dotTrace 2017.3 Help

Analyzing Async Calls

One of the downsides of asynchronous code is it's extremely difficult to profile it and analyze its performance. This is because, when an asynchronous method is executed, control is switched to a different thread and back, tangling the resulting call tree.

dotTrace dramatically simplifies the analysis of asynchronous code: It marks all async call nodes in Call Tree and groups the corresponding await time and continuation code under that node. This means that you can quickly find all "parts" of an asynchronous call in one place instead of searching for them in different call stacks.

To better understand how dotTrace treats asynchronous code, consider the following example (the code is shown on the left, the corresponding Call Tree on the right):

async calls expanded readasync

As you can see, all "parts" of the asynchronous call are shown inside the async RunAsyncOperation node:

  • The total time of RunAsyncOperation is calculated as
    119 ms = Init() 101 ms + ReadAsync() 13 ms + clr.dll 3.7 ms + FileStream.ctor() 0.6 ms
  • The Init method (101 ms) is executed on the Main thread, therefore, its time is added to the total time of RunAsyncOperation.
  • The ReadAsync is started on the Main thread (13 ms) but the subsequent task is run on a thread pool. Thus, the time of the Task execution node (819 ms) is shown grey and is not added to the total time of RunAsyncOperation.
    async calls task execution
  • The await time in our case is equal to the Task execution time (819 ms) but in real life it can be higher as it also includes the time the task waits in schedule.
  • The continuations node is a continuation code which in our case consists of a single ProcessFile method (301 ms). As this call is executed on the thread pool, its time is also shown gray and is not added to the total time of RunAsyncOperation.

Identifying the slowest async calls in Methods and Subsystems

As you can see, the total async call time does not include the time of all its parts in Call Tree. But what if you want to quickly identify the 'hot spots' - the slowest methods in your app? For such kind of task it would be better if the async method's total time included the continuation code. This is actually true for the Methods and Subsystems list, but only if you stand on the All Calls node in Call Tree.

async calls expanded readasync topmethods

Backtraces of the continuation code

Of course, Backtraces of the continuation code will lead you back not just to the callback function but to the original async method. This could be very helpful, e.g., when the continuation code throws an exception and you need to identify its origin.

async calls backtraces

Filtering by async call's total time

To apply the filter by async call's total time, either double-click the call node in Call Tree or right-click the node (or its await or continuations node) and select Analyze Async Method from the context menu.

After you filter by an async method's call time, dotTrace will leave only the time intervals where the method was executed. Note that you can include or exclude continuation code intervals (as well as the await time node) by selecting the corresponding check boxes that appear in Call Tree.

Note that if you apply filters so that the continuation code will be out of scope (e.g., a filter by Main thread), the Continuations check box will not be shown.

Tasks in Call Tree

The aforementioned functionality works not only with async/await but with all tasks based on the Task class. The Run node contains the Task execution node with the task delegate:

async calls task

Asynchronous calls and events

Last but not least worth to mention Call Tree works with calls time as well as with all other types of events supported in the Timeline profiling mode, e.g., memory allocation or exceptions. For example, you can view how much memory a particular async method allocates:

async calls allocations

Last modified: 16 April 2018