more work is available.

This function can be overridden. The default implementation is suitable for a single-threaded model of execution. Adding subscriptions, timers, services, etc. with blocking callbacks will cause this function to block (which may have unintended consequences). If the time that waitables take to be executed is longer than the period on which new waitables become ready, this method will execute work repeatedly until max_duration has elapsed.

Parameters:

max_duration[in] The maximum amount of time to spend executing work, must be >= 0. 0 is potentially block forever until no more work is available.

Throws:

std::invalid_argument – if max_duration is less than 0. Note that spin_all() may take longer than this time as it only returns once max_duration has been exceeded.

virtual void spin_once(std::chrono::nanoseconds timeout = std::chrono::nanoseconds(-1))

Collect work once and execute the next available work, optionally within a duration.

This function can be overridden. The default implementation is suitable for a single-thread model of execution. Adding subscriptions, timers, services, etc. with blocking callbacks will cause this function to block (which may have unintended consequences).

Parameters:

timeout[in] The maximum amount of time to spend waiting for work. -1 is potentially block forever waiting for work.

template<typename FutureT, typename TimeRepT = int64_t, typename TimeT = std::milli>
inline FutureReturnCode spin_until_future_complete(const FutureT &future, std::chrono::duration<TimeRepT, TimeT> timeout = std::chrono::duration<TimeRepT, TimeT>(-1))

Spin (blocking) until the future is complete, it times out waiting, or rclcpp is interrupted.

Parameters:
  • future[in] The future to wait on. If this function returns SUCCESS, the future can be accessed without blocking (though it may still throw an exception).

  • timeout[in] Optional timeout parameter, which gets passed to Executor::spin_node_once. -1 is block forever, 0 is non-blocking. If the time spent inside the blocking loop exceeds this timeout, return a TIMEOUT return code.

Returns:

The return code, one of SUCCESS, INTERRUPTED, or TIMEOUT.

void cancel()

Cancel any running spin* function, causing it to return.

This function can be called asynchonously from any thread.

Throws:

std::runtime_error – if there is an issue triggering the guard condition

void set_memory_strategy(memory_strategy::MemoryStrategy::SharedPtr memory_strategy)

Support dynamic switching of the memory strategy.

Switching the memory strategy while the executor is spinning in another threading could have unintended consequences.

Parameters:

memory_strategy – Shared pointer to the memory strategy to set.

Throws:

std::runtime_error – if memory_strategy is null

bool is_spinning()

Returns true if the executor is currently spinning.

This function can be called asynchronously from any thread.

Returns:

True if the executor is currently spinning.

Protected Types

typedef std::map<rclcpp::node_interfaces::NodeBaseInterface::WeakPtr, const rclcpp::GuardCondition*, std::owner_less<rclcpp::node_interfaces::NodeBaseInterface::WeakPtr>> WeakNodesToGuardConditionsMap
typedef std::map<rclcpp::CallbackGroup::WeakPtr, const rclcpp::GuardCondition*, std::owner_less<rclcpp::CallbackGroup::WeakPtr>> WeakCallbackGroupsToGuardConditionsMap

Protected Functions

void spin_node_once_nanoseconds(rclcpp::node_interfaces::NodeBaseInterface::SharedPtr node, std::chrono::nanoseconds timeout)

Add a node to executor, execute the next available unit of work, and remove the node.

Implementation of spin_node_once using std::chrono::nanoseconds

Parameters:
  • node[in] Shared pointer to the node to add.

  • timeout[in] How long to wait for work to become available. Negative values cause spin_node_once to block indefinitely (the default behavior). A timeout of 0 causes this function to be non-blocking.

void spin_some_impl(std::chrono::nanoseconds max_duration, bool exhaustive)

Collect work and execute available work, optionally within a duration.

Implementation of spin_some and spin_all. The exhaustive flag controls if the function will re-collect available work within the duration.

Parameters:
  • max_duration[in] The maximum amount of time to spend executing work, or 0 for no limit.

  • exhaustive[in] when set to true, continue to collect work and execute (spin_all) when set to false, return when all collected work is executed (spin_some)

void execute_any_executable(AnyExecutable &any_exec)

Find the next available executable and do the work associated with it.

Parameters:

any_exec[in] Union structure that can hold any executable type (timer, subscription, service, client).

Throws:

std::runtime_error – if there is an issue triggering the guard condition

void wait_for_work(std::chrono::nanoseconds timeout = std::chrono::nanoseconds(-1))

Block until more work becomes avilable or timeout is reached.

Builds a set of waitable entities, which are passed to the middleware. After building wait set, waits on middleware to notify.

Parameters:

timeout[in] duration to wait for new work to become available.

Throws:

std::runtime_error – if the wait set can be cleared

rclcpp::node_interfaces::NodeBaseInterface::SharedPtr get_node_by_group(const WeakCallbackGroupsToNodesMap &weak_groups_to_nodes, rclcpp::CallbackGroup::SharedPtr group)

Find node associated with a callback group.

Parameters:
  • weak_groups_to_nodes[in] map of callback groups to nodes

  • group[in] callback group to find assocatiated node

Returns:

Pointer to associated node if found, else nullptr

bool has_node(const rclcpp::node_interfaces::NodeBaseInterface::SharedPtr node_ptr, const WeakCallbackGroupsToNodesMap &weak_groups_to_nodes) const

Return true if the node has been added to this executor.

Parameters:
  • node_ptr[in] a shared pointer that points to a node base interface

  • weak_groups_to_nodes[in] map to nodes to lookup

Returns:

true if the node is associated with the executor, otherwise false

rclcpp::CallbackGroup::SharedPtr get_group_by_timer(rclcpp::TimerBase::SharedPtr timer)

Find the callback group associated with a timer.

Parameters:

timer[in] Timer to find associated callback group

Returns:

Pointer to callback group node if found, else nullptr

virtual void add_callback_group_to_map(rclcpp::CallbackGroup::SharedPtr group_ptr, rclcpp::node_interfaces::NodeBaseInterface::SharedPtr node_ptr, WeakCallbackGroupsToNodesMap &weak_groups_to_nodes, bool notify = true)

Add a callback group to an executor.

virtual void remove_callback_group_from_map(rclcpp::CallbackGroup::SharedPtr group_ptr, WeakCallbackGroupsToNodesMap &weak_groups_to_nodes, bool notify = true)

Remove a callback group from the executor.

bool get_next_ready_executable(AnyExecutable &any_executable)

Check for executable in ready state and populate union structure.

Parameters:

any_executable[out] populated union structure of ready executable

Returns:

true if an executable was ready and any_executable was populated, otherwise false

bool get_next_ready_executable_from_map(AnyExecutable &any_executable, const WeakCallbackGroupsToNodesMap &weak_groups_to_nodes)

Check for executable in ready state and populate union structure.

This is the implementation of get_next_ready_executable that takes into account the current state of callback groups’ association with nodes and executors.

This checks in a particular order for available work:

  • Timers

  • Subscriptions

  • Services

  • Clients

  • Waitable

If the next executable is not associated with this executor/node pair, then this method will return false.

Parameters:
  • any_executable[out] populated union structure of ready executable

  • weak_groups_to_nodes[in] mapping of callback groups to nodes

Returns:

true if an executable was ready and any_executable was populated, otherwise false

bool get_next_executable(AnyExecutable &any_executable, std::chrono::nanoseconds timeout = std::chrono::nanoseconds(-1))

Wait for executable in ready state and populate union structure.

If an executable is ready, it will return immediately, otherwise block based on the timeout for work to become ready.

Parameters:
  • any_executable[out] populated union structure of ready executable

  • timeout[in] duration of time to wait for work, a negative value (the defualt behavior), will make this function block indefinitely

Returns:

true if an executable was ready and any_executable was populated, otherwise false

virtual void add_callback_groups_from_nodes_associated_to_executor()

Add all callback groups that can be automatically added from associated nodes.

The executor, before collecting entities, verifies if any callback group from nodes associated with the executor, which is not already associated to an executor, can be automatically added to this executor. This takes care of any callback group that has been added to a node but not explicitly added to the executor. It is important to note that in order for the callback groups to be automatically added to an executor through this function, the node of the callback groups needs to have been added through the add_node method.

virtual void spin_once_impl(std::chrono::nanoseconds timeout)

Protected Attributes

std::atomic_bool spinning

Spinning state, used to prevent multi threaded calls to spin and to cancel blocking spins.

std::shared_ptr<rclcpp::GuardCondition> interrupt_guard_condition_

Guard condition for signaling the rmw layer to wake up for special events.

std::shared_ptr<rclcpp::GuardCondition> shutdown_guard_condition_

Guard condition for signaling the rmw layer to wake up for system shutdown.

rcl_wait_set_t wait_set_ = rcl_get_zero_initialized_wait_set()

Wait set for managing entities that the rmw layer waits on.

mutable std::mutex mutex_
memory_strategy::MemoryStrategy::SharedPtr memory_strategy_

The memory strategy: an interface for handling user-defined memory allocation strategies.

std::shared_ptr<rclcpp::Context> context_

The context associated with this executor.

WeakNodesToGuardConditionsMap weak_nodes_to_guard_conditions_

maps nodes to guard conditions

WeakCallbackGroupsToGuardConditionsMap weak_groups_to_guard_conditions_

maps callback groups to guard conditions

WeakCallbackGroupsToNodesMap weak_groups_associated_with_executor_to_nodes_

maps callback groups associated to nodes

WeakCallbackGroupsToNodesMap weak_groups_to_nodes_associated_with_executor_

maps callback groups to nodes associated with executor

WeakCallbackGroupsToNodesMap weak_groups_to_nodes_

maps all callback groups to nodes

std::list<rclcpp::node_interfaces::NodeBaseInterface::WeakPtr> weak_nodes_

nodes that are associated with the executor

rclcpp::OnShutdownCallbackHandle shutdown_callback_handle_

shutdown callback handle registered to Context

std::unique_ptr<ExecutorImplementation> impl_

Pointer to implementation.

Protected Static Functions

static void execute_subscription(rclcpp::SubscriptionBase::SharedPtr subscription)

Run subscription executable.

Do necessary setup and tear-down as well as executing the subscription.

Parameters:

subscription[in] Subscription to execute

static void execute_timer(rclcpp::TimerBase::SharedPtr timer)

Run timer executable.

Do necessary setup and tear-down as well as executing the timer callback.

Parameters:

timer[in] Timer to execute

static void execute_service(rclcpp::ServiceBase::SharedPtr service)

Run service server executable.

Do necessary setup and tear-down as well as executing the service server callback.

Parameters:

service[in] Service to execute

static void execute_client(rclcpp::ClientBase::SharedPtr client)

Run service client executable.

Do necessary setup and tear-down as well as executing the service client callback.

Parameters:

service[in] Service to execute