Inconsistent data in dashboard

(Robert McIntosh) #1

Is anyone else noticing some inconsistencies in the data used in the dashboard?

I’ve been exporting data on a regular basis to keep an eye on trends. I noticed that the data used by Discourse in my dashboard, in particular for the page view trends, seems wrong.

Page views:
According to the dashboard, the total in the last 30 days is 73810 - which tallies EXACTLY with the traffic numbers exported for the last 30 days, excluding today.

However, it tells me that the trend is down: “-6% change. Was 78298 in the previous 30 day period”

… except it wasn’t. The same data export for the previous 30 days was actually 70501

It made me search to see if I could actually arrive at that figure and I discovered that I can get exactly the figure used (78298) by including the previous 32 days of data. Is there a reason for this discrepancy? Does the same happen for other trend reports?

I’ve also looked at the counts and trends for posts & threads.

The figures for 30 day periods do appear to tally, but in contrast to page views, it seems that the 30 day count for these does include ‘today’ (at the point of analysis).

This is more complicated. It appears that the latest ‘30 days’ data is actually 29 days, including today. However, the figure used for the ‘previous 30 day period’ (960) cannot be replicated either by adding the previous data for 29 or 30 days (it is in fact larger than the sum of the previous 32 days which is 925), and doesn’t match the previous calendar month either. I have no idea where this figure comes from.

I’ve not gone and tested the other figures, but since the dashboard reversed the small increase I expected in total monthly page view, this particular issue jumped out at me.

@HAWK - is this something you have looked into with regard to the new dashboard?


Nope, I wasn’t aware of it happening.

@eviltrout is making my designs a reality so it is probably something to put on his radar.

That said, @sam digs into these stats the most. Any idea what’s happening here?

(Sam Saffron) #3

I am not sure. We are going to address this with the new dashboard as @HAWK mentioned. Perhaps @neil could have a quick look, originally the trend calculations were done by him.

Note, I would expect page views to be way down due to: Smarter handling of random crawler traffic

(Neil Lalonde) #5

I made some fixes to those reports and numbers.

The start date varied from 28 to 31 days ago, so the “Last 30 Days” columns could have been counting fewer than 30 days. The end dates were calculated based on this wrong date, so could also be off. That would throw off the “previous 30 day” period.

I also fixed a problem where the individual report pages weren’t being strict about using UTC days, so a day might be defined differently there, although it probably didn’t affect sites installed with Docker.

(Robert McIntosh) #6

thanks - I will do some checking to see the differences soon

appreciate it