holoscan::IOSpec

holoscan::IOSpec::IOSpec(holoscan::IOSpec::IOSpec(
    OperatorSpec *op_spec,
    const std::string &name,
    IOType io_type,
    const std::type_info *typeinfo = &typeid(void *),
    IOSpec::IOSize size = IOSpec::kSizeOne,    IOSpec::IOSize size = IOSpec::kSizeOne,    IOSpec::IOSize size = IOSpec::kSizeOne,
    std::optional<IOSpec::QueuePolicy> policy = std::nullopt
)

Construct a new IOSpec object.

Parameters

std::shared_ptr<Resource> holoscan::IOSpec::connector() conststd::shared_ptr<Resource> holoscan::IOSpec::connector() const

Get the connector (transmitter or receiver) of this input/output.

Returns: The connector (transmitter or receiver) of this input/output.

IOSpec & holoscan::IOSpec::topic(IOSpec & holoscan::IOSpec::topic(
    const std::string &name,
    bool replace_connector = false
)

Set a topic name for this port (Publish/Subscribe connectors only).

If the connector type is kDefault, it is automatically changed to kPubSub and the topic is applied. If the connector type is already kPubSub, the topic is set on the existing connector. For any other connector type (e.g. kUCX, kDoubleBuffer), a warning is logged and the topic is ignored unless replace_connector is true, in which case the existing connector is replaced with a Pub/Sub connector.

When both publisher and subscriber use the same topic, add_flow() is optional; add_operator() for each is sufficient — Pub/Sub backend discovery matches by topic.

Returns: Reference to this IOSpec.

Parameters

IOSpec & holoscan::IOSpec::qos(IOSpec & holoscan::IOSpec::qos(
    const nvidia::gxf::QoSProfile &profile
)

Set a QoS profile for this port (Publish/Subscribe connectors only).

Uses the backend-independent nvidia::gxf::QoSProfile struct. Named presets (e.g. QoSProfile::SensorData(), QoSProfile::Reliable()) and the fluent builder API are both supported.

If the connector type is kDefault, it is automatically changed to kPubSub and the QoS is applied. If the connector type is already kPubSub, the QoS is set on the existing connector. For any other connector type, a warning is logged and the QoS is ignored.

Example:

The DDS backend maps QoSProfile through dds_qos_profiles.hpp. Other backends (e.g. Zenoh, UCX) would map the same struct to their own QoS mechanisms.

Returns: Reference to this IOSpec.

Parameters

Example

void setup(OperatorSpec& spec) override {
  spec.output<MyType>("out")
      .topic("sensor_data")
      .qos(nvidia::gxf::QoSProfile::SensorData());
  spec.input<MyType>("in")
      .topic("sensor_data")
      .qos(nvidia::gxf::QoSProfile::SensorData());
}

IOSpec & holoscan::IOSpec::queue_size(IOSpec & holoscan::IOSpec::queue_size(
    int64_t size
)

Set the queue size of the input/output port.

Note: This value is only used for initializing input ports. ‘queue_size_’ is set by the ‘OperatorSpec::input()’ method or this method. If the queue size is set to ‘any size’ (IOSpec::kAnySize in C++ or IOSpec.ANY_SIZE in Python), the connector/condition settings will be ignored. If the queue size is set to other values, the default connector (DoubleBufferReceiver/UcxReceiver) and condition (MessageAvailableCondition) will use the queue size for initialization (‘capacity’ for the connector and ‘min_size’ for the condition) if they are not set.

Note: For input ports, a queue size greater than 1 will also set the default MessageAvailableCondition min_size to the same value unless you override the port condition. This enables batched execution. Holoscan emits a warning in the default-condition case. If you only want buffering without batching, explicitly set min_size=1. Holoscan plans to introduce an explicit batch_size configuration and change queue_size/size to control queue capacity only; setting min_size explicitly now will make future migration straightforward.

Returns: The reference to this IOSpec.

Parameters

IOSpec & holoscan::IOSpec::queue_policy(IOSpec & holoscan::IOSpec::queue_policy(
    IOSpec::QueuePolicy policy
)

Set the queue policy of the input/output port.

Note: This value is only used for initializing input and output ports. ‘queue_policy_’ is set by the ‘OperatorSpec::input()’, ‘OperatorSpec::output()’ or ‘IOSpec::queue_policy’ method.

The following IOSpec::QueuePolicy values are supported:

• QueuePolicy::kPop - If the queue is full, pop the oldest item, then add the new one.
• QueuePolicy::kReject - If the queue is full, reject (discard) the new item.
• QueuePolicy::kFault - If the queue is full, log a warning and reject the new item.

Returns: The reference to this IOSpec.

Parameters