Join GitHub today
GitHub is home to over 31 million developers working together to host and review code, manage projects, and build software together.
Sign upMalformed increse/rate results on data coliding with compaction #4094
Comments
This comment has been minimized.
This comment has been minimized.
|
Delta is for gauges, not counters. You want increase. |
brian-brazil
closed this
Apr 17, 2018
This comment has been minimized.
This comment has been minimized.
|
Ok thank's for the quick reply I know that I'm supposed to use increase not delta I'm using the increase/rate the delta was used just for debugging. If you look at the first and third figures in first image those are raw counter and increase. The increase was about 3000(img1 figure1) and result of increase function is 20M(img1 figure3)? I used the delta just to illustrate what caused the raise in the increase. |
This comment has been minimized.
This comment has been minimized.
|
Pulling per-instance data via federation is not a good idea. If you believe you have found a bug, you should show it on the original scraping Prometheus and share the raw data for a single time series with the problem. |
This comment has been minimized.
This comment has been minimized.
|
Yes, point taken. I know it's not suggested due to the staleness and performance issues as you state. I'll try to reproduce this on the Prometheus instance directly scraping the data. But fyi the metric has not been reset for whole week. This should sort out problems with the staleness imho. |
This comment has been minimized.
This comment has been minimized.
qsbao
commented
May 2, 2018
|
@brian-brazil We got the same problem, I can show you the raw data.
Environment
Node's scrape interval is 20s. |
This comment has been minimized.
This comment has been minimized.
|
That's not the raw data, and that looks like a counter going backwards. Check with resets() |
This comment has been minimized.
This comment has been minimized.
qsbao
commented
May 2, 2018
|
Yes, you are right. There is a counter resets. Thank you! |
This comment has been minimized.
This comment has been minimized.
qsbao
commented
May 2, 2018
|
Hi Brian, resets() shows there are counter resets happens. But:
|
This comment has been minimized.
This comment has been minimized.
|
A break in monotonicity is presumed to be a reset back to zero due to a process restarting, which if that's not the case you get what you're seeing. There's something broken with your counter that you need to fix. |
This comment has been minimized.
This comment has been minimized.
lock
bot
commented
Mar 22, 2019
|
This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs. |



FUSAKLA commentedApr 17, 2018
Bug Report
What did you do?
Nothing
What did you expect to see?
Normal increase similar to other instances of application
What did you see instead? Under which circumstances?
I have one instance of Prometheus(1.8.0) scraping data (scrape interval 1m). This instance is federated by two other Prometheis (one 1.8.0[scrape interval 1m] second 2.2.1[scrape interval 20s]).
Out of nothing strange spikes in data appeared in graphs using increase/rate but it happens only for one scrape target. It showed up during weekend and there were no changes during that time. This happens only on the 2.2.1 Prometheus instance.
Those spikes are caused by the rate breaks in monotonicity adjustment because delta shows some weird data changes which are not present in pure data (see picture below). I don't understand where does those changes in delta result coming from? Those data points are more often than 20s which is scrape interval of the queried Prometheus instance.

Also those weird fuzzy data occurrences does correlate with the compaction interval strongly (also the retention cutoff, WAT truncate and so on..) as shown on picture below

This happens only to the one instance of given application and happens to multiple of metrics exposed by this instance. I suspected colliding labels caused by some mis configured relabel config but I found none of those.
The last thing is that the other Prometheus instance federating those data (1.8.0) is ok.
Environment
System information: Official docker image running in Kubernetes
Prometheus version: 2.2.1
Logs:
There are just logs about running successful compactions no error.