You could win up to $50,000 building Splunk apps in the Splunk>Apptitude contest. Learn more »
I've got some performance data and I want to be alerted when the avg(total_requests) split by uri rises or drops by 10%. For now, you can aggregate a few hours worth of data in the average window.
I think this question may need more detail. However if I make a couple assumptions I can attempt an answer. (total_requests I find to be a somewhat confusing field name so Im going to call it 'request_count')
Assuming you are starting from a search that looks like
<some search> | stats avg(request_count) over uri
which would give you output like:
<code>uri avg(request_count) some/uri 12.3 some/other/uri 41.4 .... </code>
And you want your alert to fire if any of the rows rises by 10%...
And assuming you want the 'rises by 10%' to be based on a time range like comparing today to yesterday, then here's a way to do it.
<some search> | eval day = if(_time > now()-86400, "today", "yesterday") | chart avg(request_count) over uri by day
eval command puts a field called 'day' onto each event, which will be 'today' for all events today, and 'yesterday' for anything older than that.
(NOTE: timerange-wise you probably want to run this search over yesterday+today using -1d@d on the earliest side and +1d@d on the latest side)
Anyway, then the
chart command after the
eval gives you a table that looks like this:
<code>uri today yesterday some/uri 12.3 10.4 some/other/uri 41.4 10.5 .... </code>
Finally throw a
where command on the end of that and you can filter the results down to only the uri's that actually had a 10% increase today over yesterday, like so:
<some search> | eval day = if(_time > now()-86400, "today", "yesterday") | chart avg(request_count) over uri by day | where today > (1.1 * yesterday)
And there you go. If that search returns any results, then that's bad, and you want to email those results to somebody.
(If you'd prefer the results in the email were actually the full list of URI's today vs yesterday, you could take that whole where clause off the end, and instead put it in the 'custom alerting condition' when you set up the alert. )
looking for a counter that decreases 1 Answer
Viewing URLs by IP over time 2 Answers
Anyone splunking Akamai log events? 2 Answers