-
Notifications
You must be signed in to change notification settings - Fork 106
Add more logging with traceID #1973
Comments
stating the obvious maybe, but this would double the logging volume. But I agree that this would be valuable enough, and many clusters don't have such a large request rate that this would be an issue.
We have to be careful with how much value we confer to such stats (whether total runtime we already have or the breakdown you want to add) Alternatively, if you seek back enough in the log, you should be able to infer which queries ran concurrently only by processing the log file. But then you'd also need to tie that back from the read node logs back into the query node's logs. Maybe I'm overthinking/idealizing things too much. I'm fine with your suggestion in general. |
However, in many cases the series/points fetched say quite a lot about the cost of the query. Using these in conjunction with the timings and the chunk pull stats is very useful. |
Before submitting a feature request please review other open requests and issues to see if someone else has already requested this feature.
Is your feature request related to a problem? Please describe.
Metrictank has good logging for incoming render requests, but it has a few shortcomings:
meta.RenderStats
logged on completionDescribe the solution you'd like
Add some logging of high-level render stats and maybe some other interesting cases and include the traceID where possible.
Describe alternatives you've considered
Having trace enabled is great for some of this, but it is fairly heavy-weight compared to logging and (depending on where you store your trace/logs) harder to search against than logs. Often, trace is sampled (we use 5% sampling). Supplementing with logs would be great.
Additional context
I put in #1972 with very basic details of what I'm talking about. I expect there might be some interesting things to add (possibly on the read nodes side as well).
The text was updated successfully, but these errors were encountered: