Tag Archives: nosql

Using Analytics in Application Insights to monitor DocumentDB Requests

Following Wikipedia, DocumentDB is

Microsoft’s multi-tenant distributed database service for managing JSON documents at Internet scale.

The throughput of the database is charged and measured in request unit per second (RUs). Therefore, when creating application on top of DocumentDB, this is a very important dimension that you should pay attention to and monitor carefully.

Unfortunately, at the time of the writing the Azure portal tools to measure your RUs usage are very poor and not really usable. You have access to tiny charts where granularity cannot be really changed.

DocumentDB monitoring charts in Azure Portal

These are the only monitoring charts available in the Azure Portal

In this blog post, I show how Application Insights Analytics can be used to monitor the RUs consumption efficiently. This is how we monitor our collections now at Keluro.

Let us start by presenting Application Insights, it defines itself here as

an extensible Application Performance Management (APM) service for web developers on multiple platforms. Use it to monitor your live web application. It will automatically detect performance anomalies. It includes powerful analytics tools to help you diagnose issues and to understand what users actually do with your app.

Let us show how to use it in a C# application that is using the DocumentDB .NET SDK.

First you need to install the Application Insights Nuget Package. Then, you need to track the queries using a TelemetryClient object, see a sample code below.

public static async Task<FeedResponse<T>> LoggedFeedResponseAsync<T>(this IQueryable<T> queryable, string infoLog, string operationId)
	var docQuery = queryable.AsDocumentQuery();
	var now = DateTimeOffset.UtcNow;
	var watch = Stopwatch.StartNew();
	var feedResponse = await docQuery.ExecuteNextAsync<T>();
	TrackQuery(now, watch.Elapsed, feedResponse.RequestCharge, "read", new TelemetryClient(), infoLog, operationId, feedResponse.ContentLocation);
	return feedResponse;

public static void TrackQuery(DateTimeOffset start, TimeSpan duration, double requestCharge, string kind, TelemetryClient tc, string infolog, string operationId, string contentLocation)
	var dependency = new DependencyTelemetry(
			"0", // Result code : we can't capture 429 here anyway
			true // We assume this call is successful, otherwise an exception would be thrown before.
	dependency.Metrics["request-charge"] = requestCharge;
	dependency.Properties["kind"] = kind;
	dependency.Properties["infolog"] = infolog;
	dependency.Properties["contentLocation"] = contentLocation ?? "";
	if (operationId != null)
		dependency.Context.Operation.Id = operationId;

The good news is that you can now effectively keep records of all requests made to DocumentDB. Thanks to a great component of Application Insights named Analytics, you can browse the queries and see their precise request-charges (the amount of RUs consumed).

You can also add identifiers (with variables such as kind and infolog in sample above) from your calling code for a better identification of the requests. Keep in mind that the request payload is not saved by Application Insights.

In the screenshot below you can list and filter the requests tracked with DocumentDB in Application Insights Analytics thanks to its amazing querying language to access data.

Getting all requests to DocumentDB in a a timeframe using application Insights Analytics

Getting all requests to DocumentDB in a a timeframe using application Insights Analytics

There is one problem with this approach is that for now, using this technique and DocumentDB .NET SDK we do not have access to the number of retries (the 429 requests). This is an open issue on Github.

Finally, Analytics allows us to create a very important chart. The accumulated RUs per second for a specific time range.
The code looks like the following one.

| where timestamp > ago(10h)
| where type == "DOCDB"
| extend requestCharge = todouble(customMeasurements["request-charge"])
| extend docdbkind = customDimensions["kind"]
| extend infolog = customDimensions["infolog"]
| order by timestamp desc
| project  timestamp, target, data, resultCode , duration, customDimensions, requestCharge, infolog, docdbkind , operation_Id 
| summarize sum(requestCharge) by bin(timestamp, 1s)
| render timechart 

And the rendered charts is as follows

Accumulated Request-Charge per second (RUs)

Accumulated Request-Charge per second (RUs)