Bearbeiten

Teilen über


How to work with a performance problem

What do you do if users complain that "it's slow"? In this section, we describe a troubleshooting process that can help to guide you to find the root cause of the problem.

A generic performance tuning process

Before you get started on solving a performance-tuning problem, it often helps to define and quantify "slow" and also negotiate acceptable values for execution time of the "slow" operations with users. Learning about the user's expectations is key when determining when the performance tuning effort is successful (or if the expectations are unrealistic).

You also need to measure how the system performs on the given operation before starting to change anything. This task is sometimes called "establishing a baseline." To define baselines for performance, and to test whether new code or extensions introduce a performance regression, you can use the Performance Toolkit extension. The extension makes it easier to simulate and compare user experiences to your baseline. The following are examples of when the extension can help:

  • When you want to ensure that new code doesn't introduce a regression.
  • In a sandbox environment when, for example, the number of users running the same process increases significantly.
  • When you want to roll out a new process, or install a new extension.

To solve a performance problem, a common pattern is to do iterations of the following tasks:

  1. Measure system performance and collecting performance data
  2. Locate a bottleneck
  3. Eliminate the bottleneck

Continue until the "slow" operations are comparable to the established baseline.

Performance analysis tools

Business Central comes with tools that can be used to analyze a performance problem. In the following, you can read about the pros and cons of the different performance tools described below.

Performance tool Properties
Page inspector Good to troubleshoot performance of a single page.
No need to enable this (always available).
End users can run the tool.
Data collection must happen live.
In-client performance profiler Good for troubleshooting a performance scenario in the web client.
No developer required to run the tool.
For more information, see In-client Performance Profiler overview.
Telemetry Can be used if you want to investigate things after they happened.
Good for analyzing patterns across sessions.
Extensive resources available (Power BI report, Jupyter notebooks, sample KQL queries).
Little performance impact to have turned on always.
Telemetry must be enabled before the performance issue occurs.
Not every single AL call is logged to telemetry as this would slow down the Business Central server.
Verbose telemetry Will give you all SQL queries for the session where you repro the issue.
Slows down the system while running.
Can inject much data into Azure Application Insights.
Data collection must happen live.
Database performance pages The pages Database Missing Indexes and Database Wait Statistics show insights into database performance and how to fix it.
AL profiler Good to troubleshoot performance of a scenario.
Detailed information on where in the code the time is spent.
No need to enable this (always available).
Requires a developer to run the tool.
Data collection must happen live.
For more information, see AL Profiler

Analyzing performance issues using the page inspector

If a specific page takes too long to load, it might be due to extensions that are adding expensive operations to the page triggers. You can use the page inspector to analyze this performance by navigating to the Extensions tab. It displays installed extensions that affect the selected page or its source table by showing the time (in milliseconds) it took to run the extension in the call stack and the number of event subscribers run in the extension.

Read more about how to use the page inspector to troubleshoot extension performance here Inspecting and Troubleshooting Pages.

Analyzing performance issues using the in-client Performance Profiler

Use the in-client Performance Profiler page to record a slow scenario that can then be analyzed to see where time was spent. The tool is simple to use and can therefore be used by end-users, admins, and consultants to do performance investigations directly in the web client, to verify performance issues, understand which extensions are at play, and the likelihood of an extension being the cause of a performance degradation.

To profile a scenario, the overall steps are the following:

  1. Start the Performance Profiler
  2. Perform the (slow) scenario
  3. Stop the Performance Profiler

For more specific information on how to start and use the in-client Performance Profiler see In-client Performance Profiler overview.

When analyzing the result, you have three options:

  1. Look at Active Apps to see in which app time is spent (did you recently install an app?).
  2. Select an app to see how time is distributed over the apps/extensions that have code running in the app.
  3. Go to Time Spent by Application Object to see the distribution of time spent in the AL call tree (this is an advanced option).
  4. Download the generated profile content and share it with a developer. The profile file can be viewed in Visual Studio Code with the standard AL Profiler editor. From there, you can use existing options to access the AL code that was slow.

The in-client Performance Profiler is a lighter tool than the AL Profiler and, as it relies on sampling, it can perform in scenarios that would otherwise take longer time when using the AL Profiler with the instrumentation option. For more specific information on how to use the in-client Performance Profiler to troubleshoot performance see In-client Performance Profiler overview. For more information about the AL Profiler, see AL Profiler.

Analyzing performance issues using telemetry

For monitoring and analyzing performance issues in the Business Central service, we recommend connecting Azure AppInsights to the environments that you want to get signals from. For more information, see Enable Sending Telemetry to Application Insights.

In the following, you can read about ways where telemetry can help troubleshoot performance issues.

Area Telemetry Why
Some page is slow (and it uses Page Background Tasks) Authorization signal Each page background task opens a new session. Any expensive action in the OnCompanyOpen trigger slows down opening new sessions.
Sign-in Authorization signal Any expensive action in the OnCompanyOpen trigger slows down opening new sessions.
Something was slow during this period of time Company lifecycle signal Check whether a copy-company operation was running while the performance issue occurred.
Something was slow during this period of time Database locks signal Maybe the performance issue was because of locking in the database.
Suddenly the XYZ page is slow Extension lifecycle signal Maybe an extension was installed that interferes with the page in question.
Some pages or reports are slow to load Long running SQL queries Investigate whether the data operations on the page or report are taking a long time to complete.
A report is slow Report signal Check whether the report is reading more data than you expected.
System UI feels slow Web service requests signal Calling your environment too aggressively with web service requests can affect performance of the system.

This page shows an overview of all currently available signals: Monitoring and Analyzing Telemetry.

Analyzing database performance issues using verbose telemetry

If you want to capture all SQL queries for a short period of time for a given session, you can enable Additional logging from the Help & Support page. The logging lets you analyze queries that take a short time to run but happen frequently.

Read more about extra logging in the Troubleshooting section of the Resources for Help and Support article in the business functionality docs.

With Business Central on-premises, you can also change the default threshold that defines long running queries. For more information, see Analyzing Long Running Operation (SQL Query) Telemetry.

Analyzing performance issues using database missing indexes

If you notice that the information retrieval from your database is slow, you can index columns that are frequently used by the application. You can see the list of potential columns that can be indexed and other useful information about them on Database Missing Indexes. This information will help you to optimize the performance of your database and application.

For more information about missing indexes and how they help you in achieving performance objectives, see Missing Indexes in Dynamics 365 Business Central.

Analyzing performance issues using database wait statistics

The database performance can also be analyzed by checking the information at Database Wait Statistics. You can see how many queries got delayed because of different wait types along with their wait times. The wait types give you an idea about the resources or operations that are responsible for the slow performance.

For more information about wait statistics in optimizing database and application performance, see Database Wait Statistics in Dynamics 365 Business Central.

Analyzing performance issues using the AL Profiler

With the AL Profiler for the AL Language extension, you can capture a performance profile of the code that was executed for a snapshot. Using the performance profiling editor view in Visual Studio Code, you can investigate the time spent on execution, using top-down and bottom-up call stack views.

Before capturing a snapshot with a performance profile, you can choose between using instrumentation or sampling for the profile. Instrumentation captures all method calls with precise timings, but has a higher performance load on the NST, which means that the snapshot will be larger and it can take longer to generate and extract the profile from the snapshot. Sampling captures method calls at predefined, repeated intervals. This has a smaller performance load on the NST, but calls shorter than the sampling interval might not be captured, and timings fall in sample buckets.

Read more about how to use the AL profiler to troubleshoot performance here AL Profiler Overview.

Example - How to deal with a performance problem in a report from an AppSource extension

This example illustrates how you can use the performance tuning process and telemetry to deal with a performance problem in a report from an AppSource extension.

  • Together with the tenant administrator/customer, you need to define "slow" and what the acceptable rendering time will be.
  • Then, use telemetry to find data about time spent in the report: long running SQL queries (eventId RT0005) and report rendering time (eventId RT0006) are the main data sources for this. Telemetry for long running AL methods (eventId RT00180) might also be useful here. Now you have baseline data.
  • Also, use telemetry on report rendering time (eventId RT0006) to find out if the report is running with data access intent ReadOnly. If not, configure it to do so. Some reports write data, so this isn't always possible.
  • Do you see long running SQL queries for the report in telemetry (eventId RT0005)? If so, you can add appropriate indexes using a table extension.
  • After configuring, data access intent and maybe having added indexes, measure rendering time and compare with the baseline. Also check if the new rendering time is good enough with respect to the acceptable rendering time that you defined together with the tenant administrator/customer. If the performance is still not good enough, you probably need to involve the extension publisher.
  • What is the SQL statement to SQL row ratio (get this using eventId RT0006)? Maybe the extension publisher can make the report more set-based when reading data. The telemetry samples repository at aka.ms/bctelemetrysamples has KQL samples that can help with this analysis.
  • If you have data to back it up, consider filing a performance support request to the extension publisher. They're probably interested in making their app better and faster. Provide info about indexes you added, the kusto queries you used and the results, and also if the report could run as ReadOnly by default.

See also

Performance overview
Performance topics For developers
Performance tips for business users
How application configurations affect performance
Performance online
Performance of on-premises installations
Moving from OnCompanyOpen