fix: ensure the xds grpc server is properly stopped

[shawnh2]

What type of PR is this?

fix: the two xds grpc server in envoy-gateway has not been properly stopped.

logs (before fix)

INFO	gateway-api	runner/runner.go:146	shutting down	{"runner": "gateway-api"}
INFO	xds-translator	runner/runner.go:102	subscriber shutting down	{"runner": "xds-translator"}
INFO	xds-server	runner/runner.go:157	subscriber shutting down	{"runner": "xds-server"}
INFO	infrastructure	runner/runner.go:75	infra subscriber shutting down	{"runner": "infrastructure"}
INFO	provider	kubernetes/controller.go:1037	grpcRoute status subscriber shutting down	{"runner": "provider"}
INFO	provider	kubernetes/controller.go:1009	httpRoute status subscriber shutting down	{"runner": "provider"}
INFO	cmd/server.go:220	shutting down
INFO	provider	kubernetes/controller.go:1093	tcpRoute status subscriber shutting down	{"runner": "provider"}
INFO	provider	kubernetes/controller.go:1121	udpRoute status subscriber shutting down	{"runner": "provider"}
INFO	provider	kubernetes/controller.go:1065	tlsRoute status subscriber shutting down	{"runner": "provider"}

# there is no log indicates the grpc server has been stopped.

logs (after fix)

INFO	gateway-api	runner/runner.go:146	shutting down	{"runner": "gateway-api"}
INFO	xds-translator	runner/runner.go:102	subscriber shutting down	{"runner": "xds-translator"}
INFO	xds-server	runner/runner.go:160	subscriber shutting down	{"runner": "xds-server"}
INFO	infrastructure	runner/runner.go:75	infra subscriber shutting down	{"runner": "infrastructure"}
INFO	cmd/server.go:220	shutting down
INFO	provider	kubernetes/controller.go:1093	tcpRoute status subscriber shutting down	{"runner": "provider"}
INFO	provider	kubernetes/controller.go:1121	udpRoute status subscriber shutting down	{"runner": "provider"}
INFO	provider	kubernetes/controller.go:1065	tlsRoute status subscriber shutting down	{"runner": "provider"}

# the grpc server stopped
INFO	xds-server	runner/runner.go:108	grpc server shutting down	{"runner": "xds-server"}  
# the ratelimit grpc server stopped if ratelimit has been enabled
INFO	global-ratelimit	runner/runner.go:103	grpc server shutting down	{"runner": "global-ratelimit"}

What this PR does / why we need it:

the grpc.Serve() method will block <-ctx.Done(), so the <-ctx.Done() will never be executed even if the ctx is canceled, causing the grpc server never be stopped.

gateway/internal/xds/server/runner/runner.go

Lines 105 to 110 in 3e0ba18

	err = r.grpc.Serve(l)
	if err != nil {
	r.Logger.Error(err, "failed to start grpc based xds server")
	}
	<-ctx.Done()

so the fix here is to let grpc.Serve() method DO NOT block <-ctx.Done(), let grpc server properly stopped.

Which issue(s) this PR fixes:

None

[codecov]

Codecov Report

Merging #1860 (7649614) into main (1fab508) will decrease coverage by 0.06%.
The diff coverage is 0.00%.

@@            Coverage Diff             @@
##             main    #1860      +/-   ##
==========================================
- Coverage   65.35%   65.29%   -0.06%     
==========================================
  Files          86       86              
  Lines       12526    12527       +1     
==========================================
- Hits         8186     8180       -6     
- Misses       3823     3830       +7     
  Partials      517      517              

Files Changed	Coverage Δ
internal/xds/server/runner/runner.go	28.70% <0.00%> (-0.27%)	⬇️

... and 4 files with indirect coverage changes

[arkodg]

LGTM thanks !