Azure Metrics Advisor client library for Python

Metrics Advisor is a scalable real-time time series monitoring, alerting, and root cause analysis platform. Use Metrics Advisor to:

• Analyze multi-dimensional data from multiple data sources
• Identify and correlate anomalies
• Configure and fine-tune the anomaly detection model used on your data
• Diagnose anomalies and help with root cause analysis

Source code | Package (Pypi) | API reference documentation | Product documentation | Samples

Getting started

Install the package

Install the Azure Metrics Advisor client library for Python with pip:

pip install azure-ai-metricsadvisor --pre

Prerequisites

• Python 2.7, or 3.6 or later is required to use this package.
• You need an Azure subscription, and a Metrics Advisor serivce to use this package.

Authenticate the client

You will need two keys to authenticate the client:

1. The subscription key to your Metrics Advisor resource. You can find this in the Keys and Endpoint section of your resource in the Azure portal.
2. The API key for your Metrics Advisor instance. You can find this in the web portal for Metrics Advisor, in API keys on the left navigation menu.

We can use the keys to create a new MetricsAdvisorClient or MetricsAdvisorAdministrationClient.

import os
from azure.ai.metricsadvisor import (
    MetricsAdvisorKeyCredential,
    MetricsAdvisorClient,
    MetricsAdvisorAdministrationClient,
)

service_endpoint = os.getenv("ENDPOINT")
subscription_key = os.getenv("SUBSCRIPTION_KEY")
api_key = os.getenv("API_KEY")

client = MetricsAdvisorClient(service_endpoint,
                            MetricsAdvisorKeyCredential(subscription_key, api_key))

admin_client = MetricsAdvisorAdministrationClient(service_endpoint,
                            MetricsAdvisorKeyCredential(subscription_key, api_key))

Key concepts

MetricsAdvisorClient

MetricsAdvisorClient helps with:

• listing incidents
• listing root causes of incidents
• retrieving original time series data and time series data enriched by the service.
• listing alerts
• adding feedback to tune your model

MetricsAdvisorAdministrationClient

MetricsAdvisorAdministrationClient allows you to

• manage data feeds
• manage anomaly detection configurations
• manage anomaly alerting configurations
• manage hooks

DataFeed

A DataFeed is what Metrics Advisor ingests from your data source, such as Cosmos DB or a SQL server. A data feed contains rows of:

• timestamps
• zero or more dimensions
• one or more measures

Metric

A DataFeedMetric is a quantifiable measure that is used to monitor and assess the status of a specific business process. It can be a combination of multiple time series values divided into dimensions. For example a web health metric might contain dimensions for user count and the en-us market.

AnomalyDetectionConfiguration

AnomalyDetectionConfiguration is required for every time series, and determines whether a point in the time series is an anomaly.

Anomaly & Incident

After a detection configuration is applied to metrics, AnomalyIncidents are generated whenever any series within it has an DataPointAnomaly.

Alert

You can configure which anomalies should trigger an AnomalyAlert. You can set multiple alerts with different settings. For example, you could create an alert for anomalies with lower business impact, and another for more important alerts.

Notification Hook

Metrics Advisor lets you create and subscribe to real-time alerts. These alerts are sent over the internet, using a notification hook like EmailNotificationHook or WebNotificationHook.

Examples

• Add a data feed from a sample or data source
• Check ingestion status
• Configure anomaly detection configuration
• Configure alert configuration
• Query anomaly detection results
• Query incidents
• Query root causes
• Add hooks for receiving anomaly alerts

Add a data feed from a sample or data source

Metrics Advisor supports connecting different types of data sources. Here is a sample to ingest data from SQL Server.

import os
import datetime
from azure.ai.metricsadvisor import MetricsAdvisorKeyCredential, MetricsAdvisorAdministrationClient
from azure.ai.metricsadvisor.models import (
        SqlServerDataFeedSource,
        DataFeedSchema,
        DataFeedMetric,
        DataFeedDimension,
        DataFeedRollupSettings,
        DataFeedMissingDataPointFillSettings
    )

service_endpoint = os.getenv("ENDPOINT")
subscription_key = os.getenv("SUBSCRIPTION_KEY")
api_key = os.getenv("API_KEY")
sql_server_connection_string = os.getenv("SQL_SERVER_CONNECTION_STRING")
query = os.getenv("SQL_SERVER_QUERY")

client = MetricsAdvisorAdministrationClient(
    service_endpoint,
    MetricsAdvisorKeyCredential(subscription_key, api_key)
)

data_feed = client.create_data_feed(
    name="My data feed",
    source=SqlServerDataFeedSource(
        connection_string=sql_server_connection_string,
        query=query,
    ),
    granularity="Daily",
    schema=DataFeedSchema(
        metrics=[
            DataFeedMetric(name="cost", display_name="Cost"),
            DataFeedMetric(name="revenue", display_name="Revenue")
        ],
        dimensions=[
            DataFeedDimension(name="category", display_name="Category"),
            DataFeedDimension(name="city", display_name="City")
        ],
        timestamp_column="Timestamp"
    ),
    ingestion_settings=datetime.datetime(2019, 10, 1),
    data_feed_description="cost/revenue data feed",
    rollup_settings=DataFeedRollupSettings(
        rollup_type="AutoRollup",
        rollup_method="Sum",
        rollup_identification_value="__CUSTOM_SUM__"
    ),
    missing_data_point_fill_settings=DataFeedMissingDataPointFillSettings(
        fill_type="SmartFilling"
    ),
    access_mode="Private"
)

return data_feed

Check ingestion status

After we start the data ingestion, we can check the ingestion status.

import datetime
from azure.ai.metricsadvisor import MetricsAdvisorKeyCredential, MetricsAdvisorAdministrationClient

service_endpoint = os.getenv("ENDPOINT")
subscription_key = os.getenv("SUBSCRIPTION_KEY")
api_key = os.getenv("API_KEY")
data_feed_id = os.getenv("DATA_FEED_ID")

client = MetricsAdvisorAdministrationClient(service_endpoint,
    MetricsAdvisorKeyCredential(subscription_key, api_key)
)

ingestion_status = client.list_data_feed_ingestion_status(
    data_feed_id,
    datetime.datetime(2020, 9, 20),
    datetime.datetime(2020, 9, 25)
)
for status in ingestion_status:
    print("Timestamp: {}".format(status.timestamp))
    print("Status: {}".format(status.status))
    print("Message: {}\n".format(status.message))

Configure anomaly detection configuration

While a default detection configuration is automatically applied to each metric, we can tune the detection modes used on our data by creating a customized anomaly detection configuration.

from azure.ai.metricsadvisor import MetricsAdvisorKeyCredential, MetricsAdvisorAdministrationClient
from azure.ai.metricsadvisor.models import (
    ChangeThresholdCondition,
    HardThresholdCondition,
    SmartDetectionCondition,
    SuppressCondition,
    MetricDetectionCondition,
)

service_endpoint = os.getenv("ENDPOINT")
subscription_key = os.getenv("SUBSCRIPTION_KEY")
api_key = os.getenv("API_KEY")
metric_id = os.getenv("METRIC_ID")

client = MetricsAdvisorAdministrationClient(
    service_endpoint,
    MetricsAdvisorKeyCredential(subscription_key, api_key)
)

change_threshold_condition = ChangeThresholdCondition(
    anomaly_detector_direction="Both",
    change_percentage=20,
    shift_point=10,
    within_range=True,
    suppress_condition=SuppressCondition(
        min_number=5,
        min_ratio=2
    )
)
hard_threshold_condition = HardThresholdCondition(
    anomaly_detector_direction="Up",
    upper_bound=100,
    suppress_condition=SuppressCondition(
        min_number=2,
        min_ratio=2
    )
)
smart_detection_condition = SmartDetectionCondition(
    anomaly_detector_direction="Up",
    sensitivity=10,
    suppress_condition=SuppressCondition(
        min_number=2,
        min_ratio=2
    )
)

detection_config = client.create_detection_configuration(
    name="my_detection_config",
    metric_id=metric_id,
    description="anomaly detection config for metric",
    whole_series_detection_condition=MetricDetectionCondition(
        condition_operator="OR",
        change_threshold_condition=change_threshold_condition,
        hard_threshold_condition=hard_threshold_condition,
        smart_detection_condition=smart_detection_condition
    )
)
return detection_config

Configure alert configuration

Then let's configure in which conditions an alert needs to be triggered.

from azure.ai.metricsadvisor import MetricsAdvisorKeyCredential, MetricsAdvisorAdministrationClient
from azure.ai.metricsadvisor.models import (
    MetricAlertConfiguration,
    MetricAnomalyAlertScope,
    TopNGroupScope,
    MetricAnomalyAlertConditions,
    SeverityCondition,
    MetricBoundaryCondition,
    MetricAnomalyAlertSnoozeCondition,
)
service_endpoint = os.getenv("ENDPOINT")
subscription_key = os.getenv("SUBSCRIPTION_KEY")
api_key = os.getenv("API_KEY")
anomaly_detection_configuration_id = os.getenv("DETECTION_CONFIGURATION_ID")
hook_id = os.getenv("HOOK_ID")

client = MetricsAdvisorAdministrationClient(
    service_endpoint,
    MetricsAdvisorKeyCredential(subscription_key, api_key)
)

alert_config = client.create_alert_configuration(
    name="my alert config",
    description="alert config description",
    cross_metrics_operator="AND",
    metric_alert_configurations=[
        MetricAlertConfiguration(
            detection_configuration_id=anomaly_detection_configuration_id,
            alert_scope=MetricAnomalyAlertScope(
                scope_type="WholeSeries"
            ),
            alert_conditions=MetricAnomalyAlertConditions(
                severity_condition=SeverityCondition(
                    min_alert_severity="Low",
                    max_alert_severity="High"
                )
            )
        ),
        MetricAlertConfiguration(
            detection_configuration_id=anomaly_detection_configuration_id,
            alert_scope=MetricAnomalyAlertScope(
                scope_type="TopN",
                top_n_group_in_scope=TopNGroupScope(
                    top=10,
                    period=5,
                    min_top_count=5
                )
            ),
            alert_conditions=MetricAnomalyAlertConditions(
                metric_boundary_condition=MetricBoundaryCondition(
                    direction="Up",
                    upper=50
                )
            ),
            alert_snooze_condition=MetricAnomalyAlertSnoozeCondition(
                auto_snooze=2,
                snooze_scope="Metric",
                only_for_successive=True
            )
        ),
    ],
    hook_ids=[hook_id]
)

return alert_config

Query anomaly detection results

We can query the alerts and anomalies.

import datetime
from azure.ai.metricsadvisor import MetricsAdvisorKeyCredential, MetricsAdvisorClient

service_endpoint = os.getenv("ENDPOINT")
subscription_key = os.getenv("SUBSCRIPTION_KEY")
api_key = os.getenv("API_KEY")
alert_config_id = os.getenv("ALERT_CONFIG_ID")
alert_id = os.getenv("ALERT_ID")

client = MetricsAdvisorClient(service_endpoint,
    MetricsAdvisorKeyCredential(subscription_key, api_key)
)

results = client.list_alerts(
    alert_configuration_id=alert_config_id,
    start_time=datetime.datetime(2020, 1, 1),
    end_time=datetime.datetime(2020, 9, 9),
    time_mode="AnomalyTime",
)
for result in results:
    print("Alert id: {}".format(result.id))
    print("Create time: {}".format(result.created_time))

results = client.list_anomalies(
    alert_configuration_id=alert_config_id,
    alert_id=alert_id,
)
for result in results:
    print("Create time: {}".format(result.created_time))
    print("Severity: {}".format(result.severity))
    print("Status: {}".format(result.status))

Query incidents

We can query the incidents for a detection configuration.

import datetime
from azure.ai.metricsadvisor import MetricsAdvisorKeyCredential, MetricsAdvisorClient

service_endpoint = os.getenv("ENDPOINT")
subscription_key = os.getenv("SUBSCRIPTION_KEY")
api_key = os.getenv("API_KEY")
anomaly_detection_configuration_id = os.getenv("DETECTION_CONFIGURATION_ID")

client = MetricsAdvisorClient(service_endpoint,
    MetricsAdvisorKeyCredential(subscription_key, api_key)
)

results = client.list_incidents(
            detection_configuration_id=anomaly_detection_configuration_id,
            start_time=datetime.datetime(2020, 1, 1),
            end_time=datetime.datetime(2020, 9, 9),
        )
for result in results:
    print("Metric id: {}".format(result.metric_id))
    print("Incident ID: {}".format(result.id))
    print("Severity: {}".format(result.severity))
    print("Status: {}".format(result.status))

Query root causes

We can also query the root causes of an incident

from azure.ai.metricsadvisor import MetricsAdvisorKeyCredential, MetricsAdvisorClient

service_endpoint = os.getenv("ENDPOINT")
subscription_key = os.getenv("SUBSCRIPTION_KEY")
api_key = os.getenv("API_KEY")
anomaly_detection_configuration_id = os.getenv("DETECTION_CONFIGURATION_ID")
incident_id = os.getenv("INCIDENT_ID")

client = MetricsAdvisorClient(service_endpoint,
    MetricsAdvisorKeyCredential(subscription_key, api_key)
)

results = client.list_incident_root_causes(
            detection_configuration_id=anomaly_detection_configuration_id,
            incident_id=incident_id,
        )
for result in results:
    print("Score: {}".format(result.score))
    print("Description: {}".format(result.description))

Add hooks for receiving anomaly alerts

We can add some hooks so when an alert is triggered, we can get call back.

from azure.ai.metricsadvisor import MetricsAdvisorKeyCredential, MetricsAdvisorAdministrationClient
from azure.ai.metricsadvisor.models import EmailNotificationHook

service_endpoint = os.getenv("ENDPOINT")
subscription_key = os.getenv("SUBSCRIPTION_KEY")
api_key = os.getenv("API_KEY")

client = MetricsAdvisorAdministrationClient(service_endpoint,
    MetricsAdvisorKeyCredential(subscription_key, api_key))

hook = client.create_hook(
    hook=EmailNotificationHook(
        name="email hook",
        description="my email hook",
        emails_to_alert=["alertme@alertme.com"],
        external_link="https://docs.microsoft.com/en-us/azure/cognitive-services/metrics-advisor/how-tos/alerts"
    )
)

Async APIs

This library includes a complete async API supported on Python 3.6+. To use it, you must first install an async transport, such as aiohttp. See azure-core documentation for more information.

from azure.ai.metricsadvisor import MetricsAdvisorKeyCredential
from azure.ai.metricsadvisor.aio import MetricsAdvisorClient, MetricsAdvisorAdministrationClient

client = MetricsAdvisorClient(
    service_endpoint,
    MetricsAdvisorKeyCredential(subscription_key, api_key)
)

admin_client = MetricsAdvisorAdministrationClient(
    service_endpoint,
    MetricsAdvisorKeyCredential(subscription_key, api_key)
)

Troubleshooting

General

The Azure Metrics Advisor clients will raise exceptions defined in Azure Core.

Logging

This library uses the standard logging library for logging.

Basic information about HTTP sessions (URLs, headers, etc.) is logged at INFO level.

Detailed DEBUG level logging, including request/response bodies and unredacted headers, can be enabled on the client or per-operation with the logging_enable keyword argument.

See full SDK logging documentation with examples here.

Next steps

More sample code

For more details see the samples README.

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit cla.microsoft.com.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Release History

1.0.0 (2021-07-06)

Breaking Changes

• Changed
  • DetectionConditionsOperator -> DetectionConditionOperator
  • cross_conditions_operator -> condition_operator
  • AnomalyAlert.created_on -> AnomalyAlert.created_time
  • AnomalyAlert.modified_on -> AnomalyAlert.modified_time
  • Anomaly.created_on -> Anomaly.created_time
  • admin_emails has been renamed to admins in NotificationHook
  • admin_emails has been renamed to admins in DataFeedOptions
  • viewer_emails has been renamed to viewers in DataFeedOptions

1.0.0b4 (2021-06-07)

New Features

• Added AzureLogAnalyticsDataFeedSource and AzureEventHubsDataFeedSource
• Update method now returns the updated object
• Added DatasourceCredentials and DatasourceCredential operations
• Added authentication type support for data feed

Breaking Changes

• Delete methods now take positional only argument as id

• update_subscription_key and update_api_key are merged into one method update_key

• Removed DataFeedOptions and moved all its properties to the DataFeed model

• Deprecated:

  • HttpRequestDataFeed
  • ElasticsearchDataFeed

• Renamed

  • AzureApplicationInsightsDataFeed -> AzureApplicationInsightsDataFeedSource
  • AzureBlobDataFeed -> AzureBlobDataFeedSource
  • AzureCosmosDBDataFeed -> AzureCosmosDbDataFeedSource
  • AzureDataExplorerDataFeed -> AzureDataExplorerDataFeedSource
  • AzureTableDataFeed -> AzureTableDataFeedSource
  • InfluxDBDataFeed -> InfluxDbDataFeedSource
  • MySqlDataFeed -> MySqlDataFeedSource
  • PostgreSqlDataFeed -> PostgreSqlDataFeedSource
  • SQLServerDataFeed -> SqlServerDataFeedSource
  • MongoDBDataFeed -> MongoDbDataFeedSource
  • AzureDataLakeStorageGen2DataFeed -> AzureDataLakeStorageGen2DataFeedSource

Dependency Updates

• Bump msrest requirement from 0.6.12 to 0.6.21

1.0.0b3 (2021-02-09)

New Features

• AAD support authentication #15922
• MetricsAdvisorKeyCredential support for rotating the subscription and api keys to update long-lived clients

Breaking Changes

• list_dimension_values has been renamed to list_anomaly_dimension_values
• update methods now return None
• Updated DataFeed.metric_ids to be a dict rather than a list

Hotfixes

• Bump six requirement from 1.6 to 1.11.0`

1.0.0b2 (2020-11-10)

Breaking Changes

• create_hook now takes as input an EmailHook or WebHook
• Anomaly has been renamed to DataPointAnomaly
• Incident has been renamed to AnomalyIncident
• IncidentPropertyIncidentStatus has been renamed to AnomalyIncidentStatus
• Alert has been renamed to AnomalyAlert
• Severity has been renamed to AnomalySeverity
• Metric has been renamed to DataFeedMetric
• Dimension has been renamed to DataFeedDimension
• EmailHook has been renamed to EmailNotificationHook
• WebHook has been renamed to WebNotificationHook
• Hook has been renamed to NotificationHook
• TimeMode has been renamed to AlertQueryTimeMode
• admins has been renamed to admin_emails on NotificationHook
• admins has been renamed to admin_emails on DataFeedOptions
• viewers has been renamed to viewer_emails on DataFeedOptions
• timestamp_list has been renamed to timestamps on MetricSeriesData
• value_list has been renamed to values on MetricSeriesData
• SeriesResult has been renamed to MetricEnrichedSeriesData
• create_anomaly_alert_configuration has been renamed to create_alert_configuration
• get_anomaly_alert_configuration has been renamed to get_alert_configuration
• delete_anomaly_alert_configuration has been renamed to delete_alert_configuration
• update_anomaly_alert_configuration has been renamed to update_alert_configuration
• list_anomaly_alert_configurations has been renamed to list_alert_configurations
• create_metric_anomaly_detection_configuration has been renamed to create_detection_configuration
• get_metric_anomaly_detection_configuration has been renamed to get_detection_configuration
• delete_metric_anomaly_detection_configuration has been renamed to delete_detection_configuration
• update_metric_anomaly_detection_configuration has been renamed to update_detection_configuration
• list_metric_anomaly_detection_configurations has been renamed to list_detection_configurations
• list_feedbacks has been renamed to list_feedback
• list_alerts_for_alert_configuration has been renamed to list_alerts
• list_anomalies_for_alert & list_anomalies_for_detection_configuration have been grouped into list_anomalies
• list_dimension_values_for_detection_configuration has been renamed to list_dimension_values
• list_incidents_for_alert & list_incidents_for_detection_configuration have been grouped into list_incidents

New Features

• __repr__ added to all models

1.0.0b1 (2020-10-07)

First preview release