Export linux PSI avgs in addition to total for finer granularity when scrape interval is higher
tomershafir opened this issue · comments
Linux PSI avgs may help better quantify time series memory pressure spikes for a higher resolution data points. For example, avg10 with a regular 1m scrape interval, or avg60 with 2m.
The procfs already reads entire PSI lines. If you agree I can contribute.
This is unnecessary as you can get the same results from rate(X[Ym]). With the counters all average values can be determined as long as the scrape interval is higher than the desired averaging.
@SuperQ isnt it the opposite? with 1m scrape interval storage will contain data points at 1m resolution, thus for a given time t you cannot avg last 10s because there are no data points
Ahh, sorry, I misunderstood your statement. You are correct, however the samples are nearly useless since they don't contain any of the intermediary data and are essentially random gauge values.
Even still, it makes no sense to ingest these values as cardinality (additional metrics) has a higher overall cost (indexing) than more frequent scrapes. Samples are greatly compressed and take up very little memory compared to the additional indexing overhead.
If you require more resolution simply increase your scrape interval to something more normal like 15s. It's very common to scrape the node_exporter at 10-15s.
This is not a feature we intend to support since the value is not there and the results are misleading.
well, I dont think its random value and an experienced user may properly tradeoff, however I can see how the calculation can mislead. thanks for the quick response!