Forward ensemble request parameters to composing models (#261)

Author: rmccorm4

src/ensemble_scheduler.cc

@@ -347,6 +347,9 @@ class EnsembleContext {
   InferenceRequest::SequenceId correlation_id_;
   uint64_t priority_;
   uint64_t timeout_;
+  // The parameters of the ensemble-level request. Use this to forward
+  // parameters to composing models.
+  std::deque<InferenceParameter> parameters_;
 
   // Objects related to the ensemble infer request
   Status ensemble_status_;
@@ -462,6 +465,7 @@ EnsembleContext::EnsembleContext(
     flags_ = lrequest->Flags();
     priority_ = lrequest->Priority();
     timeout_ = lrequest->TimeoutMicroseconds();
+    parameters_ = lrequest->Parameters();
 
     for (const auto& pr : lrequest->ImmutableInputs()) {
       const InferenceRequest::Input* input = pr.second;
@@ -951,6 +955,7 @@ EnsembleContext::InitStep(
   irequest->SetFlags(flags);
   irequest->SetPriority(priority_);
   irequest->SetTimeoutMicroseconds(timeout_);
+  irequest->SetParameters(parameters_);
 #ifdef TRITON_ENABLE_STATS
   irequest->SetSecondaryStatsAggregator(
       &request_tracker_->ContextStatsAggregator());

src/infer_request.cc

@@ -241,6 +241,15 @@ InferenceRequest::AddParameter(const char* name, const bool value)
   return Status::Success;
 }
 
+Status
+InferenceRequest::SetParameters(
+    const std::deque<InferenceParameter>& parameters)
+{
+  // NOTE: For BYTES parameters, this will shallow copy the pointer for now.
+  parameters_ = parameters;
+  return Status::Success;
+}
+
 #ifdef TRITON_ENABLE_TRACING
 Status
 InferenceRequest::TraceInputTensors(

src/infer_request.h

@@ -362,10 +362,11 @@ class InferenceRequest {
       TRITONSERVER_InferenceTraceActivity activity, const std::string& msg);
 #endif  // TRITON_ENABLE_TRACING
 
-  // Add an parameter to the request.
+  // Add a parameter to the request.
   Status AddParameter(const char* name, const char* value);
   Status AddParameter(const char* name, const int64_t value);
   Status AddParameter(const char* name, const bool value);
+  Status SetParameters(const std::deque<InferenceParameter>& parameters);
   const std::deque<InferenceParameter>& Parameters() const
   {
     return parameters_;