-
Notifications
You must be signed in to change notification settings - Fork 4.7k
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
[stats, xds] disable per-resource config update stats tracking, add per-API update_duration tracking (if necessary). #23979
Comments
@adisuissa could you no-stalebot this issue? |
Hi @stevenzzzz are you talking about this histogram? @adisuissa, is my understanding correct that we create only one |
Depends on the resource type. For CDS and LDS for example, there will be a single subscription. |
Thanks Adi, makes sense. |
yes. Instead, I think we should trace the performance at "per-XDS response level". |
I'm not sure if this claim is correct, IIUC the "per-xDS response level". For example, if Envoy has 3 routes, and only 2 are being constantly updated, then if we just update when was the last RDS update that came, we lose the information on when the other route was updated. |
I am talking about the update_duration etc, which I think is very expensive and not very useful to a cloud proxy. I think we can have an option to allow user to choose which to track: per-resource level or "per-response" level. |
Ah, thanks for the clarification. I read the title as "disable per-resource config update stats tracking", which I understood as more generic than just |
Title: disable per-resource update_duration tracking, add per-API update_duration tracking (if necessary).
Description: right now there is a histogram tracking per-resource update duration, i.e. how long it takes Envoy to consume and update a resource's config.
This stat is of less value when there are tons of resources in Envoy, what matters more is how long does Envoy take to load each discoveryResponse.
OTOH, histograms are known to consume lots of RAM and CPU.
We should allow users to disable per-resource update-duration tracking (with a runtime flag) and maybe add a per-API level update-duration tracking instead.
The text was updated successfully, but these errors were encountered: