Hi all, I am looking for the best practice how to aggregate measurements of ever-increasing counters, such as bits or packets trasmitted/received (aka SNMP IF-MIB::ifXTable), extracting as much useful data from the measurements as possible.
Let’s say I read the counters every minute, and I want to create a 5-minute aggregate with average and peak traffic. The problem is that continuous aggregates apparently cannot see the previous time bucket with the lag() function. It is discussed in the docs:
The suggested solution there uses the first() function. However, this approach selects only one value from each bucket, so it is useful only for computing average traffic, but not for peak traffic: the peak can occur either between the two measurements in the same time bucket, or between the last measurement of the previous bucket and the first measurement of the current one.
A similar problem is present even when calculating average traffic, when there is a counter reset recorded in the measurements: then we have several measurements which can be used for average traffic calculation, and two neighboring measurements which can not, because the counter difference is negative. And again, these two can either appear inside the same bucket, or as the last measurement of the previous time bucket and the first measurement of the current one. But as a whole, the average traffic estimate can be computed.
Is it possible to compute average and peak traffic using TimescaleDB continuous aggregations? If so, how?
Thanks,
-Yenya
2 posts - 2 participants