Updated Trillian/etcd/grpc to latest version

[mhutchinson]

Most of this is pretty straightforwarded, just bending the code to the slightly different shape of the dependencies. While the etcd announcements look quite different in code, I've confirmed that the output of etcdctl get trillian-ctfe --prefix is the same before/after this change, e.g.

trillian-ctfe-http/localhost:6965
{"Op":0,"Addr":"localhost:6965","Metadata":null}
trillian-ctfe-metrics-http/localhost:6965
{"Op":0,"Addr":"localhost:6965","Metadata":null}

• I have updated the CHANGELOG.
  • Adjust the draft version number according to semantic versioning rules.
• I have updated documentation accordingly.

[codecov]

Codecov Report

Merging #774 (f8c02bd) into master (3d1e6e0) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #774   +/-   ##
=======================================
  Coverage   72.94%   72.94%           
=======================================
  Files          81       81           
  Lines        7842     7842           
=======================================
  Hits         5720     5720           
  Misses       1696     1696           
  Partials      426      426           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3d1e6e0...f8c02bd. Read the comment docs.

[mhutchinson]

Hi @pgporada do you mind taking a look at this PR and letting us know if & how it'll affect you? I've tried to keep the changes minimal, but I've been forced to make a minor functional change, and I don't have the context to know how big a deal this will be.

[pgporada]

Thanks for pinging me about this. Turns out that we're doing something goofy in our ctfe configuration. I wasn't aware that I could or even should be using etcd for the ctfe. From what I can tell our ctfe's are hitting the kubernetes services api

certificate-transparency-go/trillian/ctfe/ct_server/main.go

Lines 158 to 161 in 8e37269

	} else {
	glog.Infof("Using regular DNS resolver")
	dialOpts = append(dialOpts, grpc.WithBalancerName(roundrobin.Name))
	}

ctfe config

    spec:
      containers:
        - name: trillian-ctfe
          args: [
          "-http_endpoint=0.0.0.0:6962",
          "-metrics_endpoint=0.0.0.0:6963",
          "-log_rpc_server=dns:///log-server:8090",  <==== this thing
          "-log_config=/ct-server.cfg",
          "-logtostderr",
          "-rpc_deadline=30s",
          "-max_get_entries=256",
          "-by_range=true",
          "-align_getentries=true",
          "-mask_internal_errors=true"
          ]

Our trillian log_server and log_signer's contain the following config. Will a similar code change to what you've done here also need to be done for the log_server and log_signer when a similar PR lands in the trillian repo?

        "-etcd_servers=etcd-cluster-client:2379",
        "-etcd_http_service=logserver-http",

Assuming that I switch my ctfe's to use etcd, am I correctly understanding that the config would look like:

        "-etcd_http_service=logserver-http/",
        "-etcd_metrics_service=logserver-metrics/",

You're right that feels a bit dodgy. I am onboard with a single service with two endpoints.

[pav-kv]

This is backward-compatible, as far as I can tell. LGTM.

[pav-kv]

Update the PR desc?

[mhutchinson]

I've updated the PR description 👍

[mhutchinson]

@pgporada thanks for your comments here. I went back to look at Trillian and then found out my expectation of the previous behaviour didn't match the reality. See google/trillian#2381. It turns out that the endpoints were already announced under a foo/bar format, and I've changed this PR to update etcd APIs without any functional changes.