Loading data
Three ways to give the grid its rows, all rows at once, page by page as the end nears, or only the window in view.
The grid never fetches: it tells you what is in view, and you decide what to load and how to keep it. Fetching, caching and placeholders are your app's.
All the rows
Pass rows, an array. Append to it (a new array) and the grid renders the new rows, keeping the
scroll position.
Sorting, filtering, searching and paging the rows you hold take one hook:
Rows in memory. Selecting them is yours: the
companies example sorts through the hook, and selects and deletes its
own way.
Page by page
onRowsEndReached fires when the view's last row comes within endReachedThreshold rows of the
end (10 by default), once per row count. Load the next page there and append it.
Only the window
For a dataset you cannot load in full, pass rowCount and getRow(index). The scrollbar
represents every row from the start; getRow answers the row, or undefined while it is not
loaded. A row not loaded keeps its place and renders with data-loading, so you can draw a
placeholder.
Load what onRowWindowChange reports. When rows arrive, tell the grid which with
rows.changed, through a gridRef:
getRow stays the same function, and the grid renders again only when the rows that arrived are
on screen. Rows that land far from the view cost nothing.
const gridRef = useDataGridRef<Person>();
// the cache tells which rows each block brought
useEffect(
() =>
cache.subscribe((range) =>
gridRef.current?.model.run("rows.changed", range),
),
[cache, gridRef],
);
<DataGrid.Root
rowCount={1_000_000}
getRow={cache.get}
gridRef={gridRef}
onRowWindowChange={({ rendered }) => cache.ensure(rendered.start, rendered.end)}
…
/>rows.changed takes { start, end } (end excluded), either bound alone, or nothing for every
row. Passing a new getRow still works too, and renders the grid each time.
The same call serves data that changes all the time, as a live feed does: batch the moves and
tell the grid their range once a frame. It renders the rows on screen again, once a frame
whatever the rate, and nothing for the rows that moved out of view. The
live-market example ticks a thousand rows up to ten thousand times a
second that way, its summary row following with summary-rows.changed.
Wait for the scroll to settle before loading (a short timeout): a thumb dragged across the
dataset then asks only for where it stops. See the windowed-loading
example, and windowed-columns for the same with columns, by tiles.