Performance Optimization

Analogy categoryAnalogies

A Power Apps canvas app that feels instant with 50 test records can feel unusable with 50,000 real ones, and the causes are almost always the same handful of culprits: non-delegable queries silently truncating or slowing down data retrieval, a bloated OnStart or OnVisible chain that blocks the first screen from rendering, formulas that recompute expensive expressions repeatedly instead of caching the result once, and an excess of controls or unoptimized images weighing down the client. Performance work in Power Apps is mostly about being deliberate: knowing which functions delegate to the data source, knowing when a formula runs, and measuring instead of guessing.

🏏

Cricket analogy: It is like a bowler who looks unplayable in the nets against club batters but gets carted around Eden Gardens by Suryakumar Yadav in a real T20, because the test conditions never matched match conditions.

Delegation: The Single Biggest Factor

Delegation is Power Apps' mechanism for pushing a query's filtering, sorting, and aggregation work down to the data source instead of pulling all records to the client first. Each connector supports a different set of delegable functions and operators — Dataverse and SQL Server delegate broadly, including most comparisons and some string functions, while SharePoint delegates a narrower set and historically struggles with delegating 'contains' on choice columns or complex nested conditions. When a formula uses a non-delegable function, Studio shows a blue delegation warning, and the app silently applies only the first 500 records (default, configurable up to 2000) rather than the full dataset, which produces bugs that only appear once real data volume exceeds that threshold.

🏏

Cricket analogy: It is like a stadium's DRS review being limited to only the deliveries captured by the ball-tracking cameras; anything outside that coverage zone simply can't be reviewed, no matter how confident the fielding side is.

text

// Delegable (pushed to SharePoint/Dataverse/SQL server-side)
Filter(Orders, Status = "Open" && Region = "West")
SortByColumns(Orders, "OrderDate", SortOrder.Descending)

// NON-delegable examples that trigger the blue warning:
Filter(Orders, Len(CustomerName) > 10)         // Len() not delegable on SharePoint
Filter(Orders, Left(Status, 4) = "Open")        // Left() not delegable on SharePoint
Filter(Orders, Status.Value in ["Open","Held"]) // 'in' on choice columns limited

// Fix: use a delegable equivalent
Filter(Orders, StartsWith(Status, "Open"))      // StartsWith IS delegable on many sources

Load-Time Strategy: OnStart, OnVisible, and Concurrent

Every function chained with semicolons in App.OnStart runs sequentially before the first screen paints, so an OnStart that loads five collections one after another can make a user stare at a blank splash screen for several seconds. Wrapping independent operations in Concurrent() lets Power Apps run them in parallel instead of sequentially, and moving screen-specific data loads out of OnStart into that screen's own OnVisible means the app only pays the cost for a screen when the user actually navigates to it. Named formulas (App.Formulas) go further by removing the concept of 'loading' entirely for values that can be computed lazily and cached, rather than force-loaded at startup.

🏏

Cricket analogy: It is like a T20 team sending in only the openers to pad up before the toss instead of having the entire top six get ready in the dressing room simultaneously before anyone knows who's batting.

The Monitor tool (Studio > Advanced tools > Monitor, or launched via 'Monitor' when running the app) traces every network call, formula evaluation, and control event with precise timings. It is the single best way to find a specific slow OnSelect or a data call that's firing more often than expected, rather than guessing from user complaints alone.

Client-Side Load: Controls, Images, and Collections

Every control on a screen, even one that's currently invisible via Visible property, still consumes memory and gets evaluated on every relevant property change, so screens with hundreds of controls (a common anti-pattern from over-using galleries-inside-galleries) get sluggish regardless of delegation. Large images embedded directly in the app or loaded at full camera resolution from a data source bloat both load time and memory; compressing images before upload and preferring SVG icons over bitmap images for UI chrome keeps the client light. Collections built with ClearCollect() live entirely in client memory, so caching 50,000 rows locally 'to make filtering fast' often backfires by making the whole app sluggish and memory-constrained on lower-end tablets.

🏏

Cricket analogy: It is like fielding all eleven players inside the 30-yard circle during the death overs; every fielder still has to react to the ball even though half of them are pointlessly bunched up doing nothing useful.

Avoid ClearCollect() with the full contents of a large data source just to enable client-side filtering. It defeats delegation entirely, loads the whole table into device memory, and on lower-end tablets or phones can crash the app outright. Filter server-side with delegable Filter()/Search() calls instead, and only collect the smaller, already-filtered result if local caching is genuinely needed.

Delegation pushes filtering/sorting to the data source; non-delegable formulas silently cap results at 500 (default) or up to 2000 records.
Different connectors delegate different function sets — SharePoint delegates fewer functions than Dataverse or SQL Server.
Use Concurrent() to run independent OnStart/OnVisible operations in parallel instead of sequentially blocking app launch.
Move screen-specific data loading into that screen's OnVisible rather than the global App.OnStart.
Every control consumes resources even when Visible = false; avoid excessive nested galleries and unnecessary controls.
Compress images and prefer SVGs for icons; large embedded or full-resolution images bloat load time and memory.
Use the Monitor tool to trace real network and formula timings instead of guessing which formula is slow.
Avoid caching entire large tables with ClearCollect(); it defeats delegation and can exhaust device memory.