NetSuite Map/Reduce: Why getInputData() Should Do Almost Nothing

The misunderstood stage

Most Map/Reduce documentation focuses on the map() and reduce() stages, where the actual processing happens. But getInputData() is where the most consequential design decisions are made — and where the most common performance mistakes occur.

getInputData() runs in a single execution context, before NetSuite distributes any work. Whatever happens in getInputData() cannot benefit from Map/Reduce's parallel processing. It runs once, sequentially, on one thread.

Its job is simple: tell NetSuite what records need to be processed. Return a data source, and let the framework take it from there.

What many scripts do instead

A common pattern uses getInputData() to run a search, loop through every result, and build an array before passing it to map():

function getInputData() {
    var data = [];

    search.load({ id: 'customsearch_orders' })
        .run()
        .each(function(result) {
            data.push(result.id);
            return true;
        });

    return data;
}

This works. But by the time map() starts, all the search execution has already happened in a single thread. The script has loaded and iterated every result before NetSuite has had any opportunity to distribute the workload.

If the search returns 10,000 records, getInputData() processes all 10,000 sequentially. Then map() distributes what's left. The stage that was supposed to benefit from parallelism has already done the heavy lifting before parallelism begins.

The correct approach: return a data source object

Instead of executing the search inside getInputData(), return the search object itself:

function getInputData() {
    return search.load({
        id: 'customsearch_orders'
    });
}

When you return a Search object, NetSuite's Map/Reduce framework handles execution and distribution. The framework fetches results in pages and hands them to map() workers in parallel. The script no longer pre-executes anything — it simply identifies what needs to be processed.

The same applies when using the N/query module. Return the Query object, not the executed results:

function getInputData() {
    return query.create({
        type: query.Type.SALES_ORDER
    }); // NetSuite handles execution
}

Why this matters for governance

getInputData() runs under the standard scheduled script governance limits. When you pre-execute a large search inside it, you consume governance units before any parallel processing begins.

By returning a Search or Query object instead, you defer execution to the Map/Reduce framework, which handles pagination and distribution more efficiently. The result is lower governance consumption in getInputData(), more work offloaded to map() workers, and a script that scales to larger datasets without hitting limits.

What should stay out of getInputData()

Beyond pre-executing searches, these patterns in getInputData() undermine the framework:

• Loading records to inspect them before building the work list — if you need to filter records, add the condition to the search filter, not to a loop in getInputData()
• Calling external APIs to build the input set — external calls block the single getInputData() execution; move them to map() where each worker handles its own call
• Performing business logic — any transformation or calculation that can be deferred to map() should be deferred

The simpler getInputData() is, the more work the framework can distribute.

The analogy

Think of Map/Reduce as a warehouse operation. getInputData() is the manager who creates the work order list. map() workers are the employees who execute each item on the list independently.

The manager's job is to write the list — not to start working through it before handing it off.

When getInputData() pre-executes the search, it is as if the manager processes half the orders before any workers have arrived. The work that was supposed to be distributed has already been done sequentially.

The rule

getInputData() should return one of:

• A Search object — search.load() or search.create()
• A Query object — query.create()
• A small, known array when the input set is genuinely fixed and small

It should not execute searches, load records, or perform business logic.

The best getInputData() functions are often the ones that do the least.