Node availability report based on time frame(s) and avoiding summarization of detailed stats

jwitty

I would like to create a report that utilizes raw (detailed stats) data to calculate the available uptime for a group of nodes (for this example, its a group created for our Windows servers).  When I currently create this group and apply it to the node availability report for, say, last 3 months, it shows 100%.  Even though there are a few Windows servers that over that time period were not up 100%.  My question is, how do the reports calculate this group uptime and is there a way to use the detailed data on this vs summarized data.  For settings purposes, we have our environment configured to keep detailed data for 7 days.  So does that mean the daily detailed stats are not long kept in the database once they are summarized?  Is there no way to get an accurate availability report for 30 days, 90 days etc. based on this?  I already select the "don't summarize" option within the report table data view but I am guessing my 7 day detailed polling setting will be true limitation, correct assumption?  Thanks for any feedback.  

weston.hood

You are correct, detailed data is kept for 7 days and aggregated, and therefore a loss in uptime accuracy can occur beyond 1 week. Node Availability is not necessarily the percentage of time the node is up. It's the percentage of polls that it was up, if that makes sense. Availability of 100% = node was up that poll and 0% = node was not up that poll. And those get recorded every 2 minutes and held for 7 days. After 7 days, every single polled data point is summarized into hourly averages, held onto until it reaches 30 days old. At which point its averaged into daily averages, then held onto for 365 days where it's deleted. So "group availability" is more complicated. Its based on your group's status rollup setting. But "group members availability" should just be an average of all the individual members of the group, regardless of rollup. If you choose "best status" or "mixed status" rollup I believe it just needs 1 member up to be 100% available (all others can be down, but as long as 1 is up, its 100% available) - think of this as an availability pool. If you choose 'worst status' It needs all members to be up to be 100% available. At least, this is what I'm observing by testing with a test group within perfstack. Changing the status rollup has a direct impact on group availability.

Two things that can throw off your availability reporting are 1-unmanaging nodes or 2-dependencies. Either of those essentially pauses the polling, and so during periods of the parent object down or unmanage window, those periods are not counted towards the nodes overall uptime.

Getting back to retention - What I've seen many users do is increase the retention if it's important to you - of course, this will grow your database some. But since you're just interested in availability I'd say it's not worth increasing ALL detailed stats just for the one metric (because cpu/memory/etc get increased too when you increase this). What I would probably recommend is to either set up the Report Scheduler to automate emailing you a CSV for the past week of availability. Or alternatively, you could set up a SQL Job to just select this availability data into a new DB table where you hold extended availability metrics for every single poll. You should only need to copy from [ResponseTime_CS_cur] table.