feat: Metrics API enhanced for Failure counters (#377)

Enhances existing metrics API to incorporate failure and failure counts in inference requests
---------

Co-authored-by: Kevin Wang <[email protected]>
Co-authored-by: krishung5 <[email protected]>

Author: 3 people

src/backend_model_instance.cc

@@ -36,6 +36,7 @@
 #include "backend_config.h"
 #include "backend_model.h"
 #include "cuda_utils.h"
+#include "infer_stats.h"
 #include "metrics.h"
 #include "model_config.pb.h"
 #include "numa_utils.h"
@@ -558,7 +559,8 @@ TritonModelInstance::PrepareRequestsOrRespond(
   // If any errors occurred, respond with error for each request.
   if (!status.IsOk()) {
     for (auto& r : requests) {
-      InferenceRequest::RespondIfError(r, status, true /* release_requests */);
+      InferenceRequest::RespondIfError(
+          r, status, true /* release_requests */, FailureReason::OTHER);
     }
     // Log a single error for batch of requests for better visibility
     LOG_STATUS_ERROR(status, "Requests failed pre-execution checks");
@@ -685,7 +687,16 @@ TritonModelInstance::Execute(
     for (TRITONBACKEND_Request* tr : triton_requests) {
       std::unique_ptr<InferenceRequest> ur(
           reinterpret_cast<InferenceRequest*>(tr));
-      InferenceRequest::RespondIfError(ur, status, true /* release_requests */);
+      // NOTE: If a backend both returns an error in
+      // TRITONBACKEND_ModelInstanceExecute and reports an error with
+      // TRITONBACKEND_ModelInstanceReportStatistics, this can result in double
+      // counting of the failure metric for the same request. However, it is
+      // currently not expected for this to be a common case, as the return
+      // value of TRITONBACKEND_ModelInstanceExecute is used to express
+      // ownership of the request rather than success of an inference request.
+      // See tritonbackend.h for more details on this.
+      InferenceRequest::RespondIfError(
+          ur, status, true /* release_requests */, FailureReason::BACKEND);
     }
 
     TRITONSERVER_ErrorDelete(err);

src/dynamic_batch_scheduler.cc

@@ -50,11 +50,12 @@ IsStaleState(Payload::State payload_state)
 void
 FinishSkippedRequests(
     std::vector<std::deque<std::unique_ptr<InferenceRequest>>>&& requests,
-    const Status& response_status)
+    const Status& response_status, FailureReason reason)
 {
   for (auto& queue : requests) {
     for (auto& request : queue) {
-      InferenceRequest::RespondIfError(request, response_status, true);
+      InferenceRequest::RespondIfError(
+          request, response_status, true /* release_requests */, reason);
     }
   }
 }
@@ -69,8 +70,10 @@ FinishRejectedCancelledRequests(
   const static Status rejected_status =
       Status(Status::Code::UNAVAILABLE, "Request timeout expired");
   const static Status cancelled_status = Status(Status::Code::CANCELLED);
-  FinishSkippedRequests(std::move(rejected_requests), rejected_status);
-  FinishSkippedRequests(std::move(cancelled_requests), cancelled_status);
+  FinishSkippedRequests(
+      std::move(rejected_requests), rejected_status, FailureReason::REJECTED);
+  FinishSkippedRequests(
+      std::move(cancelled_requests), cancelled_status, FailureReason::CANCELED);
 }
 
 DynamicBatchScheduler::DynamicBatchScheduler(

src/ensemble_scheduler/ensemble_scheduler.cc

@@ -81,23 +81,26 @@ class RequestTracker {
     std::lock_guard<std::mutex> lk(mtx_);
     inflight_request_counter_--;
     if (inflight_request_counter_ == 0) {
+      if (request_ != nullptr) {
 #ifdef TRITON_ENABLE_STATS
-      const auto& infer_stats = context_stats_aggregator_.ImmutableInferStats();
-      request_->ReportStatisticsWithDuration(
-          metric_reporter_, status_.IsOk(), compute_start_ns_,
-          infer_stats.compute_input_duration_ns_,
-          infer_stats.compute_infer_duration_ns_,
-          infer_stats.compute_output_duration_ns_);
-      if (status_.IsOk()) {
-        stats_aggregator_->UpdateInferBatchStatsWithDuration(
-            metric_reporter_, std::max(1U, request_->BatchSize()),
+        const auto& infer_stats =
+            context_stats_aggregator_.ImmutableInferStats();
+        request_->ReportStatisticsWithDuration(
+            metric_reporter_, status_.IsOk(), compute_start_ns_,
             infer_stats.compute_input_duration_ns_,
             infer_stats.compute_infer_duration_ns_,
             infer_stats.compute_output_duration_ns_);
-      }
+        if (status_.IsOk()) {
+          stats_aggregator_->UpdateInferBatchStatsWithDuration(
+              metric_reporter_, std::max(1U, request_->BatchSize()),
+              infer_stats.compute_input_duration_ns_,
+              infer_stats.compute_infer_duration_ns_,
+              infer_stats.compute_output_duration_ns_);
+        }
 #endif
-      InferenceRequest::Release(
-          std::move(request_), TRITONSERVER_REQUEST_RELEASE_ALL);
+        InferenceRequest::Release(
+            std::move(request_), TRITONSERVER_REQUEST_RELEASE_ALL);
+      }
     }
     return (inflight_request_counter_ == 0);
   }
@@ -1136,7 +1139,8 @@ EnsembleContext::FinishEnsemble(std::unique_ptr<InferenceResponse>&& response)
                 "more "
                 "ensemble steps can be made");
         InferenceRequest::RespondIfError(
-            request_tracker_->Request(), ensemble_status_);
+            request_tracker_->Request(), ensemble_status_,
+            false /* release_requests */, FailureReason::OTHER);
       } else {
         request_tracker_->Request()->ResponseFactory()->SendFlags(
             TRITONSERVER_RESPONSE_COMPLETE_FINAL);
@@ -1149,7 +1153,8 @@ EnsembleContext::FinishEnsemble(std::unique_ptr<InferenceResponse>&& response)
           ensemble_status_);
     } else {
       InferenceRequest::RespondIfError(
-          request_tracker_->Request(), ensemble_status_);
+          request_tracker_->Request(), ensemble_status_,
+          false /* release_requests */, FailureReason::OTHER);
     }
   }
 

src/infer_request.cc

@@ -421,10 +421,25 @@ InferenceRequest::Run(std::unique_ptr<InferenceRequest>& request)
   return status;
 }
 
+FailureReason
+stringToFailureReason(const std::string& error_type)
+{
+  if (error_type == "REJECTED") {
+    return FailureReason::REJECTED;
+  }
+  if (error_type == "CANCELED") {
+    return FailureReason::CANCELED;
+  }
+  if (error_type == "BACKEND") {
+    return FailureReason::BACKEND;
+  }
+  return FailureReason::OTHER;
+}
+
 void
 InferenceRequest::RespondIfError(
     std::unique_ptr<InferenceRequest>& request, const Status& status,
-    const bool release_request)
+    const bool release_request, FailureReason reason)
 {
   if (status.IsOk()) {
     return;
@@ -442,7 +457,10 @@ InferenceRequest::RespondIfError(
       InferenceResponse::SendWithStatus(
           std::move(response), TRITONSERVER_RESPONSE_COMPLETE_FINAL, status),
       (request->LogRequest() + "failed to send error response").c_str());
-
+#ifdef TRITON_ENABLE_STATS
+  request->ReportErrorStatistics(
+      request->model_raw_->MetricReporter().get(), reason);
+#endif
   // If releasing the request then invoke the release callback which
   // gives ownership to the callback. So can't access 'request' after
   // this point.
@@ -452,20 +470,6 @@ InferenceRequest::RespondIfError(
   }
 }
 
-void
-InferenceRequest::RespondIfError(
-    std::vector<std::unique_ptr<InferenceRequest>>& requests,
-    const Status& status, const bool release_requests)
-{
-  if (status.IsOk()) {
-    return;
-  }
-
-  for (auto& request : requests) {
-    RespondIfError(request, status, release_requests);
-  }
-}
-
 Status
 InferenceRequest::Release(
     std::unique_ptr<InferenceRequest>&& request, const uint32_t release_flags)
@@ -1371,6 +1375,21 @@ InferenceRequest::ValidateBytesInputs(
 }
 
 #ifdef TRITON_ENABLE_STATS
+
+void
+InferenceRequest::ReportErrorStatistics(
+    MetricModelReporter* metric_reporter, FailureReason reason)
+{
+  INFER_STATS_DECL_TIMESTAMP(request_end_ns);
+  model_raw_->MutableStatsAggregator()->UpdateFailure(
+      metric_reporter, request_start_ns_, request_end_ns, reason);
+  if (secondary_stats_aggregator_ != nullptr) {
+    secondary_stats_aggregator_->UpdateFailure(
+        nullptr /* metric_reporter */, request_start_ns_, request_end_ns,
+        reason);
+  }
+}
+
 void
 InferenceRequest::ReportStatistics(
     MetricModelReporter* metric_reporter, bool success,
@@ -1407,10 +1426,12 @@ InferenceRequest::ReportStatistics(
     }
   } else {
     model_raw_->MutableStatsAggregator()->UpdateFailure(
-        metric_reporter, request_start_ns_, request_end_ns);
+        metric_reporter, request_start_ns_, request_end_ns,
+        FailureReason::BACKEND);
     if (secondary_stats_aggregator_ != nullptr) {
       secondary_stats_aggregator_->UpdateFailure(
-          nullptr /* metric_reporter */, request_start_ns_, request_end_ns);
+          nullptr /* metric_reporter */, request_start_ns_, request_end_ns,
+          FailureReason::BACKEND);
     }
   }
 }
@@ -1443,10 +1464,12 @@ InferenceRequest::ReportStatisticsWithDuration(
     }
   } else {
     model_raw_->MutableStatsAggregator()->UpdateFailure(
-        metric_reporter, request_start_ns_, request_end_ns);
+        metric_reporter, request_start_ns_, request_end_ns,
+        FailureReason::OTHER);
     if (secondary_stats_aggregator_ != nullptr) {
       secondary_stats_aggregator_->UpdateFailure(
-          nullptr /* metric_reporter */, request_start_ns_, request_end_ns);
+          nullptr /* metric_reporter */, request_start_ns_, request_end_ns,
+          FailureReason::OTHER);
     }
   }
 }
@@ -1850,5 +1873,4 @@ operator!=(
 {
   return !(lhs == rhs);
 }
-
 }}  // namespace triton::core

src/infer_request.h

@@ -590,7 +590,8 @@ class InferenceRequest {
   // 'release_request' is true 'request' is returned as nullptr.
   static void RespondIfError(
       std::unique_ptr<InferenceRequest>& request, const Status& status,
-      const bool release_request = false);
+      const bool release_request = false,
+      FailureReason reason = FailureReason::OTHER);
 
   // Send an error response to a set of 'requests'. If 'status' is
   // Success then no responses are sent and the requests are not
@@ -603,7 +604,8 @@ class InferenceRequest {
   // returned with all nullptrs.
   static void RespondIfError(
       std::vector<std::unique_ptr<InferenceRequest>>& requests,
-      const Status& status, const bool release_requests = false);
+      const Status& status, const bool release_requests = false,
+      FailureReason reason = FailureReason::OTHER);
 
   // Release the request. Call the release callback and transfer
   // ownership of the request to the callback. On return 'request' is
@@ -673,6 +675,16 @@ class InferenceRequest {
       const uint64_t compute_start_ns, const uint64_t compute_input_end_ns,
       const uint64_t compute_output_start_ns, const uint64_t compute_end_ns);
 
+  // Report the error statistics to stats collectors associated with the
+  // request.
+  // FIXME: A separate function may not be necessary here, but is being used
+  // cautiously in case of unforeseen issues such as possibly capturing a trace
+  // twice. This should be revisited and better tested to see if the
+  // ReportStatistics function can be used as-is for the newly captured failure
+  // cases.
+  void ReportErrorStatistics(
+      MetricModelReporter* metric_reporter, FailureReason reason);
+
   // Report the statistics to stats collectors associated with the request.
   // Duration and timestamps provide two granularities for stats collectors.
   void ReportStatisticsWithDuration(

src/infer_stats.cc

@@ -36,10 +36,28 @@ namespace triton { namespace core {
 
 #ifdef TRITON_ENABLE_STATS
 
+// This function converts FailureReason enum values to std::string
+std::string
+failureReasonToString(FailureReason reason)
+{
+  switch (reason) {
+    case FailureReason::REJECTED:
+      return "REJECTED";
+    case FailureReason::CANCELED:
+      return "CANCELED";
+    case FailureReason::BACKEND:
+      return "BACKEND";
+    case FailureReason::OTHER:
+      return "OTHER";
+    default:
+      return "OTHER";
+  }
+}
+
 void
 InferenceStatsAggregator::UpdateFailure(
     MetricModelReporter* metric_reporter, const uint64_t request_start_ns,
-    const uint64_t request_end_ns)
+    const uint64_t request_end_ns, FailureReason reason)
 {
   std::lock_guard<std::mutex> lock(mu_);
 
@@ -48,7 +66,8 @@ InferenceStatsAggregator::UpdateFailure(
 
 #ifdef TRITON_ENABLE_METRICS
   if (metric_reporter != nullptr) {
-    metric_reporter->IncrementCounter("inf_failure", 1);
+    std::string reason_str = failureReasonToString(reason);
+    metric_reporter->IncrementCounter("inf_failure_" + reason_str, 1);
   }
 #endif  // TRITON_ENABLE_METRICS
 }

src/infer_stats.h

@@ -39,6 +39,9 @@
 
 namespace triton { namespace core {
 
+// Define the FailureReason enum within the triton::core namespace
+enum class FailureReason { REJECTED, CANCELED, BACKEND, OTHER };
+
 class MetricModelReporter;
 
 
@@ -136,7 +139,7 @@ class InferenceStatsAggregator {
   // Add durations to Infer stats for a failed inference request.
   void UpdateFailure(
       MetricModelReporter* metric_reporter, const uint64_t request_start_ns,
-      const uint64_t request_end_ns);
+      const uint64_t request_end_ns, FailureReason reason);
 
   // Add durations to infer stats for a successful inference request.
   void UpdateSuccess(

src/metric_model_reporter.cc

@@ -29,6 +29,7 @@
 #ifdef TRITON_ENABLE_METRICS
 
 #include "constants.h"
+#include "infer_stats.h"
 #include "triton/common/logging.h"
 
 // Global config group has 'name' of empty string.
@@ -101,6 +102,13 @@ MetricReporterConfig::ParseQuantiles(std::string options)
 //
 // MetricModelReporter
 //
+const std::map<FailureReason, std::string>
+    MetricModelReporter::failure_reasons_map = {
+        {FailureReason::REJECTED, "REJECTED"},
+        {FailureReason::CANCELED, "CANCELED"},
+        {FailureReason::BACKEND, "BACKEND"},
+        {FailureReason::OTHER, "OTHER"}};
+
 Status
 MetricModelReporter::Create(
     const ModelIdentifier& model_id, const int64_t model_version,
@@ -189,7 +197,6 @@ MetricModelReporter::InitializeCounters(
 {
   // Always setup these counters, regardless of config
   counter_families_["inf_success"] = &Metrics::FamilyInferenceSuccess();
-  counter_families_["inf_failure"] = &Metrics::FamilyInferenceFailure();
   counter_families_["inf_count"] = &Metrics::FamilyInferenceCount();
   counter_families_["inf_exec_count"] =
       &Metrics::FamilyInferenceExecutionCount();
@@ -227,6 +234,15 @@ MetricModelReporter::InitializeCounters(
       counters_[name] = CreateMetric<prometheus::Counter>(*family_ptr, labels);
     }
   }
+
+  // Initialize failure metrics with reasons
+  for (const auto& reason_pair : failure_reasons_map) {
+    std::map<std::string, std::string> extended_labels = labels;
+    extended_labels["reason"] = reason_pair.second;
+    counters_["inf_failure_" + reason_pair.second] =
+        CreateMetric<prometheus::Counter>(
+            Metrics::FamilyInferenceFailure(), extended_labels);
+  }
 }
 
 void

src/metric_model_reporter.h

@@ -94,6 +94,8 @@ class MetricModelReporter {
   // Lookup summary metric by name, and observe the value if it exists.
   void ObserveSummary(const std::string& name, double value);
 
+  static const std::map<FailureReason, std::string> failure_reasons_map;
+
  private:
   MetricModelReporter(
       const ModelIdentifier& model_id, const int64_t model_version,