4 ms·
There seem to be lots of people here that are offended by the "Node.js cancer" line. I am not defending his choice of words, but in TFA he wrote "Unix can do th
by IgorPartola 15y ago
There seem to be lots of people here that are offended by the "Node.js cancer" line. I am not defending his choice of words, but in TFA he wrote "Unix can do that all by itself with out all the stupid Node.js cancer.". I take it as, "there is no reason to have node.js involved since simpler tools can accomplish the same task". This says nothing about Node.js itself. In fact, try re-reading that same sentence but substituting a different technology from the following list: [MySQL, MS SQL, PHP, Perl, Visual Basic]. Does it sound slightly more reasonable now?
From what I gather his gripe was with the fact that the original Hubot was over-engineered. I agree that it was. Not only that, but the original Hubot came with almost no useful documentation, and a pretty horrible bootstraping mechanism (scripts that download other scripts onto a server over HTTP should never be run in prod environments). Getting it to run one step away from how it was intended (specifically on IRC) required digging deep into the code since the command line options/environment variables that were supposed to be passed in to make it tick were not exposed in any docs. This daemon process did not background itself properly, writing a PID file, etc. The whole thing was a cool proof of concept, but running it in production would make me very nervous. On top of that, it did require Redis, for some ungodly reason. SQLite would have been a much nicer choice since it's much more easily obtained (at least on CentOS) and does not require a separate process or having to configure. (Offtopic: has anyone had good experiences with SQLite style no-sql data stores? Something like MongoDB, but storing stuff on disk and accessing it in-process?).
Don't get me wrong. Hubot is awesome. It's just that it's only awesome at GitHub, where it integrates into all sorts of ecosystem and the culture. I tried it for an afternoon, and by default it comes with no useful plugins (mustache generators notwithstanding) because the plugins are the trivial part when you have a great infrastructure already. Imagine if I handed you the source to bash, but you and everyone you know uses MS DOS. Yes, you can see that bash has potential to be powerful, but until you have utilities like find, ls, grep, awk, sed, ed, cat, and so on.
- semanticist 15y agoA simple file-based key-value store would be something like GDBM? It's been around since approximately forever and has been used in a ton of things. NoSQL doesn't have to mean new and trendy.
- IgorPartola 15y agoAgreed. I was thinking of an object store where you get to index properties of objects and perform arbitrary queries. New, old,etc is not a concern.
- tombell 15y agoNews just in: cool and fun side project doesn't do things the 'serious' way. Hubot doesn't "require" Redis, don't load the `redis-brain.coffee` script. Problem solved. Hubot is a framework for building your own useful things on top of it. Sure it doesn't provide a billion and one ways to do useful thing, you add those yourself. If it doesn't work for you, don't use it simple as. Given it is a fun side project a company can use, I don't think the concept of "production" or "development" comes into it. You just run it. Plenty of people have run it without issues, and anyone that have had issues has been helped to have them resolved. Feel free to send a pull request to improve documentation rather than just slate it :). Hubot is what YOU make it, not what you can just download and hope it brings useful things to your team.
- IgorPartola 15y agoI understand that Hubot is more framework than finished product. My point is that when I am evaluating whether to implement it fir the company I work for vs spending time and resources on some other side project, I am having a hard time making a case for this particular framework. As for production vs dev: Hubit is either your tool for deploying code or it is not. If there are many ways to deploy the same piece of code you are doing it wrong. Now consider being the Ops guy who has to deploy a critical security fix and Hubot took a cigarette break because it never started properly or some other bug in the code caused it to fail.
- tombell 15y agoUnderstandable. But surely you would have a fall back way to deploy given your main way doesn't work? I can deploy my apps via Hubot, but I can also deploy them without Hubot because I have an abstraction that Hubot uses to deploy my apps. And as far as I know GitHub have also something similar in place. A bug in any piece of software is going to cause issues, Hubot isn't an exception.
- IgorPartola 15y agoYes. And I agree that any piece of software in you stack can ahve bugs that make life difficult. My issue is that (1) Hubot is supposed to be central to your ecosystem and (2) when I looked at the code at first I saw a style of deployment and docs that made me think that running it would be more trouble than it's worth.
- cldwalker 15y ago* I've been running this on heroku for free for weeks. Why would anyone be nervous about that? * As tombell pointed out, you don't need Redis * The docs are ok but compared to just ranting docs for this fork they're awesome. * The docs for IRC usage have moved - https://github.com/nandub/hubot-irc/ https://github.com/nandub/hubot-irc/ * Who cares if Hubot is over-engineered? It has tests. This ranting fork has none. Moving on ...
- IgorPartola 15y agoI would not be running it on Heroku Redis requirement seemed to be strickt based on the docs when Hubot first hit HN Agreed on ranting, but docs are not OK. Docs talk about how awesome it is, but not about how to get it running Those docs were not there AFAIK when it first hit HN. Just because something has unit tests does not mean its paradigm is sound or that it scales or is reliable. This "fork" is not great either by the criteria I use. On top of that both are simply for fun projects not billed as bullet proof battle hardened pieces of software. However, I can see the OP's point of view: Hubot is over-engineered, with a convoluted deploy process outside of Heroku and other design shortcomings. The OP attempted to improve on the idea with a different design, and arguably failed.
- tombell 15y agoIt's actually fairly simple to get running locally, I'll take a look at improving the documentation. 1. Download the latest version from the downloads section 2. Install dependencies with `npm install` 3. Export any environment variables for your adapters and/or scripts 4. `bin/hubot -a <adapter>` That's as simple as it is to run in a terminal. You can use many things to manage the actual process e.g. Foreman/Upstart/Forever.
- icebraining 15y agoFor a in-process key-value store, I'd use BerkeleyDB [1], I think it's pretty much the default solution and it's used by plenty of projects. The free license is copyleft, though, and the commercial license is pricey. [1]: https://en.wikipedia.org/wiki/Berkeley_DB https://en.wikipedia.org/wiki/Berkeley_DB