| tstats count where index=<your-index-here> earliest=-3d@d latest=now() by _time span=15m | eval date_wday=strftime(_time,"%A"), date_hourmin=strftime(_time,"%H:%M") | search date_wday!=Saturday date_wday!=Sunday | eval current_weekday=strftime(now(),"%A") | eval previous_working_day=case(match(current_weekday,"Monday"),"Friday",match(current_weekday,"Tuesday"),"Monday",match(current_weekday,"Wednesday"),"Tuesday",match(current_weekday,"Thursday"),"Wednesday",match(current_weekday,"Friday"),"Thursday") | table _time count date_wday date_hourmin current_weekday previous_working_day | where date_wday=current_weekday OR date_wday=previous_working_day | chart sum(count) as count by date_hourmin, date_wday Ok that is the closest I could get to what you originally tried. However, there are some flaws with this solution you may want to consider. Specifically partial time bins can not be filtered out without using the time chart command. So it could look like the count for the most recent time span has dangerously dropped when in reality you only have 2 or 3 minutes of the 15 minute window to measure. Working with the timewrap command is more correct way to do this as you can leverage timechart which allows you to disable partial windows. You will find though that filtering out week-ends and the -3d@d makes for odd visualizations. index=<your-index-here> date_wday!=saturday date_wday!=sunday earliest=-3d@d latest=+1d@d | timechart span=15m partial=f count | timewrap 1day align=end Splunk time extracts date_* fields for you already. The +1d@d is only important if you want your graph to go midnight to midnight, replace with now() if you are ok with the visualization start and end moving as the day progresses.
... View more