Datadog tracing

.gitignore

@@ -188,7 +188,7 @@ testing/scripts/tensorflow
 testing/scripts/run.log
 testing/scripts/my-model/
 wrappers/s2i/python/_python/
-wrappers/s2i/python-conda/_python/
+incubating/wrappers/s2i/java-jni/_python/
 
 seldon-controller/go
 

.pre-commit-config.yaml

@@ -9,4 +9,4 @@ repos:
     rev: stable
     hooks:
       - id: black
-        args: ['python/', 'testing/', 'operator/helm', 'operator/seldon-operator/hack', '--exclude', '(testing/scripts/proto|seldon_core/proto/|.eggs)']
+        args: ['python/', 'testing/', '--exclude', '(testing/scripts/proto|seldon_core/proto/|.eggs)']

Makefile

@@ -28,7 +28,7 @@ run_core_builder_in_host:
 			-v /var/run/docker.sock:/var/run/docker.sock \
 			-v $${HOME}/.m2:/root/.m2 \
 			-v $(SELDON_CORE_LOCAL_DIR):/work \
-			seldonio/core-builder:0.17 bash
+			seldonio/core-builder:0.18 bash
 
 
 run_core_builder_in_minikube:
@@ -37,7 +37,7 @@ run_core_builder_in_minikube:
 			-v /var/run/docker.sock:/var/run/docker.sock \
 			-v /home/docker/.m2:/root/.m2 \
 			-v $(SELDON_CORE_VM_DIR):/work \
-			seldonio/core-builder:0.17 bash
+			seldonio/core-builder:0.18 bash
 
 show_paths:
 	@echo "local: $(SELDON_CORE_LOCAL_DIR)"

components/alibi-detect-server/Dockerfile

@@ -1,7 +1,7 @@
 FROM registry.access.redhat.com/ubi8/python-36
 LABEL name="Seldon Alibi Detect Server" \
       vendor="Seldon Technologies" \
-      version="1.2.3-dev" \
+      version="1.3.0-dev" \
       release="1" \
       summary="Alibi Detect Server for Seldon Core" \
       description="The Alibi Detect Server provides outlier, drift and adversarial detection services for Seldon Core"

components/alibi-explain-server/Dockerfile

@@ -1,7 +1,7 @@
 FROM registry.access.redhat.com/ubi8/python-36
 LABEL name="Seldon Alibi Wrapper" \
       vendor="Seldon Technologies" \
-      version="1.2.3-dev" \
+      version="1.3.0-dev" \
       release="1" \
       summary="Alibi Explainer Wrapper for Seldon Core" \
       description="Allows Seldon Core inference models to run with a black box model explanation model from the Alibi:Explain project"

components/alibi-explain-server/alibiexplainer/__main__.py

@@ -33,6 +33,7 @@
 EXPLAINER_FILENAME = "explainer.dill"
 KERAS_MODEL = "model.h5"
 
+
 def main():
     args, extra = parse_args(sys.argv[1:])
     # Pretrained Alibi explainer
@@ -60,12 +61,11 @@ def main():
         alibi_model,
         Protocol(args.protocol),
         args.tf_data_type,
-        keras_model
+        keras_model,
     )
     explainer.load()
     ExplainerServer(args.http_port).start(explainer)
 
+
 if __name__ == "__main__":
     main()
-
-

components/routers/case_study/models/xgb_model/requirements.txt

@@ -1,4 +1,2 @@
-scikit-learn==0.20.1
-numpy==1.15.1
-scipy==1.1.0
-xgboost==0.81
+numpy >= 1.8.2
+xgboost==1.2.0

components/seldon-request-logger/Dockerfile

@@ -1,7 +1,7 @@
 FROM registry.access.redhat.com/ubi8/python-36
 LABEL name="Seldon Request Logger" \
       vendor="Seldon Technologies" \
-      version="1.2.3-dev" \
+      version="1.3.0-dev" \
       release="1" \
       summary="The payload logger for Seldon Core" \
       description="The Seldon Payload Logger allows request and response payloads from a Seldon Core inference graph to be processed and sent to an ELK endpoint"

components/storage-initializer/Dockerfile

@@ -1,7 +1,7 @@
 FROM registry.access.redhat.com/ubi8/python-36
 LABEL name="Storage Initializer" \
       vendor="Seldon Technologies" \
-      version="1.2.3-dev" \
+      version="1.3.0-dev" \
       release="1" \
       summary="Storage Initializer for Seldon Core" \
       description="Allows Seldon Core to download artifacts from cloud and local storage to a local volume"

core-builder/Dockerfile

@@ -40,6 +40,7 @@ RUN wget https://storage.googleapis.com/kubernetes-release/release/v1.16.2/bin/l
 RUN \
     apt-get update -y && \
     apt-get install -y vim && \
+    apt-get install -y jq && \
     apt-get install -y build-essential && \
     apt-get remove -y --auto-remove && apt-get clean -y && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
 

core-builder/Makefile

@@ -1,5 +1,5 @@
 DOCKER_IMAGE_NAME=seldonio/core-builder
-DOCKER_IMAGE_VERSION=0.17
+DOCKER_IMAGE_VERSION=0.18
 
 build_docker_image:
 	cp ../testing/scripts/dev_requirements.txt .

doc/requirements_docs.txt

@@ -8,3 +8,4 @@ nbsphinx>=0.4.2
 nbsphinx-link>=1.2.0
 ipykernel>=5.1.0
 ipython>=7.2.0
+pygments>=2.4.1,<3

doc/source/analytics/routers.md

@@ -9,7 +9,15 @@ Currently we provide two reference implementations of routers in Python. Both ar
 * [Thompson Sampling](https://github.com/SeldonIO/seldon-core/tree/master/components/routers/thompson-sampling)
 
 ## Implementing custom routers
-A router component must implement a ```Route``` method which will return one of the children that the router component is connected to for routing an incoming request. Optionally a ```SendFeedback``` method can be implemented to provide a mechanism for informing the router on the quality of its decisions. This would be used in adaptive routers such as multi-armed bandits, refer to the [epsilon-greedy](https://github.com/SeldonIO/seldon-core/tree/master/components/routers/epsilon-greedy) example for more detail.
+A router component must implement a ```Route``` method which will return one of the children that the router component is connected to for routing an incoming request. The options for the return value for a custom router at present are
+
+ * -1 : Route to all children
+ * -2 : Route to no children and return the current request as the response
+ * N >= 0 : Route to child N
+
+The response for REST calls should be returned as a SeldonMessage with the payload containing the route value or a JSON array containing a single integer.
+
+Optionally a ```SendFeedback``` method can be implemented to provide a mechanism for informing the router on the quality of its decisions. This would be used in adaptive routers such as multi-armed bandits, refer to the [epsilon-greedy](https://github.com/SeldonIO/seldon-core/tree/master/components/routers/epsilon-greedy) example for more detail.
 
 As an example, consider writing a custom A/B/C... testing component with a user-specified number of children and routing probabilities (two-model routing is already supported in Seldon Core: [RANDOM_ABTEST](../reference/seldon-deployment.md#proto-buffer-definition)). In this scenario because the routing logic is static there is no need to implement ```SendFeedback``` as we will not be dynamically changing the routing by providing feedback for its routing choices. On the other hand, an adaptive router whose routing is required to change dynamically by providing feedback will need to implement the ```SendFeedback``` method.
 

doc/source/examples/notebooks.rst

@@ -26,6 +26,7 @@ Python Language Wrapper Examples
    Sagemaker SKLearn Example <sagemaker_sklearn>   
    TFserving MNIST <tfserving_mnist>
    Statsmodels Holt-Winter's time-series model <statsmodels>
+   Runtime Metrics & Tags <runtime_metrics_tags>
 
 Specialised Framework Examples
 -----

doc/source/examples/runtime_metrics_tags.nblink

@@ -0,0 +1,3 @@
+{
+  "path": "../../../examples/models/runtime_metrics_tags/runtime_metrics_tags.ipynb"
+}

doc/source/graph/distributed-tracing-dd.md

@@ -0,0 +1,83 @@
+# Distributed Tracing with Datadog
+
+You can use Open Tracing to trace your API calls to Seldon Core. By default Jaeger is supported ([see here](distributed-tracing.md)), but Datadog can also be used. 
+
+Datadog is only supported in the Executor and the Python wrapper at this time.
+
+## Install Datadog
+
+You will need to install the Datadog Agent on your Kubernetes cluster. Follow their [documentation](https://docs.datadoghq.com/agent/kubernetes/?tab=helm). Ensure that APM is enabled in the deployment, see [here](https://docs.datadoghq.com/agent/kubernetes/apm/?tab=helm).
+
+## Configuration
+
+You will need to annotate your Seldon Deployment resource with environment variables to make tracing active and set the appropriate Datadog configuration variables.
+
+  * For each Seldon component you run (e.g., model transformer etc.) you will need to add environment variables to the container section.
+
+
+### Python Wrapper Configuration
+
+Add an environment variable: `TRACING` with value `1` to activate tracing.
+
+To ensure that Datadog tracing is used, set the environment variable `DD_ENABLED` to `1`
+
+For a complete list of available environment variables, see the [Datadogs python documentation](https://docs.datadoghq.com/tracing/setup/python/#configuration) for the model wrapper, and [Datadogs Go documentation](https://docs.datadoghq.com/tracing/setup/go/) for the executor, but the relevant ones are below:
+* `DD_AGENT_HOST=<host>` (defaults to `localhost`)
+* `DATADOG_TRACE_AGENT_PORT=<port>` (defaults to `8126`)
+* `DD_SERVICE=<svc>` (will default to either `executor`, or the name of your Python class)
+* `DD_TAGS=<key:value,key2:value2`
+* `DD_SAMPLE_RATE:1` (defaults to 1, keeping all traces)
+    * _NOTE: This is a non-standard environment variable, meaning its sepcific to Seldon._
+
+
+
+An example is show below:
+
+```yaml
+apiVersion: machinelearning.seldon.io/v1
+kind: SeldonDeployment
+metadata:
+  name: dd-tracing-example
+  namespace: seldon
+spec:
+  name: dd-tracing-example
+  predictors:
+  - componentSpecs:
+    - spec:
+        containers:
+        - env:
+          - name: TRACING
+            value: '1'
+          - name: DD_AGENT_HOST
+            valueFrom:
+              fieldRef:
+                fieldPath: status.hostIP
+          - name: DATADOG_TRACE_AGENT_PORT
+            value: '8126'
+          - name: DD_SAMPLE_RATE
+            value: 0.75 # Keep 75% of traces
+          image: seldonio/mock_classifier_rest:1.3
+          name: model1
+        terminationGracePeriodSeconds: 1
+    graph:
+      children: []
+      endpoint:
+        type: REST
+      name: model1
+      type: MODEL
+    name: tracing
+    replicas: 1
+    svcOrchSpec:
+      env:
+      - name: TRACING
+        value: '1'
+      - name: DD_AGENT_HOST
+        valueFrom:
+          fieldRef:
+            fieldPath: status.hostIP
+      - name: DATADOG_TRACE_AGENT_PORT
+        value: '8126'
+      - name: DD_SAMPLE_RATE
+        value: 0.9 # Keep 90% of traces
+```
+

doc/source/graph/distributed-tracing.md

@@ -2,6 +2,8 @@
 
 You can use Open Tracing to trace your API calls to Seldon Core. By default we support Jaeger for Distributed Tracing, which will allow you to obtain insights on latency and performance across each microservice-hop in your Seldon deployment.
 
+Datadog tracing is also supported, see [Distributed Tracing with Datadog](distributed-tracing-dd.md)
+
 ## Install Jaeger
 
 You will need to install Jaeger on your Kubernetes cluster. Follow their [documentation](https://www.jaegertracing.io/docs/1.18/operator/)

doc/source/index.rst

@@ -86,6 +86,7 @@ Documentation Index
    :caption: Incubating Projects
 
    Java Language Wrapper <java/README.md>
+   Java (JNI) Language Wrapper [ALPHA] <java-jni/README.md>
    R Language Wrapper [ALPHA] <R/README.md>
    NodeJS Language Wrapper [ALPHA] <nodejs/README.md>
    Go Language Wrapper [ALPHA] <go/go_wrapper_link.rst>
@@ -156,6 +157,7 @@ Documentation Index
    Seldon Deployment CRD <reference/seldon-deployment.md>
    Service Orchestrator <graph/svcorch.md>
    Kubeflow <analytics/kubeflow.md>
+   Concepts <reference/concepts.md>
 
 .. toctree::
    :maxdepth: 1