NRQL queries that use TIMESERIES and LIMIT now sort results randomly. This seems to be new today. I had a query for some time that would make a timeseries of hourly transaction counts for the top 5 apps. This allows me to see which of our 25+ sites is serving more requests than others. Starting today, the five chosen seem to be random. That forces me to expand the LIMIT to include all apps too see the true picture, and it’s a lot harder to see what’s spiking due to all the lines.
Here’s the NRQL.
SELECT COUNT(*) AS 'Hits' FROM Transaction WHERE name NOT LIKE '%monitor.php' AND host NOT LIKE 'staging%' AND appName NOT LIKE 'SVR_%' AND appName NOT LIKE 'ENV_%' FACET appName SINCE 1 week ago TIMESERIES 1 hour LIMIT 100
I believe until today it was sorting on the COUNT(*) descending, perhaps by the last datapoint for each row. If this is not a bug that will be fixed, then I’d propose an ORDER BY clause that allows specifying by name or column order as can be done with SQL (e.g. ORDER BY appName or ORDER BY 1).