Orchestrator router

[yondonfu]

What does this pull request do? Explain your changes. (required)

This PR adds a new buildable binary livepeer_router. The binary runs a gRPC server that accepts GetOrchestrator requests from Bs and forwards OrchestratorInfo responses from Os. A router is configured with a set of Os and is responsible for routing a requesting B to one of the Os by passing the OrchestratorInfo response from that O to the B. The B will then submit segments to the URI indicated in the Transcoder field of the OrchestratorInfo response. If any Os go offline, are at capacity or are have network transit issues, the router can route a different O to the B. This router should also help mainnet node operators start running multiple Os with a single on-chain registered ETH address instead of running each O with a different on-chain registered ETH address.

At the moment, the router sends a received GetOrchestrator request to all of its Os and forwards the first OrchestratorInfo response back to the B. As a result, a requesting B will be routed to the O that is closest to the router. A more sophisticated routing algorithm can be introduced separately. For example, an improved routing algorithm might involve doing a geoip lookup on the Os and the requesting B and routing the B to the O with the smallest distance based on lat/lng coordinates.

I chose to include this router functionality in a separate binary because the livepeer binary already has a lot of configuration options and I was hesitant to introduce another "node type" there especially since this is a relatively early concept. Plus this allows for some experimentation with this functionality without having to incorporate it into the code for the livepeer binary just quite yet. If there is a desire to move this functionality into the livepeer binary I think we can consider doing so later.

Specific updates (required)

See commit history.

How did you test each of these updates (required)

Tested manually.

Checklist:

• README and other documentation updated
• Node runs in OSX and devenv
• All tests in ./test.sh pass

[yondonfu]

This CLI tool can be used in k8s environments for liveness and startup probes instead of vanilla HTTP probes which won't work right out of the box for a gRPC server. See livepeer/livepeer-infra#416.

[yondonfu]

This is required for the gRPC server to implement the gRPC health checking protocolhttps://github.com/grpc/grpc/blob/master/doc/health-checking.md) which allows us to use tools like grpc_health_probe to check the health of the server.

[yondonfu]

This function is pretty similar to some of the code in the discovery package so an initial thought was to use the code there that implements the common.OrchestratorPool interface. But, there are some additional dependencies with that approach such as a broadcaster interface, capabilities interface and suspender interface which are not requirements here (yet). So, for now, I chose to just add this relatively small piece of code for now.

[kyriediculous]

LGTM after squashing the fixup commits 👍

[darkdarkdragon]

One small comment, otherwise looks good.

[darkdarkdragon]

LGTM

[darkdarkdragon]

though CircleCI is failing with cp: cannot stat './livepeer_router': No such file or directory

[yondonfu]

though CircleCI is failing with cp: cannot stat './livepeer_router': No such file or directory

Ah looks like the intermediate Docker image was overwritten by another PR. Fixed by re-building to make sure the intermediate Docker image for CI contains the livepeer_router binary.