Conversation
Pull latest changes from the master
Pull from envoy:master
63a8528 to
9665d9e
Compare
load_report.proto Signed-off-by: Vishal Powar <vishalpowar@rodete-desktop-imager.corp.google.com>
9665d9e to
8c74fb2
Compare
merge from envoy master
There was a problem hiding this comment.
Can you comment on examples of this will be used; e.g. what would Envoy put in this? What is the key space for the metadata?
There was a problem hiding this comment.
Ideally this would contain the same information that we populate in LbEndpoint.metadata.
Thinking of this as a custom response to the Load Balancer for the Endpoint. e.g. if Load Balancer is using some sort of 'affinity', it would pass down it down to envoy via LbEndpoint.metadata. And should be copied back in the report to the balancer, for some correlation from Assignment to Load Report at the balancer.
That said, I agree that this is not an exact definition of how this field will be populated.
I will remove it.
There was a problem hiding this comment.
I think that Endpoint has a lot of information beyond just the endpoint identifier these days, even though it started of as a simple identifier. I think core.Address is the only actual identifier in there. Should we use that instead?
There was a problem hiding this comment.
Yes, core.Address is enough. Will change
There was a problem hiding this comment.
Where did we end up with on the "let's eventually deprecate cluster granularity stats and have the management server aggregate the endpoint stats"?
There was a problem hiding this comment.
I just caught up on this, ignore this comment.
There was a problem hiding this comment.
I thought we did not want to disable a scenario where Server (Balancer) might not need per endpoint granurality information. In which case we can avoid sending all the Endpoint information to the Server.
To make this work I have added the following in LoadStatsResponse below
bool report_endpoint_granularity = 3;
There was a problem hiding this comment.
FWIW I think it makes sense to make this be configurable. If users don't want all the extra data there seems no reason to spend resources serializing it.
67461c6 to
b95a0ef
Compare
htuch
left a comment
There was a problem hiding this comment.
LGTM modulo tiny nit. Is everyone else good with this? @mattklein123 ?
There was a problem hiding this comment.
Nit *true* to match Envoy docs style.
load_report.proto Signed-off-by: Vishal Powar <vishalpowar@rodete-desktop-imager.corp.google.com>
b95a0ef to
1756646
Compare
For an explanation of how to fill out the fields, please see the relevant section
in PULL_REQUESTS.md
Description:
Risk Level: None
Testing: None
Docs Changes:
Release Notes:
Fixes: #4036