v3.6.0 Release

Version 3.6.0 is one of the larger releases in a while. As well as a long list of fixes and performance work, it introduces the Delta Analyzer - a new preview tool for looking at the Delta table metadata sitting behind a Direct Lake model. This post also covers the two diagramming features that arrived in v3.5.0 but have not been written up until now.

Delta Analyzer (Preview)

When you are troubleshooting a Direct Lake model, the performance you see in DAX Studio is often driven by things that live outside the semantic model. How many parquet files a Delta table is spread across, how large the row groups are, whether V-Order was applied when the files were written, and whether rows have been soft deleted with deletion vectors all affect how quickly Direct Lake can transcode and scan a column.

The new Delta Analyzer reads that metadata directly from OneLake and presents it per table and per column, so you can see file counts, small file counts, row group sizing, compressed and uncompressed sizes, encoding and codec, V-Order, deletion vectors and liquid clustering all in one place - right next to the query timings you are already collecting.

It also raises a set of considerations for each table, such as flagging tables that would benefit from compacting small files or from having their deletion vectors purged. These are deliberately framed as guidance rather than hard rules, since the right answer always depends on your data volumes and query patterns. You can copy them to the clipboard as Markdown or export them to a file.

Because it is a preview feature you need to switch it on first via Options > Preview > Delta Analyzer > Show Delta Analyzer. Once enabled, the Delta Analyzer button appears in the Metrics group on the Advanced ribbon tab.

Note that the UI may still change before this feature comes out of preview - feedback is welcome.

Read more about the Delta Analyzer

Two diagramming features you may have missed

These two are not new in 3.6.0 - they quietly appeared in v3.5.0 - but they have never been documented, so if you have not turned them on yet this is a good time to take a look. Both live under Options > Preview > Diagrams and both are off by default.

Model Diagram

The Model Diagram draws the connected model on a single canvas, showing tables, columns and relationships, with cardinality and filter direction markers on each relationship line. It is handy when you are working with a model you did not build and you want to see how the tables hang together, or when you want to isolate just the tables involved in the query you are tuning.

If you have run View Metrics first, the diagram can also show column cardinality or column size beside each column and sort the columns in each table by those values, which makes it easy to spot the expensive columns at a glance. It also works offline against a .daxx file that has VertiPaq Analyzer data in it.

Tables can be dragged around and the layout is remembered per model, there are five layout algorithms to choose from, and you can add text annotations before exporting the diagram to a PNG file or copying it to the clipboard.

Enable Show Model Diagram and the Model Diagram button appears in the Model group on the Advanced ribbon tab.

Read more about the Model Diagram

Storage Engine Dependencies

Storage Engine Dependencies takes a different approach. Instead of diagramming the model, it parses the xmSQL captured by the Server Timings trace and diagrams only what the Storage Engine actually touched for the query you just ran.

Each table box lists the columns the parser found, badged with how they were used - join key, filter, selected, expression, callback or aggregate - and the lines between the tables show the joins that the Storage Engine performed, along with their hit counts. The table headers are coloured as a heat map so the hot spots stand out, and you can switch the heat metric between CPU time, row count and hit count.

Enable Show Storage Engine Dependencies and a Dependencies button appears on the Server Timings pane once you have run a query with the trace enabled.

Read more about Storage Engine Dependencies

Both of these are still preview features, so the user interface may change before they are finished

• feedback on either of them is very welcome.

Benchmarking from the command line

dscmd.exe has gained a benchmark command, so the cold and warm cache benchmarking that has been available in the UI for a while can now be scripted or run as part of a build pipeline.

dscmd.exe benchmark c:\temp\results.csv -s localhost\tabular -d "Adventure Works" -f query.dax --cold 5 --warm 5

It writes per-iteration detail rows plus summary statistics to a CSV file, prints a summary table with the Formula Engine / Storage Engine split and row counts to the console, and supports --silent for unattended use. There are also --role and --effective-user options for benchmarking a query under row level security.

Read more about the benchmark command

DAX user-defined function support

DAX Studio now understands model-defined DAX user-defined functions when expanding measures and their dependencies. If a measure references a user-defined function, DAX Studio adds the required FUNCTION definition to the query before the measure definitions, including any functions on which it depends.

This also works when expanding a measure inline. Existing function definitions are detected so that they are not added to the query more than once.

More reliable command-line authentication

Authentication in dscmd.exe has been updated to make unattended execution more predictable, especially on machines where multiple Entra accounts are available. Use -u or --userid to select the account explicitly and --non-interactive to prevent a sign-in prompt. If the selected account cannot authenticate silently, the command exits with a non-zero exit code instead of waiting for user input.

The new auth command lets you sign in and cache an account without printing its access token. It can also list the accounts available from Windows and the DAX Studio cache, or act as a pre-flight check before an unattended job.

dscmd auth -u user@contoso.com --non-interactive

Read more about the auth command

A more consistent command line

The command line parsing for both daxstudio.exe and dscmd.exe has been rebuilt on top of Spectre.Console.Cli. Arguments and command names are now case-insensitive, --option=value syntax works, and the DOS style /option form as well as legacy single dash long options like -file are still accepted so existing shortcuts and scripts keep working. Values that need their casing preserved - file paths, uri values and base64 payloads - are left exactly as you typed them.

The startup parameters for daxstudio.exe are now documented too.

Read more about the startup parameters

Version notifications have moved into the app

The old notification tray icon has been removed, and new version notifications are now shown as a pill in the DAX Studio title bar instead. There is also a new opt-in Show Pre-Release Notifications setting for anyone who wants to hear about preview builds as well as stable releases.

Read more about version notifications

Updates

• Added the Delta Analyzer preview feature for Direct Lake models
• Added a benchmark command to dscmd for scripted cold and warm cache benchmarking
• Added support for including DAX user-defined functions when expanding measures and dependencies
• Added a secure auth command and non-interactive Entra authentication support to dscmd
• Command line arguments for both daxstudio.exe and dscmd.exe are now case-insensitive
• Version notifications now appear in the main window instead of the notification tray
• Added an option to be notified about pre-release builds
• Added bulk Order By controls to the Query Builder so you can toggle sorting, and flip the sort direction, for all columns at once
• Added multi row copy support to the Physical and Logical Query Plan panes
• Added a search box to the recent files list in the File menu
• Improved the Browse Workspaces experience
• The Entra ID sign-in prompt now attaches to the correct window
• Added zoom support to the Server Timings and Custom Trace panes
• Updated to the new DAX Formatter endpoint
• Updated to Fluent.Ribbon v11, which refreshes the ribbon styling across the light, dark and high contrast themes
• Switched to Microsoft.Data.SqlClient and to a shared HttpClient
• Added ApplicationContext to QueryEnd trace events
• Converted the solution to SDK style projects with central package management, and updated a large number of dependencies
• Improved application startup and general performance, including loading query history without blocking startup
• The installer now overwrites all files when upgrading

Fixes

• Fix #1453 Incorrect code completion for object names containing numeric digits
• Fix #1456 Unable to copy Physical or Logical Query Plan rows
• Fix #1458 Bulk options in the Query Builder to adjust the Order By
• Fix #1460 Callbacks without parameters were not being detected, so they appeared as plain strings rather than highlighted text
• Fix #1477 Use the correct less-than-or-equal and greater-than-or-equal symbols in data grid filters
• Fix handling of escaped callback names in the xmSQL parser
• Fix search not working in the database dialog
• Fix the trace column cache not being refreshed on reconnect
• Fix expired access tokens when reusing trace connections
• Fix View As reconnect races, remap caching, and stale busy states
• Fix group column alignment
• Fix an edge case with Server Timings and Clear Cache on Run when events arrive out of order
• Fix the save prompt when closing modified document tabs
• Fix a race condition when building the options list
• Fix a crash when validating hotkeys
• Fix invalid window handle crashes
• Fix a crash in DaxStudio.Checker
• Fix assembly binding errors in dscmd
• Harden opening SQL Profiler
• Handle WPF render thread COM errors and a shutdown race condition
• Fix a minor bug in the query history server filter
• Improve the robustness of the Excel version check