#Grafana timeslice not working Patch#
#Grafana timeslice not working password#
#Grafana timeslice not working series#

It isn’t clear to me whether you don’t understand: We don't add features merely because one user asserts it's useful to them, features cost us maintenance and cost our users cognitive load.” If you can't convince me that this has a valid use case that makes sense both operationally and statistically, it has extremely little chance of gettingĪdded. Quantile_over_time(0.9,jmx_memory)” will throw an error (“expected type vector in aggregation expression, got matrix”) because it does not support timeseries (matrix) arguments.įixing that seems like an overall better approach so that you can treat vectors and matrixes equally in aggregate queries.

I would like to calculate the p90 JMX memory usage across the latest values for a time-slice of the data and then apply a filter to them in a single query. Suppose, I have a load of gauge metrics across a fleet of 50 hosts, collected every 15 seconds. Flow chart showing the syntax of the LATEST ON keyword.

#Grafana timeslice not working series#

The benefit of doing it is a much more natural and powerful query interface all of the xxx_over_time functions can be deprecated and the facilities available to time series become much more powerful. I'd like to understand whether there is a philosophical reason this has not been done already it seems that all of the existing aggregations would apply to time series equally as they do to the existing 'instant's. However, as I was adding additional functionality to this I came to the realization that what I was doing was essentially rewriting the aggregation logic which already exists but only supports instant points.įrom what I've been able to tell, getting the aggregation functions to support timeseries is not that hard, although all 'instant' readings need to be silently upconverted to time series with a single entry.

#Grafana timeslice not working Patch#

Since there is nothing in the APIs to do this, I looked into writing a function to do the timeseries aggregations which I could provide a patch for.

#Grafana timeslice not working password#

That's annoying since that's typically the data that I find useful from a dashboarding perspective if a single host is running slow that is not interesting as long as the overall service performance remains within the latency bounds I've outlined. Start by selecting Add Data Source and choosing the PostgreSQL option in the SQL group: In the configuration screen, supply the Host, Database, User, and Password for your Timescale Cloud. For example there is no way to say 'give me an average over this time series by metric name, but combine all the hosts together in a single dataset.' However, these functions don't allow for aggregations within the time series. The latest version of prometheus provides the xxx_over_time functions (quantile_over_time) in order to do this. Nevertheless, a YIELDC would be cleaner, so give some thought to whether you think it would be worthwhile. Or maybe using a conditional backwards GOTO or something (making sure to clear the jump condition). TimeSlice small and them doing something that would trigger a yield. I've been looking into methods to do time series aggregations in prometheus where I, for example, what to work out a quantile over a time range. I was thinking not in terms of having a YIELD but of making.