Don't solve the solved problems(blog.collections.me)
blog.collections.me
Don't solve the solved problems
http://blog.collections.me/post/46373756896/dont-solve-the-solved-problems
37 comments
I read this post as less "don't reinvent the wheel", more "if you're building a car, don't build your own wheel when you could buy one from a wheel manufacturer" - maybe if you did it yourself you could make improvements, but it's far more likely that a company where the problem you're trying to solve is their bread and butter will provide a better solution.
I think the core problem is that the question "build or buy?" is formulated in a way such that it rests along the incorrect axis. It's like giving people advice on whether they should go outside or not, whether they should fall in love or not, whether they should start a company or not, etc.
The proper answer is: "it depends." But that's unsatisfying because it's too complex. But in reality you can't eliminate that complexity, it's an irreducible complexity. Just as there isn't one simple, objective system which you can consult to reliably get an answer to the question "should you go outside or stay indoors?" there isn't one for whether you should use something off the shelf or build something yourself. Look at, for example, SpaceX, who built the first generation of engines for their Falcon 9 rockets using turbopumps (one of the most complex components of a high performance rocket) provided by 3rd party companies. However, in the subsequent generation of engines (the Merlin-1D) SpaceX has designed and built the turbopumps themselves, and this is responsible for significant improvements in the engine.
Ultimately I think it's more important to take a step back and make a conscious effort to avoid making big decisions based on instinct or inertia and begin to cultivate a set of processes, elements of corporate culture, and habits which encourage thoughtful analysis and introspection about these choices. People should be more thorough about evaluating the advantages and disadvantages of doing things different ways and be mindful of when it doesn't seem to make much difference. And also mindful of the opportunity to change your mind in the future. Sometimes it can be worthwhile to rebuild a tool that already exists in the industry just for the purposes of acquiring the experience and knowledge of having done so. And sometimes even when taking something in house might easily save millions of dollars a year it still might not be worth it due to other opportunity costs.
It's a complicated question, and it's always going to be a complicated question. Instead of pretending it's an easy question people should be working on acquiring the skills to make sure they can answer the question well with some degree of reliability.
The proper answer is: "it depends." But that's unsatisfying because it's too complex. But in reality you can't eliminate that complexity, it's an irreducible complexity. Just as there isn't one simple, objective system which you can consult to reliably get an answer to the question "should you go outside or stay indoors?" there isn't one for whether you should use something off the shelf or build something yourself. Look at, for example, SpaceX, who built the first generation of engines for their Falcon 9 rockets using turbopumps (one of the most complex components of a high performance rocket) provided by 3rd party companies. However, in the subsequent generation of engines (the Merlin-1D) SpaceX has designed and built the turbopumps themselves, and this is responsible for significant improvements in the engine.
Ultimately I think it's more important to take a step back and make a conscious effort to avoid making big decisions based on instinct or inertia and begin to cultivate a set of processes, elements of corporate culture, and habits which encourage thoughtful analysis and introspection about these choices. People should be more thorough about evaluating the advantages and disadvantages of doing things different ways and be mindful of when it doesn't seem to make much difference. And also mindful of the opportunity to change your mind in the future. Sometimes it can be worthwhile to rebuild a tool that already exists in the industry just for the purposes of acquiring the experience and knowledge of having done so. And sometimes even when taking something in house might easily save millions of dollars a year it still might not be worth it due to other opportunity costs.
It's a complicated question, and it's always going to be a complicated question. Instead of pretending it's an easy question people should be working on acquiring the skills to make sure they can answer the question well with some degree of reliability.
I think your example also speaks to the argument I put forward to some extent - SpaceX initially chose to use a 3rd party turbopump as building an efficient turbopump isn't their objective. However, at a later stage when the 3rd party turbopumps became an issue (whether that be because they were faulty, not efficient enough or any number of reasons) SpaceX decided to build their own.
This seems like the correct way round to me for the majority of cases where subcomponents of your product could be externalised - a 3rd party should be the primary go-to choice unless a company cannot be found that fits the requirements. The issue I see in development (and have been guilty of in the past) is simply building everything when libraries could easily cover the use case - "not invented here" syndrome. That's the real problem.
[edit] Also I think it's clear that there'll always be cases where the default doesn't apply, for example when working with a new field/concept or where the 3rd party options are too far from the mark, but for the majority of development (especially web development) there's likely a library/service to match many of your requirements.
This seems like the correct way round to me for the majority of cases where subcomponents of your product could be externalised - a 3rd party should be the primary go-to choice unless a company cannot be found that fits the requirements. The issue I see in development (and have been guilty of in the past) is simply building everything when libraries could easily cover the use case - "not invented here" syndrome. That's the real problem.
[edit] Also I think it's clear that there'll always be cases where the default doesn't apply, for example when working with a new field/concept or where the 3rd party options are too far from the mark, but for the majority of development (especially web development) there's likely a library/service to match many of your requirements.
What you are missing is that every library has its own dependencies and set of libraries. Gitlab uses nginx, unicorn, mysql, and a few other technologies. Ganglia uses apache, python, perl, and a few others. If you continue to add components you end up with seven different web servers, application servers, and little spare system resources and we havent even started talking about how to get these components to play nicely together! In conclusion, yes, it is nice if you can simply outsource to a library, but many times you have to bend over backwards to do so and it simply isnt worth it.
It depends a lot on scale. If you're just throwing together a web app that isn't going to have crazy amounts of traffic or business then maybe a big messy dependency chain is ok. But every one of those dependencies is a risk and a liability, and the more valuable and significant your project becomes the more important it is to lock down risk and reduce liabilities. Which means tracking down and eliminating dependencies.
One of the things at play here is the Dunning-Kruger effect. It's pretty common for there to be lots of different libraries and little tools and whatnot which duplicate existing tools but do so poorly. And part of the reason why is that it's often easy for people to look at those other tools, see the faults in them and think to themselves that they could do so much better with very little effort. But often this is wrong due to an inflated sense of ones own capabilities.
That's a hell lot of things to change, break, go down, lose your data, start charging ridiculous prices, be discontinued, etc... I'm all for SaaS and not re-inventing wheels, but I'd be nervous if I had to rely on so many external dependencies.
I think the vendor lock-in (cough heroku) is the bigger risk rather than reliability.
I've heard this a lot lately. Heroku uses super standard stuff (nginx,varnish,postgres,etc). I could move my stuff off Heroku in a matter of days. I used to do a lot of sysadmin work. I go with them because I can't devote time to sysadmin work anymore. Same reason I don't run my own mail servers anymore.
I'd argue mail servers (as SMTP endpoints) are more interchangeable than Heroku. They have a completely proprietary configuration and deployment process.
I can understand that competent sysadmins can recreate heroku's underlying hosting, but there's still a decent switching cost (especially if your whole team is accustomed to the heroku way). I've consulted with several companies that have 5-figure a month hosting on heroku that they can't migrate after all their investment.
If SendGrid changes my plan (or somehow degrades), it's a simple config change to send my email through another provider. A more apt comparison would be if your mail provider used a custom ReST endpoint instead of SMTP (say, as MailChimp does).
I can understand that competent sysadmins can recreate heroku's underlying hosting, but there's still a decent switching cost (especially if your whole team is accustomed to the heroku way). I've consulted with several companies that have 5-figure a month hosting on heroku that they can't migrate after all their investment.
If SendGrid changes my plan (or somehow degrades), it's a simple config change to send my email through another provider. A more apt comparison would be if your mail provider used a custom ReST endpoint instead of SMTP (say, as MailChimp does).
I agree. If you don't have any good sysadmin in your team, you are likely to be stuck at heroku. But then again, if you didn't have a sysadmin, you wouldn't have built it yourself. Starting at heroku or similar clone, still makes sense if your new thing may not make it past the first year. I guess it also depends what the business model is.
Presumably when any of those things happen, that's when you bring it in-house.
You should bring it in-house as soon as it becomes critical for your business. Only startups that aren't profitable yet should bet their skin on so many hosted services as long as they allow them to grow faster (that's not a given, sometimes SaaS/PaaS are complicated to use and unstable).
When you are raking in profits, you'd better use some of them to secure your business and make it independent of such risks. Which is the point the blog post seems to be trying to make.
When you are raking in profits, you'd better use some of them to secure your business and make it independent of such risks. Which is the point the blog post seems to be trying to make.
I don't think that's a viable solution. Some of these are quite complicated and replicating them in-house is not something you just up and do in a short time. That tactic might expose you to significant downtime.
I agree they take a long time to replicate. Which is why it doesn't make any sense to me to suggest that you should build them in house before you get your product out to the world.
It depends, sometimes you really do need the big complicated thing and you don't have time/money to develop it in house.
Sometimes though you can make do with something significantly simpler if you restate the problem slightly and move the pieces around.
You might be able to do the core of the job with a cron job and some command line tools rather than the fancy SaaS platform with graphs and everything.
Building on top of the fancy SaaS APIs and then trying to scale down when you want to move away is a lot more difficult than starting with the simpler approach and then scaling up when you really need that extra functionality.
Believe me, I've done it in both directions.
Sometimes though you can make do with something significantly simpler if you restate the problem slightly and move the pieces around.
You might be able to do the core of the job with a cron job and some command line tools rather than the fancy SaaS platform with graphs and everything.
Building on top of the fancy SaaS APIs and then trying to scale down when you want to move away is a lot more difficult than starting with the simpler approach and then scaling up when you really need that extra functionality.
Believe me, I've done it in both directions.
I was hoping for a slightly higher-level look at what problems not to solve. It seems that Collections.me solves a solved problem - convenient online storage. It iterates a bit on it, but you seem to be building something that is largely a pile of already-solved problems, both in its implementation and use cases. Just my opinion, of course, and that is how many services advance, but I thought I'd voice it in light of the purpose of your post.
Try http://www.codinghorror.com/blog/2008/10/programming-is-hard...
The OP doesn't really talk about "solved problems", it talks about the companies core competencies.
The OP doesn't really talk about "solved problems", it talks about the companies core competencies.
most of these services wouldn't be around if they had taken that advice ;)
I just so happen to be working on a product that significantly overlaps with Collections.
I've spent almost a year on it and don't think I'm going to take their advice if it's all the same to them.
I've spent almost a year on it and don't think I'm going to take their advice if it's all the same to them.
OP: thanks for mentioning us (https://circleci.com)!
That's a lot more 3rd party services than most use - do you find major differences in reliability?
That's a lot more 3rd party services than most use - do you find major differences in reliability?
I second most of the listed services. Just because you can code or host your own systems, it doesn't mean you should. Any time spent on extraneous internal devices is a disservice to your core and customers.
Lastly, shameless plug but I think we have a better accounting solution in http://cheqbook.com
"MongoDB is how we store things"
Is mongodb really a good solution for storing things? IIRC one concern about coinbase (YC) is that they used mongodb for storing financial information.
Is mongodb really a good solution for storing things? IIRC one concern about coinbase (YC) is that they used mongodb for storing financial information.
We like to embrace different technologies and use each for its strength. MongoDB is excellent for a whole class of use cases and we like it a lot, but we also use other persistence solutions as well. Check out Polyglot persistence post by Martin Fowler, it's an awesome read. http://martinfowler.com/bliki/PolyglotPersistence.html
MongoDB is web-scale
Edit: http://www.youtube.com/watch?v=b2F-DItXtZs
Edit: http://www.youtube.com/watch?v=b2F-DItXtZs
Stripe uses mongodb, if its good enough for them, its probably fine for financial info. As I recall, there is some aspect of mongodb that lets it scale up without maintainance, making it good for a service that just can't go down.
How do you get any of the benefits of redis over a Wan link? So confused by a hosted redis solution...
I have the same question...
Latency to another server in the same AWS availability zone is 1ms.
So it's really just a deployment solution for EC2?
I think other providers have similar speeds, and that the redis and mongo providers provision on lots of provider's machines. But I'm not sure how far that goes.
The Bolt-On Engineering for Dummies.
If I can ask, what's the general monthly run rate of those services you use combined?
How about the unimpleted problems ? Or does that fall under unsolved ?
This speaks better to how well heroku is able to sell add-ons.
"Don't reinvent the wheel." Classic advice. And it has some merit. But it's a broken idea.
Google? Not the first search company. The iPod? Not the first mp3 player. The iPhone? Not the first smartphone. Hacker news? Not the first tech oriented community site.
"Don't reinvent the wheel...
unless you think you can do much better."
Also, if you're merely iteratively improving the wheel, do so whenever you like. We don't drive around on hand-carved roughly circular hunks of wood any longer, and that's good.