Serilog Timings

Serilog's support for structured data makes it a great way to collect timing information. It's easy to get started with in development, because the timings are printed to the same output as other log messages (the console, files, etc.) so a metrics server doesn't have to be available all the time.

Serilog Timings is built with some specific requirements in mind:

One operation produces exactly one log event (events are raised at the completion of an operation)
Natural and fully-templated messages
Events for a single operation have a single event type, across both success and failure cases (only the logging level and Outcome properties change)

This keeps noise in the log to a minimum, and makes it easy to extract and manipulate timing information on a per-operation basis.

Important

SerilogTimings is maintained, but "done". Its simple, lightweight, portable, opinionated API is considered a feature, and it's unlikely any further changes will be made. For integration with .NET's tracing system, including out-of-the-box support for ASP.NET Core, HttpClient, and SqlClient tracing, check out SerilogTracing, the logical successor to SerilogTimings, influenced strongly by the SerilogTimings API. Read this migration guide if you're interested in porting an existing codebase to the new library.

Installation

The library is published as SerilogTimings on NuGet.

Install-Package SerilogTimings -DependencyVersion Highest

The package works on all currently-supported .NET versions.

Getting started

Before your timings will go anywhere, install and configure Serilog.

Types are in the SerilogTimings namespace.

using SerilogTimings;

The simplest use case is to time an operation, without explicitly recording success/failure:

using (Operation.Time("Submitting payment for {OrderId}", order.Id))
{
    // Timed block of code goes here
}

At the completion of the using block, a message will be written to the log like:

[INF] Submitting payment for order-12345 completed in 456.7 ms

The operation description passed to Time() is a message template; the event written to the log extends it with " {Outcome} in {Elapsed} ms".

All events raised by SerilogTimings carry an Elapsed property in milliseconds
Outcome will always be "completed" when the Time() method is used

All of the properties from the description, plus the outcome and timing, will be recorded as first-class properties on the log event.

Operations that can either succeed or fail, or that produce a result, can be created with Operation.Begin():

using (var op = Operation.Begin("Retrieving orders for {CustomerId}", customer.Id))
{
	// Timed block of code goes here

	op.Complete();
}

Using op.Complete() will produce the same kind of result as in the first example:

[INF] Retrieving orders for customer-67890 completed in 7.8 ms

Additional methods on Operation allow more detailed results to be captured:

    op.Complete("Rows", orders.Rows.Length);

This will not change the text of the log message, but the property Rows will be attached to it for later filtering and analysis.

If the operation is not completed by calling Complete(), it is assumed to have failed and a warning-level event will be written to the log instead:

[WRN] Retrieving orders for customer-67890 abandoned in 1234.5 ms

In this case the Outcome property will be "abandoned".

To suppress this message, for example when an operation turns out to be inapplicable, use op.Cancel(). Once Cancel() has been called, no event will be written by the operation on either completion or abandonment.

Use with `ILogger`

If a contextual ILogger is available, the extensions methods TimeOperation() and BeginOperation() can be used to write operation timings through it:

using (logger.TimeOperation("Submitting payment for {OrderId}", order.Id))
{
    // Timed block of code goes here
}

These otherwise behave identically to Operation.Time() and Operation.Begin().

`LogContext` support

If your application enables the Serilog LogContext feature using Enrich.FromLogContext() on the LoggerConfiguration, Serilog Timings will add an OperationId property to all events inside timing blocks automatically.

This is highly recommended, because it makes it much easier to trace from a timing result back through the operation that raised it.

Levelling

Timings are most useful in production, so timing events are recorded at the Information level and higher, which should generally be collected all the time.

If you truly need Verbose- or Debug-level timings, you can trigger them with Operation.At() or the OperationAt() extension method on ILogger:

using (Operation.At(LogEventLevel.Debug).Time("Preparing zip archive"))
{
    // ...

When a level is specified, both completion and abandonment events will use it. To configure a different abandonment level, pass the second optional parameter to the At() method.

Caveats

One important usage note: because the event is not written until the completion of the using block (or call to Complete()), arguments to Begin() or Time() are not captured until then; don't pass parameters to these methods that mutate during the operation.

How does this relate to SerilogMetrics?

SerilogMetrics is a mature metrics solution for Serilog that includes timings as well as counters, gauges and more. Serilog Timings is an alternative implementation of timings only, designed with some different stylistic preferences and goals. You should definitely check out SerilogMetrics as well, to see if it's more to your tastes!