roblox-testing

name	roblox-testing
description	Testing, debugging, and profiling Roblox experiences — Developer Console, Output, logging discipline, pcall and assertion patterns, TestEZ unit tests, the MicroProfiler (client and server), Scene Analysis, Script Profiler, memory diagnostics, network debugging, Luau type checking, and connection cleanup. Use when something is broken, slow, or unreliable, or when setting up test workflows.
last_reviewed	"2026-06-17T00:00:00.000Z"

roblox-testing

Official sources (always check these for the latest):

• https://create.roblox.com/docs/en-us/studio/developer-console
• https://create.roblox.com/docs/en-us/performance-optimization/identify
• https://create.roblox.com/docs/en-us/performance-optimization/microprofiler
• https://create.roblox.com/docs/en-us/performance-optimization/scene-analysis
• https://create.roblox.com/docs/en-us/studio/optimization/memory-usage
• https://create.roblox.com/docs/en-us/studio/optimization/scriptprofiler

This skill is about finding and fixing problems, not just writing code. It focuses on the tools and habits that separate working experiences from broken ones.

When to use this skill

Activate when:

• Something is not behaving as expected (scripts, UI, physics, data, networking).
• Frame rate, memory, or server heartbeat is degrading.
• Setting up unit tests or reproducible test cases.
• Trying to isolate whether a bug is on the client or server.
• Debugging a live issue using the Developer Console.

Cross-reference:

• roblox-core/SKILL.md for services and script contexts.
• roblox-networking/SKILL.md for debugging remote flows and network ownership.
• roblox-datastores/SKILL.md for DataStore retry patterns and debugging data store errors.
• roblox-physics/SKILL.md for debugging physics ownership and sleep.
• roblox-npcs/SKILL.md for debugging NPC behavior.

The debugging mindset

1. Reproduce it. If you can't reproduce it, you can't fix it.
2. Isolate it. Remove systems until the bug disappears; the last thing removed is the cause.
3. Measure it. Use tools instead of guessing.
4. Fix one thing at a time. Verify the fix and add a regression test if possible.
5. Log defensively. Good logs make future debugging faster.

Logging discipline

Use print, warn, and error deliberately:

• print for normal diagnostics.
• warn for recoverable problems you should notice.
• error for programming errors that should stop execution.

Include context in log messages:

warn(string.format("[DataStore] Save failed for %d: %s", userId, tostring(err)))

Avoid logging secrets, player data, or PII.

pcall and assertions

Wrap fallible calls, especially cloud services, HTTP, DataStores, Marketplace:

local ok, result = pcall(function()
    return someService:DoSomething()
end)
if not ok then
    warn("DoSomething failed:", result)
end

Use assert for internal invariants that should never fail:

assert(config.MaxSpeed > 0, "MaxSpeed must be positive")

Developer Console

Open with F9 in-game or in Studio play mode.

Tabs:

• Log — client/server output, errors, warnings.
• Memory — categorized memory usage.
• Network — HTTP and service requests.
• Server Stats — heartbeat, ping, data ping.
• Script Profiler — record script CPU usage.
• MicroProfiler — capture server dumps.

Toggle Client/Server views to see which side emitted output.

Unit testing with TestEZ

The standard Roblox testing framework is TestEZ. It supports nested describe/it blocks, lifecycle hooks, async tests, and a rich matcher API.

TestEZ example:

local ReplicatedStorage = game:GetService("ReplicatedStorage")
local TestEZ = require(ReplicatedStorage.DevPackages.TestEZ)

local MathUtils = require(ReplicatedStorage.MathUtils)

describe("MathUtils", function()
    it("clamps values", function()
        expect(MathUtils.clamp(5, 0, 10)).to.equal(5)
        expect(MathUtils.clamp(-1, 0, 10)).to.equal(0)
        expect(MathUtils.clamp(11, 0, 10)).to.equal(10)
    end)
end)

TestEZ.TestBootstrap:run({ game.ReplicatedStorage.Tests })

For legacy in-Studio tests, TestService:Error(msg) can mark failures visibly in Studio output, but most projects should prefer TestEZ. The scripts folder contains a minimal TestRunner.lua shim only for environments where TestEZ is unavailable.

Luau type checking

Use the Luau language server (luau-lsp) or the Luau type checker in Studio to catch errors before runtime.

Best practices:

• Add type annotations to module exports and pure functions.
• Run luau-lsp analyze in CI or before committing.
• Enable --!strict for new modules.
• Do not over-type untyped engine APIs; prefer casts at boundaries.

Studio Debugger

Use Studio's built-in debugger for step-through debugging:

• Set breakpoints by clicking the gutter next to a line number.
• Run in Play mode and use the debugger controls to step over, into, or out.
• Inspect the call stack, local variables, and upvalues in the debugger panel.
• The debugger works in both client and server contexts when Studio runs both.

xpcall and debug.traceback

Use xpcall with debug.traceback to capture full stack traces from failures:

local ok, err = xpcall(function()
    riskyOperation()
end, debug.traceback)
if not ok then
    warn(err)
end

This is especially useful at top-level entry points, scheduled callbacks, and connection handlers where pcall alone discards the stack.

Connection cleanup

Leaked RBXScriptConnection objects are a common source of memory leaks and stale state. Prefer explicit cleanup patterns:

• Use a Maid/Janitor-style collector:

local Maid = require(path.to.Maid)
local maid = Maid.new()

maid:GiveTask(workspace.ChildAdded:Connect(onChild))
maid:GiveTask(RunService.Heartbeat:Connect(onStep))

maid:Destroy() -- disconnects everything

• Use :Once() for one-shot event handlers.
• Use Instance.Destroying to trigger cleanup when an instance is removed:

obj.Destroying:Connect(function()
    cleanup()
end)

• Always pair :Connect() with a matching :Disconnect() or destruction.

Prefer task over spawn/delay

The task library is more predictable than legacy spawn, delay, and wait:

-- good
task.wait(1)
task.delay(1, callback)
task.spawn(coroutineOrFn)

-- avoid
spawn(callback)
delay(1, callback)
wait(1)

task.defer is useful for yielding to the next resumption cycle without opening a new thread.

Log rate limiting and secrets hygiene

• Never log secrets, keys, user identifiers, or personal information.
• Sanitize values before logging:

local function sanitize(value)
    if typeof(value) == "Instance" and value:IsA("Player") then
        return "Player(" .. tostring(value.UserId) .. ")"
    end
    return value
end

• Rate-limit noisy logs to avoid flooding the console and network:

local lastLog = 0
local function throttledLog(message)
    local now = os.clock()
    if now - lastLog > 5 then
        lastLog = now
        warn(message)
    end
end

Network ownership

Incorrect network ownership causes jittery physics and replication lag. Visualize ownership and assign it explicitly:

-- Server
part:SetNetworkOwner(player)

-- Client-side visualization (debug only)
local ownershipLabel -- create BillboardGui or use DebugDraw

Use BasePart:GetNetworkOwner() to inspect ownership. Vehicles and held items should usually be owned by the controlling player.

DataStore retry pattern

Wrap DataStore calls with exponential backoff and jitter:

local function dataStoreWithRetry(fn, maxAttempts)
    maxAttempts = maxAttempts or 5
    for attempt = 1, maxAttempts do
        local ok, result = pcall(fn)
        if ok then
            return true, result
        elseif attempt == maxAttempts then
            return false, result
        else
            warn("DataStore attempt " .. attempt .. " failed; retrying...")
            task.wait(2 ^ attempt * 0.1 + math.random() * 0.5)
        end
    end
end

Always call DataStores from the server and validate serialization before saving.

Memory snapshot comparison

Use the Developer Console Memory tab or Scene Analysis to compare snapshots:

1. Capture a baseline snapshot in a stable state.
2. Play through the action that may leak (spawn/despawn enemies, open/close UI).
3. Return to the stable state and capture a second snapshot.
4. Compare PlaceMemory, Luau heap, and instance counts.
5. Investigate categories that did not return to baseline; look for unparented instances in Scene Analysis.

MicroProfiler

Open with Ctrl+F6 (⌘+F6) in Studio or the desktop client.

Use it to:

• Find frame-time spikes.
• Identify whether a bottleneck is script compute, physics, or rendering.
• Capture server dumps from the Developer Console.
• Add custom labels with debug.profilebegin/debug.profileend.

Key colors:

• Orange — worker thread (scripts, physics, animations) bottleneck.
• Blue — render thread bottleneck.
• Red — GPU wait / render complexity.

Scene Analysis

Available in Studio under Window → Performance Summary → Scene Analysis.

Views:

• Script memory — per-script Luau heap.
• Unparented instances — potential memory leaks held by scripts.
• Instance composition — counts by category.
• Audio/Animation memory — asset memory usage.
• Triangle composition — draw call breakdown.

Scene Analysis is a Studio UI tool; there is no public SceneAnalysisService API.

Script Profiler

Records CPU time per script. Use it when MicroProfiler points to scripts but you need to know which script.

Common bug categories

Symptom	Likely causes
Script silently fails	Missing pcall, error swallowed, wrong script context
Data not saving	DataStore called from client, non-serializable value, no pcall
Remote not working	Wrong side, handler not connected, argument mismatch
Lag spikes	Pathfinding every frame, too many particles, unbatched loops
Physics jitter	Wrong network ownership, assembly splits, conflicting constraints
NPCs stuck	Blocked path not recomputed, bad agent params, streaming issues
UI doesn't update	Property not replicated, wrong parent, layout order
Memory grows forever	Leaked connections, unparented instances, cached assets

Network debugging

• Check Network tab in Developer Console for HTTP/DataStore failures.
• Use Shift+F3 in-game for network debug stats.
• Distinguish network ping (round-trip time) from data ping (replication queue).
• Simulate latency/jitter with Studio Network Simulation (Alt+S).

Load time debugging

Measure load time:

local start = os.clock()
game.Loaded:Connect(function()
    print("Loaded in", os.clock() - start)
end)

Enable Print Join Size Breakdown in Studio Settings → Network to see the largest replicated instances.

Scripts

• scripts/TestRunner.lua — a minimal TestEZ fallback with nested suites, lifecycle hooks, matchers, async support, and TestService integration.
• scripts/Logger.lua — a structured logger with level filtering and guarded formatting.
• scripts/DebugDraw.lua — utility for drawing rays, points, and boxes in 3D for visual debugging.

How to proceed

1. Reproduce the issue reliably.
2. Check logs and errors in the Developer Console.
3. Determine client vs server scope.
4. Use MicroProfiler/Script Profiler for performance issues.
5. Use Scene Analysis for memory leaks and scene composition.
6. Add targeted logging or tests to confirm the fix.
7. Verify on low-end devices and with network simulation when relevant.
8. Compare memory snapshots before and after suspected leaks.

Reference index

• common-bugs-and-fixes.md
• debugging-tools.md
• performance-profiling.md
• testing-patterns.md