Added support for Envoy slow start mode.

[tsaarni]

This PR adds support for slow start mode, described here in Envoy documentation. Envoy to gradually increase the amount of traffic targeted to a newly added upstream endpoint. This can be useful for example with JVM based applications, that might otherwise get overwhelmed during JIT warm-up period.

The feature is configured via struct HTTPProxy.spec.routes.services.slowStart:

apiVersion: projectcontour.io/v1
kind: HTTPProxy
metadata:
  name: slow-start
spec:
  virtualhost:
    fqdn: www.example.com
  routes:
    - services:
        - name: java-app
          port: 80
          slowStartPolicy:
            window: 3s
            aggression: 1.0
            minWeightPercent: 10

Slow start is off by default. To enable it, only window is mandatory. Other parameters are optional.

Fixes #2296
Fixes #4687

Signed-off-by: Tero Saarni [email protected]

[codecov]

Codecov Report

Merging #4772 (24ae5e2) into main (e4a0156) will increase coverage by 0.04%.
The diff coverage is 88.70%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4772      +/-   ##
==========================================
+ Coverage   75.98%   76.02%   +0.04%     
==========================================
  Files         140      140              
  Lines       16787    16849      +62     
==========================================
+ Hits        12755    12809      +54     
- Misses       3780     3788       +8     
  Partials      252      252              

Impacted Files	Coverage Δ
internal/envoy/v3/cluster.go	93.65% <73.07%> (-2.37%)	⬇️
internal/dag/dag.go	96.62% <100.00%> (+0.04%)	⬆️
internal/dag/httpproxy_processor.go	93.07% <100.00%> (+0.20%)	⬆️
internal/envoy/cluster.go	100.00% <100.00%> (ø)
internal/sorter/sorter.go	98.46% <0.00%> (-0.52%)	⬇️

[tsaarni]

Manual test procedure for this PR is described here https://gist.github.com/tsaarni/a22e3d69bf6c31ff0597e9f78454868a.

[skriss]

@tsaarni just a couple nits but overall this looks great and pretty straight-forward, thanks! I think we can include this in the upcoming 1.23 release.

[tsaarni]

Thank you for good comments @skriss!

@tsaarni just a couple nits but overall this looks great and pretty straight-forward, thanks! I think we can include this in the upcoming 1.23 release.

That would be great! :-)

BTW I see there is problem with actions currently but I will re-run/bump this PR to trigger the checks later, unless it sorts out itself automatically.

[skriss]

thanks for the updates @tsaarni, LGTM. I'll leave for @sunjayBhatia to take a look if he wants.

[sunjayBhatia]

interesting this is also configurable via RTDS, not sure if we will need to take advantage of that though

[tsaarni]

True, I did not see how to take advantage of it either, but it seems that (almost) all Runtime* message types have validation rules in .proto which makes the RuntimeKeys required. For example in this case RuntimeDouble.runtime_key has validation rule:

string runtime_key = 2 [(validate.rules).string = {min_len: 1}];

that causes config to be rejected if runtime key is not given:

[2022-10-06 06:03:23.608][1][warning][config] [source/common/config/grpc_subscription_impl.cc:126] 
gRPC config for type.googleapis.com/envoy.config.cluster.v3.Cluster rejected:
Proto constraint validation failed
(ClusterValidationError.RoundRobinLbConfig: embedded message failed validation | 
caused by RoundRobinLbConfigValidationError.SlowStartConfig: embedded message failed validation |
caused by SlowStartConfigValidationError.Aggression: embedded message failed validation |
caused by RuntimeDoubleValidationError.RuntimeKey: value length must be at least 1 characters)

So I have assumed these are mandatory, even if we would never use them.

Curiously, there are at least some cases where it is not required, like when we configure ratelimit here. In this case the message definition is missing the validation rule. Maybe that is just by mistake, since that is the only runtime_key that is missing validation in base.proto.

[sunjayBhatia]

LGTM, just made a little comment on doc wording 👍🏽

[tsaarni]

LGTM, just made a little comment on doc wording 👍🏽

Very good clarification, thank you for the review!