skill-performance

name	skill-performance
description	AL performance optimization patterns for Business Central. Use when optimizing queries with SetLoadFields, working with FlowFields and CalcFields, profiling codeunits, or resolving performance issues.

Skill: AL Performance Optimization

Purpose

Identify, analyze, and fix performance bottlenecks in AL code: inefficient queries, FlowField issues, loop anti-patterns, and data-volume problems in Business Central.

When to Load

This skill should be loaded when:

• A page, report, or batch process is slow or timing out
• CPU profiling is needed to identify hotspots
• A static triage of the codebase is requested for performance issues
• A feature involves large-dataset processing or high-frequency code paths
• AL0896 (circular FlowField) errors appear
• Architecture review requires performance analysis of a new design

Core Patterns

Pattern 1: SetLoadFields + Early Filtering

Always filter before finding, and load only needed fields. Order matters.

// ✅ Correct: SetRange first, SetLoadFields before Find
Item.SetRange("Third Party Item Exists", false);
Item.SetLoadFields("Item Category Code", Description);
if Item.FindSet() then
    repeat
        // Only "Item Category Code" and Description loaded from DB
    until Item.Next() = 0;

// ❌ Wrong: SetLoadFields after SetRange (ignored), loads all fields
Item.SetLoadFields("Item Category Code");
Item.SetRange("Third Party Item Exists", false);
Item.FindFirst();

// ❌ Wrong: No filter — full table scan
procedure GetCustomersByCity(CityFilter: Text): Integer
var
    Customer: Record Customer;
    Count: Integer;
begin
    if Customer.FindSet() then       // loads entire Customer table
        repeat
            if Customer.City = CityFilter then
                Count += 1;
        until Customer.Next() = 0;
end;

// ✅ Correct: Filter pushed to DB
procedure GetCustomersByCity(CityFilter: Text): Integer
var
    Customer: Record Customer;
begin
    Customer.SetRange(City, CityFilter);
    Customer.SetRange(Blocked, Customer.Blocked::" ");
    exit(Customer.Count());
end;

Pattern 2: Set-Based Aggregation (CalcSums / CalcFields)

Avoid manual loops for aggregation — push the sum to the database.

// ❌ Loop accumulation — N rows fetched and processed in AL
procedure GetTotalSales(CustomerNo: Code[20]): Decimal
var
    Entry: Record "Cust. Ledger Entry";
    Total: Decimal;
begin
    Entry.SetRange("Customer No.", CustomerNo);
    if Entry.FindSet() then
        repeat
            Total += Entry.Amount;
        until Entry.Next() = 0;
    exit(Total);
end;

// ✅ CalcSums — single aggregation query at DB level
procedure GetTotalSales(CustomerNo: Code[20]): Decimal
var
    Entry: Record "Cust. Ledger Entry";
begin
    Entry.SetRange("Customer No.", CustomerNo);
    Entry.CalcSums(Amount);
    exit(Entry.Amount);
end;

For FlowFields accessed outside a page context, always call CalcFields before reading:

Customer.SetLoadFields("Balance (LCY)");
Customer.Get(CustomerNo);
Customer.CalcFields("Balance (LCY)");   // required — not auto-calculated in code

Pattern 3: Temporary Tables / Dictionary / List

Pre-load data once, then process in-memory multiple times.

// ✅ Temporary table — structured record data, multi-pass processing
procedure ProcessSalesData(var TempSalesLine: Record "Sales Line" temporary)
var
    SalesLine: Record "Sales Line";
begin
    SalesLine.SetLoadFields("No.", Quantity, "Unit Price", Amount);
    if SalesLine.FindSet() then
        repeat
            TempSalesLine := SalesLine;
            TempSalesLine.Insert();
        until SalesLine.Next() = 0;

    // Process in-memory — zero additional DB hits
    ApplyDiscounts(TempSalesLine);
    CalculateTotals(TempSalesLine);
end;

// ✅ Dictionary — key-value lookup cache
procedure CacheCustomerNames(): Dictionary of [Code[20], Text]
var
    Customer: Record Customer;
    Cache: Dictionary of [Code[20], Text];
begin
    Customer.SetLoadFields("No.", Name);
    if Customer.FindSet() then
        repeat
            Cache.Add(Customer."No.", Customer.Name);
        until Customer.Next() = 0;
    exit(Cache);
end;

// ✅ List — simple value collection
procedure GetBlockedCustomerNos(): List of [Code[20]]
var
    Customer: Record Customer;
    Result: List of [Code[20]];
begin
    Customer.SetRange(Blocked, Customer.Blocked::All);
    Customer.SetLoadFields("No.");
    if Customer.FindSet() then
        repeat
            Result.Add(Customer."No.");
        until Customer.Next() = 0;
    exit(Result);
end;

Pattern 4: Batch Processing and Avoiding Nested DB Calls

Move database operations outside loops; batch writes.

// ❌ Nested DB call inside loop — O(n²) database hits
if MainTable.FindSet() then
    repeat
        OtherTable.SetRange(Field, MainTable.Field);
        if OtherTable.FindSet() then          // DB call per iteration
            repeat
                // process
            until OtherTable.Next() = 0;
    until MainTable.Next() = 0;

// ✅ Pre-load into temp table, then join in memory
if OtherTable.FindSet() then
    repeat
        TempOther := OtherTable;
        TempOther.Insert();
    until OtherTable.Next() = 0;

if MainTable.FindSet() then
    repeat
        if TempOther.Get(MainTable.Field) then
            // process — no DB hit
    until MainTable.Next() = 0;

Batch writes — collect changes, write once:

// ✅ Calculate all values first, then single Modify
procedure UpdateCustomerStats(CustomerNo: Code[20])
var
    Customer: Record Customer;
    TotalBalance: Decimal;
    LastPaymentDate: Date;
begin
    CalculateCustomerTotals(CustomerNo, TotalBalance, LastPaymentDate);

    Customer.SetLoadFields("Balance (LCY)", "Last Payment Date");
    if Customer.Get(CustomerNo) then begin
        Customer."Balance (LCY)" := TotalBalance;
        Customer."Last Payment Date" := LastPaymentDate;
        Customer.Modify(true);   // single write
    end;
end;

Scale-aware processing:

procedure UpdatePricesForItems(var Item: Record Item)
begin
    if Item.Count() > 1000 then
        UpdatePricesInBatches(Item)    // job queue / chunked
    else
        UpdatePricesDirectly(Item);
end;

Pattern 5: FlowField Optimization (AL0896)

Circular FlowField references cause infinite evaluation (AL0896 error):

❌ Circular dependency:
Table Customer: FlowField "Total Sales" → CalcFormula from Sales Statistics
Table Sales Statistics: FlowField "Customer Balance" → CalcFormula from Customer

Compiler error AL0896: recursive dependency detected

Resolution strategies (in order of preference):

1. Break the chain — convert one FlowField to a regular field updated via trigger
2. Use CalcSums — replace FlowField with explicit CalcSums in code
3. Restructure — move the calculation to a dedicated codeunit called on demand
4. SumIndexFields — use SIFT keys for frequently summed values

Avoid FlowFields:

• With complex nested CalcFormula expressions
• Called inside repeat...until loops (call CalcFields once, outside loop if possible)
• Where the source table is large and unfiltered

Workflow

Step 1: Triage (Static Code Analysis)

Scan the codebase before profiling to identify structural issues:

Patterns to detect manually or with search + problems:

• FindSet() / FindFirst() without preceding SetRange / SetFilter
• SetLoadFields placed after SetRange (wrong order)
• Database calls (Get, FindSet, FindFirst) inside repeat...until
• CalcFields inside loops
• FlowField CalcFormula referencing tables that reference back (AL0896)
• Commit inside loops (locks + performance risk)
• Missing keys for columns used in SetRange

Severity assessment:

• 🔴 Critical — DB call in loop over large table, circular FlowField, timeout-causing query
• 🟡 High — Missing SetLoadFields in high-frequency path, nested loops
• 🟢 Medium — Suboptimal aggregation (loop instead of CalcSums)
• 🔵 Low — Best-practice suggestion, minimal current impact

Step 2: Profile (Runtime Measurement)

CPU profiling is a VS Code AL command / human step — there is no agent tool for it on this surface. Ask the human to capture a CPU profile (VS Code AL profiler) and share the result, then analyze it.

Analyze the profile for:

• Hotspots — procedures with highest cumulative execution time
• Frequency — procedures called most often (especially in loops)
• DB operations — expensive FindSet / Get calls
• FlowField evaluations — unexpectedly costly CalcFields

Compare before/after optimizations by re-profiling after each fix.

⚠️ Human Gate — cleanup: Before the human clears the profile codelenses in VS Code, confirm all findings are documented (clearing codelenses is a VS Code action, not an agent tool here).

Step 3: Fix

Apply fixes in priority order (critical first). For each fix:

1. Apply targeted change (Pattern 1–5 above)
2. Rebuild: Bash: al compile (al workspace compile for a multi-project workspace)
3. Re-profile to verify improvement

Performance targets:

Metric	Target
Page load	< 2 seconds
Report generation	< 10 seconds (standard dataset)
API response	< 500 ms
Batch processing	> 1,000 records/minute

Step 4: Document Findings

For significant optimizations, create a triage report at .github/plans/perf-triage-<scope>.md:

# Performance Triage — <Scope>

**Date**: YYYY-MM-DD
**Analyzed**: [path or objects]

## Findings

| Severity | Location | Issue | Recommendation |
|---|---|---|---|
| 🔴 | `File.al:42` | Nested DB call in loop | Pre-load into TempTable |
| 🟡 | `File.al:87` | Missing SetLoadFields | Add SetLoadFields("No.", Name) |

## Changes Applied

1. [Change] — before/after metric
2. [Change] — before/after metric

## Remaining Items

[Issues not yet fixed, prioritized]

⚠️ Human Gate — report: Review findings before saving; confirm no sensitive code patterns are exposed.

References

• SetLoadFields — Microsoft Docs
• Performance Best Practices for AL
• AL0896 — Circular FlowField
• SumIndexFields (SIFT)
• CPU Profiler

Constraints

• This skill covers analysis and fix patterns — it does NOT duplicate the passive rules in al-performance.md (auto-applied to all .al files)
• Do NOT modify source code automatically during triage — generate report and get approval first
• Do NOT clear profile codelenses without human confirmation
• Do NOT save triage report without human gate review
• For runtime debugging of a specific issue (breakpoints, snapshots) → load skill-debug.md
• For event subscriber performance issues → load skill-events.md