DEV Community

PBIDocs
PBIDocs

Posted on

"Why Did My Power Query Refresh Suddenly Get Slower?"

#powerquery #performance #troubleshooting

There's no error message for this one — the refresh just finishes. It's just slower. A query that used to take ten seconds now takes ten minutes, nothing was obviously changed, and there's nothing in the UI shouting about why.

Last week: refresh in 8 seconds
This week: refresh in 6 minutes
No error. No warning. Nothing "broke" in the way an error implies.
Enter fullscreen mode Exit fullscreen mode

This almost always means query folding stopped somewhere in the query — silently, since a broken fold isn't an error condition, just a much slower execution path.

How to Actually Check

Right-click any step in Applied Steps and look at View Native Query. If it's available, everything up to and including that step is still folding to the source. The moment it's grayed out, folding has already stopped at or before that step.

Source          -> View Native Query available
Filtered Rows   -> View Native Query available
Added Custom    -> View Native Query grayed out   <- folding stopped here
Renamed Columns -> still grayed out (nothing after a break can fold again)
Enter fullscreen mode Exit fullscreen mode

Work backward from the last step, checking each one, until the option is available again — that's the exact boundary where folding broke.

Cause 1: A New Step Was Added in the Wrong Position

The most common cause of a query that "used to be fast." Adding a new step doesn't insert it at the end of a logical plan — it inserts it exactly where Applied Steps shows it, and everything downstream inherits whatever folding state that step leaves behind.

Before:  Source -> Filter (folds) -> Select Columns (folds) -> Changed Type (folds)
After:   Source -> Filter (folds) -> Added Custom Column (doesn't fold) -> Select Columns -> Changed Type
                                            ^
                                   everything from here runs locally now
Enter fullscreen mode Exit fullscreen mode

Fix: move steps that can't fold (custom columns, complex conditional logic) as late in the sequence as possible — after filtering and column selection, not before. See Ordering Steps to Preserve Folding.

Cause 2: A Custom Column With Row-by-Row Logic

Table.AddColumn with an each expression referencing M functions that have no equivalent in the source's native query language can't be translated back — the entire step, and everything after it, has to run locally.

#"Added Custom" = Table.AddColumn(
    Source, "Flag",
    each if Text.Contains([Notes], "urgent") then "Y" else "N"
)
Enter fullscreen mode Exit fullscreen mode

This is often unavoidable — not every transformation has a source-side equivalent — but it's worth knowing it's the trade being made, and placing it as late as possible so it affects the smallest number of rows and downstream steps.

Cause 3: Table.Buffer in the Wrong Spot

Table.Buffer() forces full materialization into memory — useful for stabilizing a volatile source, but it also ends folding immediately at that point, even if every step before and after it would otherwise fold cleanly.

Source (folds) -> Filter (folds) -> Table.Buffer -> Group (doesn't fold, runs locally)
Enter fullscreen mode Exit fullscreen mode

See Table.Buffer — this one is easy to miss specifically because it doesn't look like a transformation at all, just a performance-sounding function name.

Cause 4: A Merge Against a Non-Folding Source

Merging a folding query (say, a SQL table) with a query from a source that can't fold (an Excel file, a CSV, an API call) means the combined result can't be pushed back to a single source's native query language — there's no one engine that understands both halves.

SQL query (folds) + Excel query (never folds)
        |
        merged
        |
Result: doesn't fold, regardless of how well the SQL side folds alone
Enter fullscreen mode Exit fullscreen mode

This is sometimes unavoidable (the data genuinely lives in two different places), but it's worth knowing the merge itself is where folding ends, not something to debug further downstream.

Common Mistakes

Assuming a slow refresh means the source is just slow. It's easy to blame the database or the network before checking whether the query itself stopped folding — check View Native Query before escalating to infrastructure.

Adding steps in whatever order feels natural, not a folding-aware order. Applied Steps records the order things were built, not necessarily the order they should stay in — reordering after the fact is normal and often the entire fix.

Not re-checking folding after adding new steps to a previously-fast query. A query that folded perfectly last month can silently stop folding the moment one new step is added — checking once at the start isn't enough if the query keeps evolving.

Originally published on PBIDocs — Power BI documentation covering DAX, Power Query, data modeling, and Microsoft Fabric.

Top comments (0)