-
Notifications
You must be signed in to change notification settings - Fork 197
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
Marking of 'suppress_redundant' and 'heartbeat_interval' for deprecation #114
base: master
Are you sure you want to change the base?
Conversation
Hi @dplore @robshakir and others, I wanted to revive this topic with a focus on the @earies I wonder when you refer to
what solutions you had in mind? |
I think Ebben is suggesting a mechanism such as gRPC keepalive would be more efficient than using As far as deprecating existing gNMI specification, we should understand is what dependencies or use cases exist regarding Ebben, regarding a "data refresh" use case (presumably one use case for |
Apologies I must have missed the activity on this PR but want to loop back on the topic. While the PR contents are somewhat stale, it's still probably the best spot to continue discussion for now... Regarding liveliness - yes I was referring to gRPC keepalives for that problem statement (essentially at a lower layer in the stack) Regarding
Are both sufficient? Possibly - would like to hear opinions as it's obvious this brings an additional set of subscriptions into the mix... Otherwise we keep I'll leave |
I think there are some big operators that have plans to use |
Well I'd still currently put this in the bucket of "not well defined" even if the desire and perceived use-case is there the more I think about this. An However let's consider the As a client, there are 2 choices to marshal a timestamp to the ultimate TSDB record. Time of reception or the timestamp embedded in the payload w/ the latter generally being the recommended truthful of the two (the true source). In normal
There is also the consideration of race conditions from the implementation should actual change events occur during the same window in which the |
I don't have a strong thought on whether this should be removed, other than if there are folks that are using it -- let's please make sure that the specification is robust to the deployed cases.
In the case that someone designs a caching collector that avoids memory exhaustion by timing out records over age X, then I think there are cases where |
As has been discussed prior OOB, I would like to initiate this PR to kick off
RFC regarding deprecation and possible future removal of the
suppress_redundant
andheartbeat_interval
fields in favor of alternateoptions to various problem statements of determining liveliness.
NOTE: This PR is incomplete within this repository as well as the potential
removal from the openconfig-telemetry YANG schema that can be determined after
consensus for deprecation/removal.
Related problem statements:
volume subscriptions)
FW, etc..)
Should the instruction of a 'keepalive' exist at the RPC/message layer?
If so, then does it make sense to duplicate work in the RPC design everytime
these characteristics are shared?
There are various other services related to OpenConfig/P4 that share similar
characteristics but do not implement RPC level instructions for keepalives
Is it best served by the protocol stacks - gRPC/HTTP2 infra (keepalive/PING)?
If so, are all cases and language bindings accounted for where this would
suffice?