Client
netft::Client owns configuration discovery, one RDT session, recovery, health accounting, and sample delivery. Constructing it validates configuration but does not start network activity.
netft::Client
class NETFT_API Client;Member functions
Client
explicit Client(Config config);| Parameter | Type | Default |
|---|---|---|
config | Config | Required |
~Client
~Client();start
void start(SampleCallback callback);| Parameter | Type | Default |
|---|---|---|
callback | SampleCallback | Required |
Returns: void
stop
void stop() noexcept;Returns: void
Qualifiers: noexcept
bias
void bias();Returns: void
wait_for_first_sample
bool wait_for_first_sample(std::chrono::duration<double> timeout);| Parameter | Type | Default |
|---|---|---|
timeout | std::chrono::duration<double> | Required |
Returns: bool
faulted
[[nodiscard]] bool faulted() const noexcept;Returns: bool
Qualifiers: constnoexcept
fault_code
[[nodiscard]] FaultCode fault_code() const noexcept;Returns: FaultCode
Qualifiers: constnoexcept
health
[[nodiscard]] HealthSnapshot health() const;Returns: HealthSnapshot
Qualifiers: const
latest_sample
[[nodiscard]] std::optional<Sample> latest_sample() const;Returns: std::optional<Sample>
Qualifiers: const
Lifecycle and concurrency
Call start() once for a run and keep its callback bounded. The callback executes on the native receive worker, so blocking storage or application work must be transferred to an application-owned bounded queue. stop() is idempotent and non-throwing.
Destruction normally waits for worker shutdown. Destruction initiated by the sample callback cannot join its own thread, so shutdown and storage reclamation are deferred; the callback must not access the client after initiating destruction.
The type is neither copyable nor movable. latest_sample() does not consume delivery, and health() returns a point-in-time snapshot. Use structured state, fault code, status, and counters for decisions rather than parsing last_error.
Readiness
wait_for_first_sample() is a bounded readiness check for the current run. A false result establishes only that a sample was not observed before the boundary; inspect faulted(), fault_code(), and health() to distinguish timeout, stop, and a terminal fail-stop fault.