DistinctCount Approximation or Exact

[t-zhguan]

I'm connected to a Kusto cluster using DirectQuery and noticed that when I used DistinctCount to create a measure and visualize said measure, the results are sometimes off by +/- 1-2%. When I use visual filters to manually look at top/bottom n values, the actual number of distinct entries match the query results I get from Kusto.

Just wondering if the results from the distinctcount function is exact or approximate and if it's normal to expect some slight margin of error when using aggregate function on larger data sets through directquery.