-
Notifications
You must be signed in to change notification settings - Fork 84
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Indexer-less measure processing in metricd #548
Comments
this seems like a potentially useful performance optimisation or just fixing list_metrics in general. when writing to local disk, list_metrics in process_new_measures can take up 1/3 of the runtime. this is probably less significant if storage is remote but 1/3 of the time seems strange since it's just listing a single metric... granted we're talking single digit millisecond. attached is a profile pushing 300 datapoints to metric with low policy |
There's also an upside to this, is the ability for metricd to still work even if the indexer is down. |
did you have a plan for this?
i guess in both cases it gets kinda funky when you start updating timespan of policy |
My think is rather to dump archive policy requirement before doing this, i.e. #547. Once this one is done, each metric needs to store its own aggregation definition and the question is where to store it. If you store it in the storage engine, then you solve this issue. I don't have more details for now so there might be blockers (?). |
Currently, metricd
MetricProcessor
needs an indexer connection for the only reason of accessing archive policy information for each metric.It seems that metricd could not use an indexer connection at all if the archive policy content was duplicated to the metric in the aggregate storage.
This has other implication, especially for metrics using archive policies.
However, in a world where archive policies would be gone (check #547), that might be feasible.
The text was updated successfully, but these errors were encountered: