Last 10 rows on big measurement

I've an issue since the migration to influxdb 2. We have ameasurement with thousands of data by day and I need to know the last 10 rows in a month.
So previously, we have the following query : select * from payload where time > now() - 4w and time < now() order by desc limit 10

Now we translated it with :
from(bucket: "mybucket")
|> range(start: -4w, stop: now())
|> filter(fn: (r) => r._measurement == "payload")
|> pivot(rowKey:["_time"], columnKey: ["_field"], valueColumn: "_value")
|> group()
|> sort(columns: ["_time"], desc: true)
|> limit(n: 10)

But the time is very long !
It seems to be the "sort" function is very expensive.

Have you an idea to get my results faster ?

Start by reproducing the Flux query over a large measurement, then examine the range, pivot, group, sort, and limit stages to identify where execution time is spent. Compare the query’s behavior with the reported legacy query and measure any candidate change; done means the last 10 rows for the requested period are returned with a demonstrated performance improvement.