MINIFICPP-2892 Tensor extensions based on sonos/tract

CONTROLLERS.md

@@ -33,6 +33,7 @@ limitations under the License.
 - [RocksDbStateStorage](#RocksDbStateStorage)
 - [SmbConnectionControllerService](#SmbConnectionControllerService)
 - [SSLContextService](#SSLContextService)
+- [TractModelService](#TractModelService)
 - [UpdatePolicyControllerService](#UpdatePolicyControllerService)
 - [VolatileMapStateStorage](#VolatileMapStateStorage)
 - [XMLReader](#XMLReader)
@@ -359,6 +360,22 @@ In the list below, the names of required properties appear in bold. Any other pr
 | Use System Cert Store      | false                 | true<br/>false                                                                                                                                           | Whether to use the certificates in the OS's certificate store                                                                            |
 
 
+## TractModelService
+
+### Description
+
+Provides a shared, CPU-optimized neural network for inference. Supports ONNX (`.onnx`) and NNEF (directory or tarball) models; the format can be auto-detected from the file extension or set explicitly.
+
+### Properties
+
+In the list below, the names of required properties appear in bold. Any other properties (not in bold) are considered optional. The table also indicates any default values, and whether a property supports the NiFi Expression Language.
+
+| Name                | Default Value | Allowable Values       | Description                                                                                                                                                                                                                                                                                                                                                 |
+|---------------------|---------------|------------------------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| **Model File Path** |               |                        | Absolute path to the model on the edge device. For ONNX this is a `.onnx` file; for NNEF this is a `.nnef.tgz` archive, a `.nnef` tarball, or the root directory of an unpacked NNEF model. The model is loaded, parsed, and compiled for the host CPU once when the controller service is enabled; subsequent inference calls reuse the compiled runnable. |
+| **Model format**    | Auto          | Auto<br/>Onnx<br/>Nnef | Format of the file/directory referenced by 'Model File Path'. 'Auto' picks Onnx when the path ends in `.onnx` and Nnef when it ends in `.nnef`, `.nnef.tgz`, `.nnef.tar`, `.nnef.tar.gz`, or points at a directory. Set explicitly when the path uses a non-standard extension.                                                                             |
+
+
 ## UpdatePolicyControllerService
 
 ### Description

behave_framework/pyproject.toml

@@ -9,7 +9,8 @@ dependencies = [
     "PyYAML==6.0.3",
     "humanfriendly==10.0",
     "cryptography==50.0.0",
-    "pyjks==20.0.0"
+    "pyjks==20.0.0",
+    "certifi>=2026.7.22"
 ]
 
 [tool.setuptools]

minifi_rust/extensions/minifi_rs_playground/src/processors/asciify_german.rs

@@ -21,7 +21,7 @@ use crate::processors::asciify_german::relationships::FAILURE;
 use minifi_native::macros::ComponentIdentifier;
 use minifi_native::{
     FlowFileStreamTransform, GetProperty, InputStream, Logger, MinifiError, OutputStream,
-    ProcessError, RouteErrorExt, Schedule, TransformStreamResult,
+    ProcessError, Schedule, TransformStreamResult,
 };
 
 mod relationships;
@@ -56,8 +56,9 @@ impl FlowFileStreamTransform for AsciifyGerman {
                 0xC3 => {
                     let mut next = [0u8; 1];
                     if input_stream.read(&mut next)? == 0 {
-                        Err(MinifiError::custom("Truncated multi-byte sequence at EOF"))
-                            .route_err_to_failure()?
+                        return Err(ProcessError::route_to_failure(
+                            "Truncated multi-byte sequence at EOF",
+                        ));
                     }
                     match next[0] {
                         0xA4 => output_stream.write_all(b"ae")?, // ä

minifi_rust/extensions/minifi_rs_playground/src/processors/get_file.rs

@@ -27,7 +27,7 @@ use crate::processors::get_file::properties::{
 use minifi_native::macros::ComponentIdentifier;
 use minifi_native::{
     GetProperty, IoState, Logger, MinifiError, OnTriggerResult, ProcessContext, ProcessError,
-    ProcessSession, Schedule, Trigger, debug, info, trace, warn,
+    ProcessSession, Schedule, Trigger, debug, trace, warn,
 };
 use std::collections::VecDeque;
 use std::error;
@@ -179,7 +179,7 @@ impl GetFileRs {
         logger: &L,
         path: &Path,
     ) -> Result<(), MinifiError> {
-        info!(logger, "GetFile process {:?}", path);
+        trace!(logger, "GetFile process {:?}", path);
         let mut ff = session
             .create()
             .expect("Successful FlowFile creation is expected");

minifi_rust/extensions/minifi_tensor/Cargo.toml

@@ -0,0 +1,25 @@
+[package]
+name = "minifi_tensor"
+version = "0.1.0"
+edition = "2024"
+
+[lib]
+crate-type = ["cdylib"]
+
+[features]
+low-level-processors = []
+
+[dependencies]
+minifi_native = { path = "../../minifi_native" }
+strum = "0.28.0"
+strum_macros = "0.28.0"
+image = "0.25.10"
+imageproc = "0.27.0"
+tract = "0.23.4"
+ndarray = "0.17.2"
+serde = { version = "1.0.229", features = ["derive"] }
+serde_json = "1.0.151"
+
+[dev-dependencies]
+minifi_native = { path = "../../minifi_native", features = ["test-utils"] }
+tempfile = "3.27.0"

minifi_rust/extensions/minifi_tensor/features/basic.feature

@@ -0,0 +1,33 @@
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements.  See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License.  You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+@SUPPORTS_WINDOWS
+Feature: minifi_tensor extension loads
+
+  Scenario: All processors and the TractModelService register cleanly
+    Given log property "logger.org::apache::nifi::minifi::core::extension::ExtensionManager" is set to "TRACE,stderr"
+    And log property "logger.org::apache::nifi::minifi::core::ClassLoader" is set to "TRACE,stderr"
+
+    When the MiNiFi instance starts up
+
+    Then the Minifi logs contain the following message: "Registering class 'ImageToTensor' at '/minifi_tensor'" in less than 10 seconds
+    And the Minifi logs contain the following message: "Registering class 'InvokeTractModel' at '/minifi_tensor'" in less than 1 seconds
+    And the Minifi logs contain the following message: "Registering class 'FilterBoundingBoxes' at '/minifi_tensor'" in less than 1 seconds
+    And the Minifi logs contain the following message: "Registering class 'ClassifyOutput' at '/minifi_tensor'" in less than 1 seconds
+    And the Minifi logs contain the following message: "Registering class 'TractModelService' at '/minifi_tensor'" in less than 1 seconds
+    And the Minifi logs contain the following message: "Registering class 'ClassifyImage' at '/minifi_tensor'" in less than 1 seconds
+    And the Minifi logs contain the following message: "Registering class 'DetectObject' at '/minifi_tensor'" in less than 1 seconds
+    And the Minifi logs do not contain errors
+    And the Minifi logs do not contain warnings

minifi_rust/extensions/minifi_tensor/features/classification.feature

@@ -0,0 +1,128 @@
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements.  See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You under the Apache License, Version 2.0
+# (the "License"); you may not use this file except in compliance with
+# the License.  You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+@SUPPORTS_WINDOWS
+Feature: Image classification with MobileNetV2
+
+  # Based on https://github.com/sonos/tract/tree/main/examples/onnx-mobilenet-v2
+  Scenario: Grace Hopper image is classified using MobileNetV2 ONNX and imagenet labels (ImageToTensor + InvokeTract + ClassifyOutput)
+    Given a host resource file "grace_hopper.jpg" is copied to the "/tmp/input/grace_hopper.jpg" path in the MiNiFi container
+    And a host resource file "mobilenetv2-7.onnx" is copied to the "/tmp/models/mobilenetv2-7.onnx" path in the MiNiFi container
+    And a host resource file "imagenet_slim_labels.txt" is copied to the "/tmp/models/imagenet_slim_labels.txt" path in the MiNiFi container
+
+    And a TractModelService controller service named "MobileNet" is set up and the "Model File Path" property set to "/tmp/models/mobilenetv2-7.onnx"
+    And the "Model format" property of the MobileNet controller service is set to "Onnx"
+
+    And a GetFile processor with the "Input Directory" property set to "/tmp/input"
+    And the "Keep Source File" property of the GetFile processor is set to "false"
+
+    And an ImageToTensor processor with the "Target width" property set to "224"
+    And the "Target height" property of the ImageToTensor processor is set to "224"
+    And the "Resize filter" property of the ImageToTensor processor is set to "Bilinear"
+    And the "Resize mode" property of the ImageToTensor processor is set to "Stretch"
+    And the "Color format" property of the ImageToTensor processor is set to "RGB"
+    And the "Tensor shape format" property of the ImageToTensor processor is set to "CHW"
+    And the "Mean" property of the ImageToTensor processor is set to "0.485, 0.456, 0.406"
+    And the "Standard Deviation" property of the ImageToTensor processor is set to "0.229, 0.224, 0.225"
+    And the "Pixel divisor" property of the ImageToTensor processor is set to "255"
+
+    And an InvokeTractModel processor with the "Tract model service" property set to "MobileNet"
+
+    And a ClassifyOutput processor with the "Top K" property set to "3"
+    And the "Score activation" property of the ClassifyOutput processor is set to "Softmax"
+    And the "Labels file path" property of the ClassifyOutput processor is set to "/tmp/models/imagenet_slim_labels.txt"
+    # ImageNet slim labels start with a dummy entry on line 0
+    And the "Label index offset" property of the ClassifyOutput processor is set to "1"
+    And the "Confidence Threshold" property of the ClassifyOutput processor is set to "0.0"
+
+    And a PutFile processor with the "Directory" property set to "/tmp/output"
+
+    And a LogAttribute processor with the "FlowFiles To Log" property set to "0"
+    And LogAttribute is EVENT_DRIVEN
+
+    And the "success" relationship of the GetFile processor is connected to the ImageToTensor
+    And the "success" relationship of the ImageToTensor processor is connected to the InvokeTractModel
+    And the "success" relationship of the InvokeTractModel processor is connected to the ClassifyOutput
+    And the "success" relationship of the ClassifyOutput processor is connected to the LogAttribute
+    And the "success" relationship of the LogAttribute processor is connected to the PutFile
+    And ImageToTensor's failure relationship is auto-terminated
+    And InvokeTractModel's failure relationship is auto-terminated
+    And ClassifyOutput's failure relationship is auto-terminated
+    And PutFile's success relationship is auto-terminated
+    And PutFile's failure relationship is auto-terminated
+
+    When the MiNiFi instance starts up
+
+    # Grace Hopper wears a US Navy uniform; MobileNetV2 on ImageNet consistently
+    # picks "military uniform" (occasionally "suit", "Windsor tie", or
+    # "bulletproof vest" as close runner-ups). Match any of those to keep the
+    # test resilient to small numeric differences across tract versions.
+    Then the Minifi logs match the following regex: "key:class.top1.name value:(military uniform|bulletproof vest|suit|Windsor tie)" in less than 60 seconds
+    And the Minifi logs match the following regex: "key:class.count value:[1-3]" in less than 1 seconds
+    And the Minifi logs contain the following message: "key:mime.type value:application/json" in less than 1 seconds
+    And at least one file in "/tmp/output" content match the following regex: "\"class_name\":\"(military uniform|bulletproof vest|suit|Windsor tie)\"" in less than 30 seconds
+    And the Minifi logs do not contain errors
+
+  # Based on https://github.com/sonos/tract/tree/main/examples/onnx-mobilenet-v2
+  Scenario: Grace Hopper image is classified using MobileNetV2 ONNX and imagenet labels (ClassifyImage)
+    Given a host resource file "grace_hopper.jpg" is copied to the "/tmp/input/grace_hopper.jpg" path in the MiNiFi container
+    And a host resource file "mobilenetv2-7.onnx" is copied to the "/tmp/models/mobilenetv2-7.onnx" path in the MiNiFi container
+    And a host resource file "imagenet_slim_labels.txt" is copied to the "/tmp/models/imagenet_slim_labels.txt" path in the MiNiFi container
+
+    And a TractModelService controller service named "MobileNet" is set up and the "Model File Path" property set to "/tmp/models/mobilenetv2-7.onnx"
+    And the "Model format" property of the MobileNet controller service is set to "Onnx"
+
+    And a GetFile processor with the "Input Directory" property set to "/tmp/input"
+    And the "Keep Source File" property of the GetFile processor is set to "false"
+
+    And a ClassifyImage processor with the "Target width" property set to "224"
+    And the "Target height" property of the ClassifyImage processor is set to "224"
+    And the "Resize filter" property of the ClassifyImage processor is set to "Bilinear"
+    And the "Resize mode" property of the ClassifyImage processor is set to "Stretch"
+    And the "Color format" property of the ClassifyImage processor is set to "RGB"
+    And the "Tensor shape format" property of the ClassifyImage processor is set to "CHW"
+    And the "Mean" property of the ClassifyImage processor is set to "0.485, 0.456, 0.406"
+    And the "Standard Deviation" property of the ClassifyImage processor is set to "0.229, 0.224, 0.225"
+    And the "Pixel divisor" property of the ClassifyImage processor is set to "255"
+    And the "Tract model service" property of the ClassifyImage processor is set to "MobileNet"
+    And the "Top K" property of the ClassifyImage processor is set to "3"
+    And the "Score activation" property of the ClassifyImage processor is set to "Softmax"
+    And the "Labels file path" property of the ClassifyImage processor is set to "/tmp/models/imagenet_slim_labels.txt"
+    And the "Label index offset" property of the ClassifyImage processor is set to "1"
+    And the "Confidence Threshold" property of the ClassifyImage processor is set to "0.0"
+
+    And a PutFile processor with the "Directory" property set to "/tmp/output"
+
+    And a LogAttribute processor with the "FlowFiles To Log" property set to "0"
+    And LogAttribute is EVENT_DRIVEN
+
+    And the "success" relationship of the GetFile processor is connected to the ClassifyImage
+    And the "success" relationship of the ClassifyImage processor is connected to the LogAttribute
+    And the "success" relationship of the LogAttribute processor is connected to the PutFile
+    And ClassifyImage's failure relationship is auto-terminated
+    And PutFile's success relationship is auto-terminated
+    And PutFile's failure relationship is auto-terminated
+
+    When the MiNiFi instance starts up
+
+    # Grace Hopper wears a US Navy uniform; MobileNetV2 on ImageNet consistently
+    # picks "military uniform" (occasionally "suit", "Windsor tie", or
+    # "bulletproof vest" as close runner-ups). Match any of those to keep the
+    # test resilient to small numeric differences across tract versions.
+    Then the Minifi logs match the following regex: "key:class.top1.name value:(military uniform|bulletproof vest|suit|Windsor tie)" in less than 60 seconds
+    And the Minifi logs match the following regex: "key:class.count value:[1-3]" in less than 1 seconds
+    And the Minifi logs contain the following message: "key:mime.type value:application/json" in less than 1 seconds
+    And at least one file in "/tmp/output" content match the following regex: "\"class_name\":\"(military uniform|bulletproof vest|suit|Windsor tie)\"" in less than 30 seconds
+    And the Minifi logs do not contain errors