cmd: Refactor main() through helpers.

[kyriediculous]

Is your feature request related to a problem? Please describe.

• We should prevent the main function from terminating before deferred functions can be executed. Currently we are mainly using glog.Fatal which is equivalent to calling os.Exit(1), this exits the program not allowing any deferred functions to run.

• In some statements panic is used, which executes deferred functions after throwing but it is not idiomatic to panic in such contexts.

  panic: The amount of pixels per unit must be greater than 0, provided 0 instead
  goroutine 1 [running]:
  main.main()
  /Users/nico/go/src/github.com/livepeer/go-livepeer/cmd/livepeer/livepeer.go:327 +0x56f6

• Idiomatically the main function should be kept short and purely used for startup of the program. Startup of services should be initiated through helpers functions.

• We could refactor the cluttered main to simply glog.Error + return instead of glog.Fatal / os.Exit(1) but this makes testing the returned error codes in test_args.sh impossible

Additionally refactoring main makes it easier to maintain the main function and add functionality for the startup of services as needed.

Describe the solution you'd like Extract the startup of various services into helpers. Such as parseFlags startDBservice startEthClient setPriceInfo startMediaServer initS3storage startLivepeerNode

All error statements should be refactored to match the following pattern

glog.Errorf("...")
return

If the helper returns it's deferred functions will execute. We can then safely os.Exit() in the main after encountering an error

eg.

if err := parseFlags(); err != nil {
   os.Exit(1)
}

This is the same pattern used by Cobra: eg.

// Execute adds all child commands to the root command and sets flags appropriately.
// This is called by main.main(). It only needs to happen once to the rootCmd.
func Execute() {
    if err := rootCmd.Execute(); err != nil {
        fmt.Println(err)
        os.Exit(1)
    }
}

Additional context See this discussion in #933

[j0sh]

Generally I agree that the livepeer.go could use some cleanup ; see https://github.com/livepeer/go-livepeer/issues/710

Idiomaticly the main function should be kept short.

I don't really agree with this reason in itself; while it certainly helps to break out related pieces of logic into their own functions (since that's what functions are for, after all), doing this for the sake of "keeping functions short" usually results in unnecessarily deep call stacks and/or more overly-long helper functions, or code that's so disjoint it impedes clarity.

Extract the startup of various services into helpers. Such as parseFlags

To achieve the "shape" you're talking about implies putting nearly all common state in a giant struct or making it all global, or having functions with lots of parameters and return values.

I'd be OK with putting the parsed flags themselves in a struct (since that's also a way for us to sneak in features such as configuring via file (https://github.com/livepeer/go-livepeer/issues/101), but I'm curious if there is additional common state that we'd end up passing everywhere?

See this discussion in #933

Just curious: for the the pricing flags from that discussion, is there anything prior to that check which really ought to be defer'd to completion? If so, could the pricing flags be checked a bit earlier, so we can still exit safely? In the short term, this would be good to keep in mind as a rule when reviewing changes to flags.

os.Exit(1)

It's customary for (Unix) programs to return error codes based on the type of failure; this gives external tools some more information about why a program may have failed. I've thought about using this to help make our CLI tests more specific, but that's a separate issue.

[kyriediculous]

doing this for the sake of "keeping functions short"

To clarify, I'm talking explicitly about the main function, not other functions. In my opinion, the main function should be used to fire up the program and nothing else.

Perhaps I should remove the numbering in the original post, as I did not want to imply that this is the most important reason. The most important reason should be maintainability and being able to do proper error handling (as was the conclusion after the discussion I had with Yondon), the fact that it becomes short is just a positive side effect I guess.

In my opinion Cobra generated CLI programs are a great example of this.

To achieve the "shape" you're talking about implies putting nearly all common state in a giant struct or making it all global, or having functions with lots of parameters and return values.

I was personally thinking of making those global rather than a wrapping struct. Using following syntax:

var (
    network  *string
    rtmpAddr *string
    cliAddr  *string

)

Altough these don't need to be pointers if we dereference when calling flag.Type() in a possible parseFlags helper.

Just curious: for the the pricing flags from that discussion

Yeah on second consideration when skimming through the main function these could be better placed but would need some duplication of e.g. offchain/onchain checks and what type of node is being ran. The thing is that the checks imposed on the pricing flags actually use panic which implies safe exit but the returned errors aren't as nice.

In other places in the main we are still using fatal for e.g. serviceURI checks, etc. where exit isn't happening safely.

It's customary for (Unix) programs to return error codes based on the type of failure; this gives external tools some more information about why a program may have failed.

Oh yes, that would be nice. I didn't think of that. We could scope that in, considering both issues aren't priorities but just code optimisations down the line.

[cyberj0g]

+1 for refactoring the main function by adding proper data structures and moving away inline functions. It's 959 lines currently, not easy to work on. There are no good reasons to have any function this long.