Copying mlmonitoring directory in mlflow-ubuntu20.04-py39-cpu-inference environment

assets/inference/environments/mlflow-ubuntu20.04-py39-cpu-inference/context/Dockerfile

@@ -13,8 +13,8 @@ ENV AML_APP_ROOT="/var/mlflow_resources"
 ENV AZUREML_ENTRY_SCRIPT="mlflow_score_script.py"
 
 USER root
-# Copying of mlmonitoring will add once testing is completed.
-# COPY mlmonitoring /var/mlflow_resources/mlmonitoring
+
+COPY mlmonitoring /var/mlflow_resources/mlmonitoring
 
 # We'll copy the HF scripts as well to enable better handling for v2 packaging. This will not require changes to the
 # packages installed in the image, as the expectation is that these will all be brought along with the model.

assets/inference/environments/mlflow-ubuntu20.04-py39-cpu-inference/context/mlmonitoring/__init__.py

@@ -0,0 +1,7 @@
+"""For init."""
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+from .collector import Collector
+
+__all__ = ["Collector"]

assets/inference/environments/mlflow-ubuntu20.04-py39-cpu-inference/context/mlmonitoring/collector.py

@@ -0,0 +1,28 @@
+"""For collector."""
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+from typing import Callable, Any
+
+from .collector_json import JsonCollector
+from .context import CorrelationContext
+
+
+class Collector:
+    """For collector class."""
+
+    def __init__(
+            self,
+            *,
+            name: str,
+            on_error: Callable[[Exception], Any] = None
+    ):
+        """For init."""
+        self._impl = JsonCollector(name=name, on_error=on_error)
+
+    def collect(
+            self,
+            data,  # supported type: Union[pd.DataFrame]
+            correlation_context: CorrelationContext = None) -> CorrelationContext:
+        """For collect."""
+        return self._impl.collect(data, correlation_context)

assets/inference/environments/mlflow-ubuntu20.04-py39-cpu-inference/context/mlmonitoring/collector_base.py

@@ -0,0 +1,34 @@
+"""For Collector base."""
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+import logging
+
+from .init import init, is_sdk_ready
+from .config import get_config
+from .context import CorrelationContext, get_context_wrapper
+
+
+class CollectorBase:
+    """For CollectorBase."""
+
+    def __init__(
+            self,
+            model_version: str):
+        """For init."""
+        if not is_sdk_ready():
+            init(model_version)
+
+        self.logger = logging.getLogger("mdc.collector")
+        self.config = get_config()
+
+    def _response(
+            self,
+            context: CorrelationContext,
+            success: bool,
+            message: str) -> CorrelationContext:
+        """For response."""
+        if self.config.is_debug():
+            return get_context_wrapper(context, success, message)
+
+        return context

assets/inference/environments/mlflow-ubuntu20.04-py39-cpu-inference/context/mlmonitoring/collector_json.py

@@ -0,0 +1,128 @@
+"""For collector json."""
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+import sys
+import random
+from typing import Callable, Any
+
+from .payload import PandasFrameData
+from .payload.payload import build_payload
+from .queue import get_queue
+
+from .collector_base import CollectorBase
+from .context import CorrelationContext, get_context
+
+try:
+    import pandas as pd
+except ImportError:
+    pass
+
+
+def _build_log_data_by_type(data):
+    """For build log data by type."""
+    if 'pandas' in sys.modules and isinstance(data, pd.DataFrame):
+        return PandasFrameData(data)
+
+    raise TypeError("data type (%s) not supported, "
+                    "supported types: pandas.DataFrame"
+                    % type(data).__name__)
+
+
+def _raise_if_exception(e: Exception):
+    """For raise if exception."""
+    raise e
+
+
+class JsonCollector(CollectorBase):
+    """For JsonCollector."""
+
+    def __init__(
+            self,
+            *,
+            name: str,
+            on_error: Callable[[Exception], Any] = None
+    ):
+        """For init."""
+        super().__init__("default")
+        self.name = name
+        if on_error:
+            self.on_error = on_error
+        else:
+            self.on_error = _raise_if_exception
+
+        self._validate_mdc_config()
+
+    def _validate_mdc_config(self):
+        """For validate mdc config."""
+        if not self.name or len(self.name) <= 0:
+            # unexpected drop
+            msg = "collection name is not provided"
+            self.on_error(Exception(msg))
+            return False, msg
+
+        config = self.config
+        if config is None:
+            # unexpected drop
+            msg = "data collector is not initialized"
+            self.on_error(Exception(msg))
+            return False, msg
+
+        if not config.enabled():
+            # unexpected drop
+            msg = "custom logging is not enabled, drop the data"
+            self.on_error(Exception(msg))
+            return False, msg
+
+        if not config.collection_enabled(self.name):
+            # unexpected drop
+            msg = "collection {} is not enabled, drop the data".format(self.name)
+            self.on_error(Exception(msg))
+            return False, msg
+
+        return True, None
+
+    def collect(
+            self,
+            data,  # supported type: Union[pd.DataFrame]
+            correlation_context: CorrelationContext = None) -> CorrelationContext:
+        """For collect."""
+        if correlation_context is None:
+            correlation_context = get_context()
+
+        success, msg = self._validate_mdc_config()
+
+        if not success:
+            return self._response(correlation_context, False, msg)
+
+        config = self.config
+
+        percentage = config.collection_sample_rate_percentage(self.name)
+
+        if percentage < 100:
+            if percentage <= random.random() * 100.0:
+                # expected drop
+                self.logger.debug("sampling not hit, drop the data")
+                # TBD: send empty message to mdc to collect metrics of dropped messages?
+                return self._response(correlation_context, False, "dropped_sampling")
+
+        try:
+            # build payload and put into payload queue
+            log_data = _build_log_data_by_type(data)
+        except TypeError as e:
+            # unexpected drop
+            self.on_error(e)
+            return self._response(correlation_context, False, e.args[0])
+
+        payload = build_payload(
+            self.name,
+            data=log_data,
+            model_version=config.model_version(),
+            context=correlation_context)
+
+        success, msg = get_queue().enqueue(payload)
+
+        if not success:
+            # unexpected drop
+            self.on_error(Exception(msg))
+        return self._response(correlation_context, success, msg)

assets/inference/environments/mlflow-ubuntu20.04-py39-cpu-inference/context/mlmonitoring/common/__init__.py

@@ -0,0 +1,5 @@
+"""For init."""
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+__version__ = "0.1.0"

assets/inference/environments/mlflow-ubuntu20.04-py39-cpu-inference/context/mlmonitoring/config/__init__.py

@@ -0,0 +1,7 @@
+"""For init."""
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+from .config import init_config, teardown_config, is_debug, get_config
+
+__all__ = ["init_config", "teardown_config", "is_debug", "get_config"]