Refactor the API to use go-kit

[krasi-georgiev]

Started looking at the api code and I think it can be improved by using the go-kit package.
https://github.com/go-kit/kit
https://youtu.be/JXEjAwNWays?t=1451
https://www.youtube.com/watch?v=NX0sHF8ZZgw

[krasi-georgiev]

@tomwilkie what do you have against go-kit?
I haven't used it, but it looks like a good fit.

[gauravsitlani]

@krasi-georgiev Can I take the initiative to help out .
This is the code right ?
https://github.com/prometheus/prometheus/blob/master/web/api/v2/api.go

[krasi-georgiev]

that would be great!
I started a topic in the def list, but hasn't been discusses yet so need a bit more input from the other maintainers.
https://groups.google.com/forum/#!topic/prometheus-developers/IebVSsE2_Gk

I am leaning towards gokit + swagger spec generation via annotations.
https://goswagger.io/generate/spec.html

[gauravsitlani]

@krasi-georgiev I'll go through the api code and check out go-kit + swagger and follow the thread.

[krasi-georgiev]

Thanks.
post your findings or questions in the Google groups thread.

[Bo0km4n]

Hi I’m Katsuya,
I’m thinking to participate GSoC’18, and I’d like to work on Prometheus.
I found this issue Prometheus’s ideas page,
It’s sounds very interesting.
Could you tell me more detail of this issue?
If there is anything I can do, I want to help.

[krasi-georgiev]

I did a quick research and started a dicsussion in the Prometheus mailing list.
https://groups.google.com/forum/#!topic/prometheus-developers/IebVSsE2_Gk

also started a doc to take some notes
https://docs.google.com/spreadsheets/d/1juaYH-unAYIp1-rlmYRNojxxwRswK4w7lDPHtUQj398/edit#gid=0

overall I am leaning towards go-kit with some code annotations to auto generate the code based on the code annotations/comments next to each endpoint.

if you have an opinion can post it in the mailing list to get some more comment and can also ping me on irc #prometheus-dev if you have any questions.

[krasi-georgiev]

@Bo0km4n should hang around a bit longer in the IRC. Saw your question, but you were not in the irc anymore. if you put krasi in front of the question I will get a notification and better chance to get a quick reply.

[Bo0km4n]

Oh, I’m sorry,,, I’m the first time to use IRC. Even if you skipped a private chat with the “/privmsg command”, I got an error. If you do not bother you, I’d like to contact you by email or github issue.

…

2018/03/15 4:34、Krasi Georgiev ***@***.***>のメール: @Bo0km4n <https://github.com/bo0km4n> should hang around a bit longer in the IRC. Saw your question, but you were not in the irc anymore. if you put @krasi <https://github.com/krasi> in front of the question I will get a notification and better chance to get a quick reply. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#3416 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AOYwmz5QdQz1cJHR1Il2S8T3jLebCezlks5teXCvgaJpZM4QTJWL>.

[krasi-georgiev]

it might be a bit difficult of you have more than one question , but lets try , fire away.

[Bo0km4n]

Fmm... Ok, I think I will try a conversation on irc. Could you tell me about private chat command? I tried command “/privmsg krasi hello, nice to meet you” But actually, I couldn’t send message to you. What command when you send private message to other user? Sorry to hear persistently.

[krasi-georgiev]

send me a message in the main room

[Bo0km4n]

Sorry I don't know how to enter main room. Do not you just enter the #prometheus-dev channel? 2018-03-15 5:02 GMT+09:00 Krasi Georgiev <notifications@github.com>:

…

send me a message in the main room — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#3416 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AOYwm5kdrxq11lt-b396zDDijFLN_edgks5teXdDgaJpZM4QTJWL> .

[brian-brazil]

What's the actual problem we're trying to solve here?

[krasi-georgiev]

AFAIR the code is not well organised , v1,v2, admin are all separate and difficult to follow.
versioning is difficult and non DRY , API documentation is all manual and easy to drift.

[brian-brazil]

I don't see how using a framework would resolve any of that, it'd just add an extra layer of abstraction and complicate the code. That versioning was never properly designed wouldn't be fixed either, and I personally think we should be considering deleting v2 rather than fixing it.

[krasi-georgiev]

can't be sure on a 100% , but I would like at least to try and refactor it.
If it ends up the same or more complicated than we don't have to merge it.

I am against using frameworks as well so will try to avoid it at all cost unless it brings some real time and maintaining savings.

[brian-brazil]

As is stands there's very little maintenance required on the APIs, and adding new endpoints isn't that hard.

[krasi-georgiev]

It was long ago so will check again and will close the proposal if that is the case.

[krasi-georgiev]

I had a look at the code again as part of troubleshooting a race condition and must say it is extremely unorganised , difficult to follow and difficult to troubleshoot.
It uses 2 different routers with some rules defined in the New function and others in the Run function.
Overall adding more things to the api . versioning etc. would make it really difficult to maintain in its current state.

Now I am even more convinced than before that this can get a lot better.

[krasi-georgiev]

the last dev summit it was decided to try open api for the alertmanager and it all good will consider it for Prometheus as well.