Provide a fake grpclb balancer server that can be used for testing #2338
Conversation
apolcyn
force-pushed
the
grpclb_interop_client
branch
from
77c3120
to
4b32a76
Compare
interop/fake_grpclb/fake_grpclb.go
Outdated
| grpclog.Info("Expected first request to be an InitialRequest. Got: %v", lbReq) | ||
| return status.Error(codes.Code(2), "First request not an InitialRequest") | ||
| } | ||
| // gRPC clients targetting foo.bar.com:443 can sometimes include the ":443" suffix in |
There was a problem hiding this comment.
I remember this was discussed, and is expected.
If we want to remove the port, how about https://golang.org/pkg/net/#SplitHostPort?
Or we require the port number in serviceName.
There was a problem hiding this comment.
let's chat offline. For now, I'm using SplitHostPort
| if *shortStream { | ||
| return nil | ||
| } | ||
| for { |
There was a problem hiding this comment.
The test suite that's using it needs it. But I could also leave it out now and then add it in a follow up.
It's a useful test scenario for exercising some timing edge cases, and seems to reveal some unexpected behaviors so far.
interop/fake_grpclb/fake_grpclb.go
Outdated
| var ( | ||
| port = flag.Int("port", 10000, "Port to listen on.") | ||
| backendIps = flag.String("backend_addrs", "", "Comma separated list of backend IP/port addresses.") | ||
| useALTS = flag.Bool("use_alts", false, "Listen on ALTS credentials.") |
There was a problem hiding this comment.
Merge this and TLS flag into one that takes a string?
Otherwise we will need to deal with *alts==true && *tls==true, and it's hard to add another one (insecure maybe), just like in interop client.
There was a problem hiding this comment.
hmm, I agree this is a little weird, but the current flags do allow specifying insecure (by keeping both of these flags off), and it's convenient to be able to pass the same flags to this server as for the interop server
What do you think?
apolcyn
force-pushed
the
grpclb_interop_client
branch
3 times, most recently
from
fb44f50
to
ce849f9
Compare
apolcyn
force-pushed
the
grpclb_interop_client
branch
from
ce849f9
to
c9839c0
Compare
|
a couple of minor post-review changes made per the "go vet" checker |
context: I'm working on a grpclb-in-DNS interop test suite that uses fake a local DNS servers, backends, and balancers, in grpc/grpc#16727, in order to test grpclb edge cases with respect to DNS, various credentials types, and different client languages. For this, I've found it convenient to write the fake balancer server in go.
This PR is mostly aimed providing something that can be used by grpc/grpc#16727, but I'm thinking it might be useful to have around for manual testing anyways.