Balancing Performance and Convention

markmcb writes "My development team was recently brainstorming over finding a practical solution to the problem that's haunted anyone who's ever used a framework: convention vs. customization. We specifically use Rails, and like most frameworks, it's great for 95% of our situations, but it's creating big bottlenecks for the other 5%. Our biggest worry isn't necessarily that we don't know how to customize, but rather that we won't have the resources to maintain customized code going forward; it's quite simple to update Rails as it matures versus the alternative. What have your experiences been with this problem? Have you found any best practices to avoid digging custom holes you can't climb out of?"

Comments

Expand the toolkit

[thethibs]

Would it be out of the question to assess the needs of the troublesome 5% (and perhaps the other 10% that were shoehorned into Rails) and add a framework that's more in line with those needs?

Data exchange is sufficiently mature that interaction between applications in different environments is not an issue, so all you are left with is the added overhead of supporting two frameworks; not a bad thing if you consider the added flexibility of using the framework best suited to each application. Having options is usually a good thing.

Re:Expand the toolkit

[Bearhouse]

Mod up! If the only tool you have is a hammer, you end up treating everything like a nail.

As the OP pointed point, all development environments sooner or later hit this wall. I've never found one, (and am pretty sure that one will never exist), that ticks all the boxes for a typical mixed environment (the GUIs, OLTP, distributed database management, web access, making the coffee...).

In the 80s, I worked for a company that - as was fashionable at the time - developed its own '4GL' and decreed that everyone use it. In a satellite office, we tried, but found that in certain circumstances it just plain would not do what we needed. Guess what? We went back to 'hard coding' those key functions...

Don't worry about it, (after all, same bits of windows, to name just one OS, are written in assembler to speed performance).

So as the man says, expand your toolbox and your competencies - it'll make you a better project manager (to handle the interfaces) and programmer.

Re:Expand the toolkit

[truthsearch]

so all you are left with is the added overhead of supporting two frameworks

And that's often much better than supporting one framework plus your own customizations. Once you customize it's often extremely difficult to upgrade the framework and have your patches still work. So there can be a lot of extra development time involved. Plus just the learning curve to get intimately familiar with the internals of the framework when that's often not necessary.

Hard to say, but...

[SlashDotDotDot]

Making this sort of decision really requires more technical details than you provided in the summary or the linked article.

If you do choose to make custom modifications to Rails (or any other third party dependency in your system), make them very carefully. Keep notes about what you changed, and why. Comment the modified source code with a consistent, searchable tag, like

// MYCOMPANY Begin custom changes
// Optimized this block for queries where the
// search string is only one word long, and
// measured a 20% performance increase on the
// production server under moderate load.
// 3-Jan-2009
...
// MYCOMPANY End custom changes

Also, check the third party source into your own version control system so that you can track the changes explicitly.

Since you are talking about making changes for performance reasons, be sure to follow good optimization practices. Measure first, optimize only the true bottlenecks. Measure under realistic scenarios. If optimizations are even slightly confusing or subtle, comment them thoroughly. Keep the original, unoptimized code, maybe just in comments next to the new code, or maybe in a separate "reference implementation" function so that you can fall back to a known reliable version.

Re:Switch to django and python for starters

[latentevent]

Comments like this really don't address the core issues. Frameworks, and especially heavily-developed ones like Django/Rails, are rarely cause for insurmountable performance problems. Of course, caching and setting up distributed workload processing queues will be necessary in any framework, but that's up to the development team working on a specific app rather than the framework itself.

I've been working on a large Rails application for over a year, and we haven't found any scaling issues related to Rails. Instead, there are a number of things that we've done in our application in order to help it scale for our given workload.

That said, there are a number of things that Rails will do to help you scale your application. You may want to look at how Rails 3.0 (the next release of the Rails framework) is incorporating several ideas from Merb to make the framework lighter-weight and more extensible through Rack and the concept of Rails "metals." For offline processing in a Rails app, look at plugins like workling/starling and other message queues. I think you'll find that 1) Rails doesn't have scaling issues, programmers on Rails have several challenges that they have to face to make their applications scalable (just as in any other framework) and 2) As far as providing tools to facilitate developers in scaling their apps, Rails is quite a powerful framework and one that should be given a lot of consideration by development teams.

In conclusion, though, vague comments like "avoid Rails, use Django" are just unproductive spreading of FUD and don't really contribute to any significant discussion or understanding of how to get any of the viable web application development frameworks to scale in any specific case.

Re:Switch to django and python for starters

[Forge]

After reading the linked commentary, I can say there really is a better 4th option. It was even hinted at by the author.

Write your extension/hack as a modification of Rails and/or Ruby. Send them back to the maintainers of said products and see if you can get them inserted as a standard part.

If they do not exact a significant penalty elsewhere for the improved performance in your pretty mundane scenario. It is highly probeble, your changes will be accepted.

This isn't just true for Rails. It's true for any software (except stuff maintained by immature, anal retentive morons.

If it's L/GPL, BSD or other common free license, then this adoption of customer written code is a routine matter. If it's closed source and proprietary they may feel the need to get a lawyer involved and have you sign over all ownership rights to the code you wrote to improve their product (ask for money or stock when that happens).

Either way, in most scenarios good clean extensions tend to get adopted as part of the main program.

PS: Yes. Even "The Evil Empire" (Microsoft, Sun, Apple, IBM etc...) have been known to do this.

Build vs. Buy

[michaelmalak]

What you call "convention vs. customization" is really just a special case of "build vs. buy". You can research the standard factors that go into such a decision, but the most important one is whether the part you intend to build is both a) essential to the business and b) something that would give you a competitive advantage over buying.

I agree with the above poster.

[Jane Q. Public]

I work on a major website, all done in Ruby/Rails. We, in contrast, have found that the tremendous ability to customize the framework is liberating, rather than constrictive.

Without knowing much more detail about your problem, it is difficult to give advice. Having created large, complex websites, and maintained them for long periods of time, we have yet to run into intractible problems as a result of the language or framework themselves. The fact that there are many large, complex Ruby on Rails websites being created and maintained by Fortune 500 companies also contradicts the idea that Ruby and Rails have serious, show-stopping shortcomings.

I strongly suspect that part of the problem is that you do not know your platform (in particular, Ruby) as well as you perhaps could. Ruby is very extensible via Gems that are freely available or that you can write yourself, or even just code modules. Rails is similarly, wonderfully extensible via plugins. Plugins are also freely available for just about every situation you can imagine, or again you can write your own.

Ruby is one of the most flexible languages in existence. Rails is hardly perfect, but any shortcomings are easy to code around by adding your own extensions. But expecting any framework to be complete for your purposes, before you begin, is pretty much wishful thinking.

Once again, I do not know enough about your situation. But if you expect any framework to be "plug and play" for very complex tasks, like an erector set, you will be disappointed. There is no substitute for knowledge of your platform, and programming ability.

JRuby

[BrainInAJar]

Why not deploy your app on JRuby, and if there are still performance bottlenecks, write those parts in real Java ?

Re:Switch to django and python for starters

[arth1]

And when the guy knowing Django moves on, what's the chance of finding a replacements who knows it? How much will that reduce the number of eligible applicants for a position -- will you disqualify fifty really smart guys for five mediocre ones who happen to know Django?

In my experience, being a system administrator for over a generation, what is the most maintainable is to go for standards and well-defined APIs, and document it, document it, document it. Especially document what's obvious and "standard practice", because it may be "standard practice" to do a certain thing in a certain language or certain package, those who come around ten years down the road may not know that this was standard, or why. So document, not so much what you're doing as why. Any coder worth his pizza will figure out what but be clueless as to why.

I feel compelled to add

[Jane Q. Public]

that the other serious problems we have had have been programming bugs; but those bugs were due to our own coding or design lapses, not framework limitations.

Most people who speak of "limitations" to their framework are people who expect the framework to do all their work for them: this is simply unrealistic. If that were how it worked, it would not be called a "framework", but rather a "product".

Re:Switch to django and python for starters

[arth1]

Comments like this really don't address the core issues. Frameworks, and especially heavily-developed ones like Django/Rails, are rarely cause for insurmountable performance problems.

>
No problem is insurmountable. But they are commonly cause for surmountable performance problems. And surmounting costs time and money, while not impressing those who noticed the problems or pay for fixing them.

If in doubt, KISS. Adding extra abstraction layers is generally not simplification. It's like adding another layer of management in a company, in that it is sometimes necessary, but it will rarely increase productivity or responsiveness, at least not in the long term. Especially not when a middle level manager gets replaced with someone who doesn't do things the same way.

Re:Switch to django and python for starters

[Serious Callers Only]

I worked in a small shop that used rails, we found that rails is... constricting, just for the reasons you posted to slashdot for. We looked at our options and switched to django+python. Maintenance wasn't a problem after that. I'd suggest investigating a switch now while you have an opportunity.

Out of interest, what did you find constricting about Rails? Your comment here (and the article, such as it was) are not very illuminating on this score.

I've found it fairly easy to modify, very easy to add plug-ins to, and very easy to add custom SQL queries to if necessary for performance reasons (though most of the time that's not necessary). The only area I can see it being constricting would be when trying to interface with a legacy db structure which can't be changed, in which case Django would have similar problems.

Because the article is not specific enough on the problems they'd like to work around, it asks an unanswerable question - there is no magic formula for trading off convention and performance, no magic bullet; at a certain point you have to stop relying on a framework and add something yourself if it doesn't do what you want or perform to your satisfaction. The same holds true for any framework, and though the received wisdom on Slashdot is that Rails is difficult to work with, or doesn't scale, many companies don't find that to be the case (yellow pages, etc etc).

JEE tech is too complex and fragile

[presidenteloco]

We are well into a project that has JBoss SEAM as its basis, but required significant mods to give us multi-database capability and a FLEX front end. So instead of a community maintained opinionated meta-framework, we now have our own super complex, fragile framework cluster(****).

It is also effectively a dense, opaque yarnball. The stack traces of exceptions are so long, due to the interceptor architecture, that the Java stack trace displaying algorithm gives up and prints ...

So my advice would be, at the risk of trolling, but I intend it as a serious debate position, don't start with something as excessively complex as JEE technology.

Because JEE technology is barely manageable, maybe, if you stick EXACTLY to the most popular opinionated meta-framework for it, but if you deviate off the straight and narrow path, you are on your own in a dense jungle with a dull machete.

Ask Slashdot

[starrsoft]

FTFA:

So we have two basic options: pay for bigger hardware, or spend more time writing/maintaining custom code that allows us to live on our existing hardware.

Hardware is Cheap, Programmers are Expensive:

Given the rapid advance of Moore's Law, when does it make sense to throw hardware at a programming problem? As a general rule, I'd say almost always. Consider the average programmer salary here in the US. You probably have several of these programmer guys or gals on staff. I can't speak to how much your servers may cost, or how many of them you may need. Or, maybe you don't need any--perhaps all your code executes on your users' hardware, which is an entirely different scenario. Obviously, situations vary. But even the most rudimentary math will tell you that it'd take a massive hardware outlay to equal the yearly costs of even a modest five person programming team.

Re:Language is not the answer

[quanticle]

I still remain skeptical about the benefits of model-driven-development. I recall that, at the latter end of the 1980s, developers were promised much the same thing with the push toward CASE tools. Of course, developers and managers quickly learned that such tools were only of use in a certain limited number of scenarios, and that attempting to use those tools outside of the scenarios for which they had been designed quickly led to the same problems.

Frankly, I'm not sure that "software entropy" is solvable by any sort of tool or language. Rather, it is solved by training developers to write tests, and enforcing certain minimum style standards on the code, so that any developer can look at any portion of the codebase without feeling too unfamiliar with it.

Version Control

[omb]

I hate code full of company comment cruft.

Use a good VersionControl system, git, murcurial or Subversion if you must, __but__ keep the meta-data out of the code. If you use sensible tools you get all that, and its much easier to work with the upstream,

Re:That's rich!

[Per Bothner]

Java is quite fast, thank you.

Re:Switch to django and python for starters

[shirai]

Rails has a lot of problems mixed in with a lot of greatness. For a great many projects, its greatness overcomes its probems.

Its biggest problem, IMHO, is that it does not grow well with the needs of a developer over time. Some things:

* Doesn't scale well
* Not easy to do anything outside the Rails world

I had a conversation with Ezra back when he was starting Merb, and I remember the large resistance to the idea that there were problems with Rails at all. As Merb got released and then popular, it was clear that there are holes in Rails (yes I know they are merging).

At about the same time, I started my own framework (the Caffeine framework now the Go framework) which has not been made open yet (and is not complete yet), but there are several issues that it addresses which could help (a) improve Rails or (b) create other frameworks that can fill holes. I'll mention some of the ideas below.

Note that the original goal was to use Rails for our "very cool application" (tm). Unfortunately we have some pretty lofty goals and Rails turned out to be the wrong tool. I am the CEO of a successful Internet company CityMax (about 30 employees) so I do have experience with the problems that happen after launch and success. They are not the same problems as during the startup phase.

Here are some things that an ideal framework has that I believe are missing or lacking in Rails. Again, I do want to stress that Rails has greatness and that I really respect its creator. That said, I think frameworks can grow to another level.

# Built in scalability and clustering. Should require zero changes (except some configuration) to go from single database to a cluster of databases.
# Reusable code. If you build a feature (like a message board) for one application, that same code should be reusable in another app with zero code changes. It should also be reusable multiple times in the same application. For example, a blog app might use the message board in each blog but there might also be a support message board for the whole system.
# Built in scalable file system. The file should be storable in local, Amazon S3, MogileFS, network share or whatever. Directories can be mapped to any file store (e.g. temp directories might be local, permanent storage might be S3 and archives might be a network share). Should be easy to remap directories and migrate from one data store to the other.
# No magic. Hate it. Makes it hard to figure out and debug the code.
# Custom field types for database, particularly the ability to insert files and images into a database record. The actual files and images are stored in the scalable file system and automatically managed (basically automatic garbage collection for files).
# An image field type where you can change the image sizes (e.g. thumbnail, preview, full size) at will and images are automatically regenerated in those sizes the first time they are needed. This is one of those problems that comes out as you get bigger. Say your shopping cart app stores images in 3 sizes. Now you have 10 million products and you want to add another size. Without this, you are basically stuck with the decision you made at the start.
# Make easy things easy and make hard things easy. In Rails, the solution is usually to memorize the option that allows you to do what you want. With the right design, the problems should be solvable using just Ruby and the framework. This helps keep the framework light and easy to learn over time.
# Discoverable. The framework needs to be easy to figure out on ones own. Merb was more like this. Rails was less.
# Multi-threading. This was one of the primary reasons I didn't want to use Rails originally. I understand that threading is in Rails 2 now but I'm not sure how good it is.
# CouchDB. Okay, this feature is a little cutting edge but the biggest issue after launch is migrating the database. Add a new feature that requires a new column on a table with a bajillion records and you're doing many long midnight runs