Merge pull request #18716 from vespa-engine/havardpe/avoid-crash-on-runtime-onnx-errors

avoid crash on run-time onnx errors

4 files changed, 20 insertions, 10 deletions

diff --git a/searchlib/src/tests/features/onnx_feature/fragile.onnx b/searchlib/src/tests/features/onnx_feature/fragile.onnx
index 2a05500e95b..dc650154f83 100644
--- a/searchlib/src/tests/features/onnx_feature/fragile.onnx
+++ b/searchlib/src/tests/features/onnx_feature/fragile.onnx
@@ -1,5 +1,5 @@
 
-fragile.py:b
+fragile.py:]
 
 in1
 in2out"AddfragileZ
@@ -9,7 +9,8 @@ fragile.py:b
 Z
 in2
 

	
-batchb
-out
-

	
-batchB
\ No newline at end of file
+batchb
+out
+
+

+B
\ No newline at end of file
diff --git a/searchlib/src/tests/features/onnx_feature/fragile.py b/searchlib/src/tests/features/onnx_feature/fragile.py
index e4eaf168e14..fe5851f5a63 100755
--- a/searchlib/src/tests/features/onnx_feature/fragile.py
+++ b/searchlib/src/tests/features/onnx_feature/fragile.py
@@ -6,7 +6,7 @@ from onnx import helper, TensorProto
 INPUT1 = helper.make_tensor_value_info('in1', TensorProto.FLOAT, [2])
 INPUT2 = helper.make_tensor_value_info('in2', TensorProto.FLOAT, ['batch'])
 
-OUTPUT = helper.make_tensor_value_info('out', TensorProto.FLOAT, ['batch'])
+OUTPUT = helper.make_tensor_value_info('out', TensorProto.FLOAT, [2])
 
 nodes = [
     helper.make_node(
diff --git a/searchlib/src/tests/features/onnx_feature/onnx_feature_test.cpp b/searchlib/src/tests/features/onnx_feature/onnx_feature_test.cpp
index c07ebc48604..7e80d9fc335 100644
--- a/searchlib/src/tests/features/onnx_feature/onnx_feature_test.cpp
+++ b/searchlib/src/tests/features/onnx_feature/onnx_feature_test.cpp
@@ -147,20 +147,22 @@ TEST_F(OnnxFeatureTest, input_features_and_output_names_can_be_specified) {
 TEST_F(OnnxFeatureTest, fragile_model_can_be_evaluated) {
     add_expr("in1", "tensor<float>(x[2]):[docid,5]");
     add_expr("in2", "tensor<float>(x[2]):[docid,10]");
-    add_onnx(OnnxModel("fragile", fragile_model));
+    add_onnx(OnnxModel("fragile", fragile_model).dry_run_on_setup(true));
     EXPECT_TRUE(try_compile(onnx_feature("fragile")));
     EXPECT_EQ(get(1), TensorSpec::from_expr("tensor<float>(d0[2]):[2,15]"));
     EXPECT_EQ(get(3), TensorSpec::from_expr("tensor<float>(d0[2]):[6,15]"));
 }
 
-TEST_F(OnnxFeatureTest, runtime_broken_model_can_be_set_up_without_dry_run) {
+TEST_F(OnnxFeatureTest, broken_model_evaluates_to_all_zeros) {
     add_expr("in1", "tensor<float>(x[2]):[docid,5]");
     add_expr("in2", "tensor<float>(x[3]):[docid,10,31515]");
     add_onnx(OnnxModel("fragile", fragile_model).dry_run_on_setup(false));
     EXPECT_TRUE(try_compile(onnx_feature("fragile")));
+    EXPECT_EQ(get(1), TensorSpec::from_expr("tensor<float>(d0[2]):[0,0]"));
+    EXPECT_EQ(get(3), TensorSpec::from_expr("tensor<float>(d0[2]):[0,0]"));
 }
 
-TEST_F(OnnxFeatureTest, runtime_broken_model_fails_with_dry_run) {
+TEST_F(OnnxFeatureTest, broken_model_fails_with_dry_run) {
     add_expr("in1", "tensor<float>(x[2]):[docid,5]");
     add_expr("in2", "tensor<float>(x[3]):[docid,10,31515]");
     add_onnx(OnnxModel("fragile", fragile_model).dry_run_on_setup(true));
diff --git a/searchlib/src/vespa/searchlib/features/onnx_feature.cpp b/searchlib/src/vespa/searchlib/features/onnx_feature.cpp
index 0af03f3aa86..d4756a2f8a7 100644
--- a/searchlib/src/vespa/searchlib/features/onnx_feature.cpp
+++ b/searchlib/src/vespa/searchlib/features/onnx_feature.cpp
@@ -89,7 +89,12 @@ public:
         for (size_t i = 0; i < _eval_context.num_params(); ++i) {
             _eval_context.bind_param(i, inputs().get_object(i).get());
         }
-        _eval_context.eval();
+        try {
+            _eval_context.eval();
+        } catch (const Ort::Exception &ex) {
+            LOG(warning, "onnx model evaluation failed: %s", ex.what());
+            _eval_context.clear_results();
+        }
     }
 };
 
@@ -162,6 +167,8 @@ OnnxBlueprint::setup(const IIndexEnvironment &env,
         if (!error_msg.empty()) {
             return fail("onnx model dry-run failed: %s", error_msg.c_str());
         }
+    } else {
+        LOG(warning, "dry-run disabled for onnx model '%s'", model_cfg->name().c_str());
     }
     return true;
 }