Add tracing(Opentracing + Appdash)

[NIC619]

How was it fixed?

Add some basic tracing setup(Opentracing + Appdash) to keep track of the RPC calls between different nodes in the network:

• There's one Appdash server which collects span data(i.e., logs) from different nodes and visualize them via a web UI server.
• There are relationships between spans, for example, an AddPeer span or SubscribeShard span would be the ChildOf an RPC Server span.
• Whenever a node is started, a new trace is started which includes an RPC Server span and multiple RPC call spans(e.g., AddPeer).

• Each trace and spans has it's own unique ID and one can attach extra information to the span by SetTag.

• [x] Add an RPC server Stop call to turn down the RPC server so the main span(span with name RPC Server) can terminate normally.

WIP: Look into libp2p codebase to see how they set up and utilize tracing WIP: survey for different/better visualization tools

Cute Animal Picture

[mhchia]

Reference to https://github.com/ethresearch/sharding-p2p-poc/issues/27

[mhchia]

TODOs

• [x] Avoid type conflict due to gx-ed repo between opentracing and gx/ipfs/.../opentracing
  • the temporary solution is, after doing gx-go rw, we "unwrite" the "gx/ipfs/.../opentracing" imports to "github.com/opentracing/opentracing"
• [x] fix "proto: invalid utf-8 string"
  • It occurs in addpeer and broadcastcollation, the solution is to ensure we don't store any non utf-8 string in the type string in protobuf(or, just change the type string to bytes in those use cases)
• [x] fix "duplicate proto type registered: basictracer_go.wire.TracerState"
  • it results from the fact that two protobuf messages in different repository has the same package name and the message name. In this case, the name of proto message will be registered twice, due to the call to init in *.pb.go
  • In our case, the message in ipfs/go-log conflicts with the one in basictracer_go
    • IMO ipfs should rename the proto, to avoid the conflicts when using it with other tracers importing basictracer-go
    • issue: https://github.com/ipfs/go-log/issues/40
    • [ ] Change the ipfs hash and version in package.json to the latest one, after it is published

[mhchia]

I merged this branch with master, and modified partial-gx-uw.py to print the unwritten imports. Let's see what errors occur in the CI now

[mhchia]

Ah...I just removed fstring. Hope we can get rid of the error. Thanks a lot @ChihChengLiang

[NIC619]

Seems like we forgot to fetch the dependency packages recursively while the command go test -v ./… will run the tests recursively(hence it would run into the cannot find package of XXX error.

As for the previous undefined: proto.GoGoProtoPackageIsVersion1 error, the reason seems to be that the wrong version of package gogo/protobuf/proto is fetched by gx and it's currently fixed by gx unwrite the specified packages.

[mhchia]

So we need to pass ./... to go get and go test, to build all packages under the directory, and pass . to go get in docker, since we only need the exact main package in the directory to make it build? Btw., CI passed!