[ANN] A fault-tolerant network service for meaningful GUID generation (in Go)

[bmizerany]

It's amazing how quickly Go allows me to produce so many useful, yet efficient things, so quickly, with little to no dependencies, and even less code.

http://news.ycombinator.com/item?id=3291802

[Sanjay Menakuru]

nice! pretty cool. 

one nit...

i think line 27 of https://github.com/bmizerany/noeq.go/blob/master/client.go should read n := rand.Intn(len(c.addrs))

according to documentation of rand.Intn (http://weekly.golang.org/pkg/math/rand/#Intn) its inclusive 0, exclusive of the number provided, so with the code you have now, it would never pick the last server in the list.

i could be completely mistaken however.

either way, its a testament to go's succinctness how 'welcoming' the source of this program is, considering the relatively difficult problem its solving.

Sanjay

[bmizerany]

Great catch. Thank you. Fixed.

[Sanjay Menakuru]

Also, one other thing.

You should call rand.Seed(time.Nanoseconds()) before you call the random function at all. if you look at http://weekly.golang.org/src/pkg/math/rand/rand.go?s=2823:2879#L108, they initialize the global seed to a constant of 1. so while a number gotten from rand will be pseudorandom compared to the next number from rand, the sequence of numbers from rand will be consistent for each run of the program. minimal program with which to show this can be found here: http://play.golang.org/p/q9usKyf67a

You have to run this on your machine as time.Nanoseconds() always return the same value on Google servers. Try it commenting out the seed line and run it a few times.

Sanjay

[bmizerany]

Right. I'm not sure how I feel about a library setting the global seed. I think this should be a documentation issue then, or I should use crypto/rand.

[Evan Shaw]

You could just not use the global rand.Rand object. Create your own
with rand.New and do what you want with it.

- Evan

[Sanjay Menakuru]

Yea, I was just trying to give a minimalistic solution. I normally use "rg := rand.New(rand.NewSource(*seed))", where seed is an int64 commandline flag that defaults to time.Nanoseconds(). This lets me get deterministic behaviour if I want, or by default true (pseudo) randomness.

Although I do agree that its weird that the global seed is a constant.

Sanjay

[Maxim Pimenov]

$ goinstall github.com/bmizerany/noeq

$ cd $GOROOT/src/pkg/github.com/bmizerany/noeq.go

$ gotest

rm -f _test/github.com/bmizerany/noeq.go.a

6g   -p github.com/bmizerany/noeq.go -o _gotest_.6 client.go  bench_test.go client_test.go example_test.go

client_test.go:4: can't find import: "github.com/bmizerany/assert"

make: *** [_gotest_.6] Error 1

Also, after "goinstall github.com/bmizerany/assert" is still fails:

--- FAIL: noeq.TestClientGen (0.00 seconds)

client_test.go:14: dial tcp 127.0.0.1:4444: connection refused

2011/11/30 12:47:17 ERROR: dial tcp 127.0.0.1:4444: connection refused

2011/11/30 12:47:17 ERROR: dial tcp 127.0.0.1:4444: connection refused

2011/11/30 12:47:17 ERROR: dial tcp 127.0.0.1:4444: connection refused

2011/11/30 12:47:17 FATAL: unable to connect to weiqid

gotest: "./6.out" failed: exit status 1

And the idea to keep *.6 files and binaries in your goinstallable repo seems bad to me.

[wkharold]

noeq is just a client. For it to work you need a server. That would be
noeqd - https://github.com/bmizerany/noeqd.git

... WkH

[bmizerany]

I'm sorry about that. I've updated the README. It should have been linking/referencing noeqd, not noeq.