I've tried Node when the trend started and it was okay. But the problem was that the actual script was also the server. So if the script hanged, the whole server hanged. Went back to PHP and never looked back.
It sounds like you "tried Node" but didn't learn/read enough to understand the event loop architecture (and why node is asynchronous). Node is pretty simple (and powerful) but yes if you don't know about the event loop, you'll make a blocking call (synchronous function) and end up blocking the loop (which will hang your whole process).
Not that it matters because you don't care and no one uses promises any more but it does really sadden me how much nested promise code I've found over the years.
eg this
function sad(input) {
return new Promise((resolve, reject) =>
foo(input).then((data) => {
bar(data).then((data2) => {
baz(data2).then(resolve, reject)
})
})
);
}
could quite as easily have been written like this:
That's not inherent to PHP, but rather the ecosystem it's usually used in. The standard "LAMP" stack, for example, has Apache in it for the actual server, talking to PHP using a CGI interface. So if your PHP script crashes or hangs, the server itself is still up, and capable of serving other clients.
If you set up a Node script where e.g. Express talks directly to the clients, then yes, the script crashing or hanging means the server becomes unavailable or unresponsive. However, you can also set up a layer in front of Node. See cgi-node for replicating the CGI workflow you might be used to.
There are some advantages to the standard Node model though: the program can manage its own resources, such as keeping a database connection open; it can run asynchronous maintenance tasks; it can see and report the current server load; it can easily combine HTTP(S) communication and Web socket streams; etc.
> The standard "LAMP" stack, for example, has Apache in it for the actual server, talking to PHP using a CGI interface. So if your PHP script crashes or hangs, the server itself is still up, and capable of serving other clients.
Not exactly... the standard, typical LAMP stack makes use of mod_php, so the PHP engine is in-process with one of the Apache process.
The fact that Apache has multiprocess/hybrid workers is actually why the server stays up and can serve more requests.
Some contemporary LAMP stacks use FPM, I guess; most of those in shared hosting ISPs for example because of the possibility of running the script as a user process.
I would imagine that FPM is more common than people realize because it’s also usually faster, and running scripts as a separate user is more secure. For example, if PHP is a user with read-only access to the document root, it is far more difficult for an attacker to do file injection.
php has the opposite problem where if a script hangs on io, you can be cooked (assuming other requests blocked similarly).
Different trade offs for sure, but given IO is often the bottleneck, having it force you to think about running code asynchronously can often result in code that runs more in parallel than equivalent first blush php code.
For instance, if you were writing in a function style a program that fetched entries from a web service, for each of those results, run a summarization pass using another service, and then insert summarization result in a database, a style like shown in this article might serialize the summarization / insert operations, one after the other, where you’d instead want each result to be processed in parallel.
Lots of libraries in node that make that easy and the code straightforward to reason about.
You can make this same mistake in PHP blocking operations with 1 worker process in mod_php, php-fpm, or handling sockets directly. The fix is pretty standard: run multiple processes, threads, or use non-blocking operations.
The most common PHP setups are simply running multiple processes behind a httpd daemon, so this problem is less frequently encountered. But it's still lurking at a certain concurrency level.
I mean, there are tradeoffs with both approaches. It seems to me that Node’s stateful approach is used by other web servers and languages too.
The main tradeoff is you’re now reloading the entire server for every request in PHP. If you have a massive server or framework, that might not be the fastest thing in the world.
writing/thinking in FP style since 2020. Elixir, Js, F#, Julia and even PhP. Never got back to OOP. Never needed to. I realized what a waste of time and frustration had been OOP for me for almost 15 years (Java, C#).
Btw while this code doesn't win a prize, I agree with the author that functional style is worth it in many cases, even for a language as wordy and cumbersome as PHP. I still love PHP - its deployment story is still unparalleled and its standard library came batteries included since forever.
php with lambdas eliminates a bunch of the syntax and can let map/reduce style code be colocated while still achieving the functional style benefits.
A big issue though (at least for some apps), is there is no attempt to make that code parallelized for the io parts. So if you aren’t careful, you can end up with n+1 style IO calls, making things very slow.
Personally I consider that a benefit at least as often as not. In Nystromian terms, PHP functions are colorless, and that removes a whole kind of complexity (at the cost of, like you correctly point out, potential performance problems).
You can write OOP-style in C, FP-style in C/C++/Java or even PHP. Sure.
Writing pure FP code (pushing all the side effects to the "edges of the program" in C is possible! But it's going to take a lot of discipline from the programmers: there is no safety-net and lots of dreadful boilerplate.
tl;dr: Hardware is cheap. Engineers are expensive.
> Instead of fearing the overhead in PHP for function calls,
Already in 2015 in almost all applications this is a premature optimization which is rightly known as the root of all evil. Please. Your app will talk to the network, likely run a database query which is like thousands or millions of times slower than a function call. Further, even if there is a measurable difference the cost of hardware which makes that difference go away is very likely to be smaller and much smaller at that than the cost of the engineering hours wasted during maintenance when wrestling with code which was written with a "function calls are expensive" mindset.
This doesn't mean you don't need to worry about performance and scalability but even that is going to be much easier if you have a well structured code.
Well, in my particular experience with juniors, it's astonishing the so many ways they have to make any code run slower than it has to be (besides the network and database queries), just by thinking 'nah, this is a computer and it can do millions of ops by second'. And that's why I always take the "premature optimization" dogma with a grain of salt while my life's miserable enough having to accelerate some other's stuff.
IO has a very different cost than wasting CPU cycles, so shouldn’t necessarily be lumped together when analyzing tradeoffs.
Your point still stands that fundamentally changing the design of code should limited to the hottest parts.
Probably the biggest risk to this style map/reduce code is that you end up having IO deeply nested running synchronously versus grouping operations that can be run in parallel.
I've tried Node when the trend started and it was okay. But the problem was that the actual script was also the server. So if the script hanged, the whole server hanged. Went back to PHP and never looked back.
It sounds like you "tried Node" but didn't learn/read enough to understand the event loop architecture (and why node is asynchronous). Node is pretty simple (and powerful) but yes if you don't know about the event loop, you'll make a blocking call (synchronous function) and end up blocking the loop (which will hang your whole process).
I admit, i struggled with promises, within promises, within promises. Synchronous, I understood and just worked for me.
Not that it matters because you don't care and no one uses promises any more but it does really sadden me how much nested promise code I've found over the years.
eg this
could quite as easily have been written like this: But no one seemed to be aware that you can return a promise in the `.then()` callback and have it chain neatly...async functions return promises. everyone uses promises.
That's not inherent to PHP, but rather the ecosystem it's usually used in. The standard "LAMP" stack, for example, has Apache in it for the actual server, talking to PHP using a CGI interface. So if your PHP script crashes or hangs, the server itself is still up, and capable of serving other clients.
If you set up a Node script where e.g. Express talks directly to the clients, then yes, the script crashing or hanging means the server becomes unavailable or unresponsive. However, you can also set up a layer in front of Node. See cgi-node for replicating the CGI workflow you might be used to.
There are some advantages to the standard Node model though: the program can manage its own resources, such as keeping a database connection open; it can run asynchronous maintenance tasks; it can see and report the current server load; it can easily combine HTTP(S) communication and Web socket streams; etc.
> The standard "LAMP" stack, for example, has Apache in it for the actual server, talking to PHP using a CGI interface. So if your PHP script crashes or hangs, the server itself is still up, and capable of serving other clients.
Not exactly... the standard, typical LAMP stack makes use of mod_php, so the PHP engine is in-process with one of the Apache process.
The fact that Apache has multiprocess/hybrid workers is actually why the server stays up and can serve more requests.
Some contemporary LAMP stacks use FPM, I guess; most of those in shared hosting ISPs for example because of the possibility of running the script as a user process.
I would imagine that FPM is more common than people realize because it’s also usually faster, and running scripts as a separate user is more secure. For example, if PHP is a user with read-only access to the document root, it is far more difficult for an attacker to do file injection.
php has the opposite problem where if a script hangs on io, you can be cooked (assuming other requests blocked similarly).
Different trade offs for sure, but given IO is often the bottleneck, having it force you to think about running code asynchronously can often result in code that runs more in parallel than equivalent first blush php code.
For instance, if you were writing in a function style a program that fetched entries from a web service, for each of those results, run a summarization pass using another service, and then insert summarization result in a database, a style like shown in this article might serialize the summarization / insert operations, one after the other, where you’d instead want each result to be processed in parallel.
Lots of libraries in node that make that easy and the code straightforward to reason about.
You can make this same mistake in PHP blocking operations with 1 worker process in mod_php, php-fpm, or handling sockets directly. The fix is pretty standard: run multiple processes, threads, or use non-blocking operations.
The most common PHP setups are simply running multiple processes behind a httpd daemon, so this problem is less frequently encountered. But it's still lurking at a certain concurrency level.
This is one of the most underrated features of PHP.
I mean, there are tradeoffs with both approaches. It seems to me that Node’s stateful approach is used by other web servers and languages too.
The main tradeoff is you’re now reloading the entire server for every request in PHP. If you have a massive server or framework, that might not be the fastest thing in the world.
writing/thinking in FP style since 2020. Elixir, Js, F#, Julia and even PhP. Never got back to OOP. Never needed to. I realized what a waste of time and frustration had been OOP for me for almost 15 years (Java, C#).
(2015)
Btw while this code doesn't win a prize, I agree with the author that functional style is worth it in many cases, even for a language as wordy and cumbersome as PHP. I still love PHP - its deployment story is still unparalleled and its standard library came batteries included since forever.
php with lambdas eliminates a bunch of the syntax and can let map/reduce style code be colocated while still achieving the functional style benefits.
A big issue though (at least for some apps), is there is no attempt to make that code parallelized for the io parts. So if you aren’t careful, you can end up with n+1 style IO calls, making things very slow.
Personally I consider that a benefit at least as often as not. In Nystromian terms, PHP functions are colorless, and that removes a whole kind of complexity (at the cost of, like you correctly point out, potential performance problems).
You can write OOP-style in C, FP-style in C/C++/Java or even PHP. Sure.
Writing pure FP code (pushing all the side effects to the "edges of the program" in C is possible! But it's going to take a lot of discipline from the programmers: there is no safety-net and lots of dreadful boilerplate.
Impracticle, but possible.
tl;dr: Hardware is cheap. Engineers are expensive.
> Instead of fearing the overhead in PHP for function calls,
Already in 2015 in almost all applications this is a premature optimization which is rightly known as the root of all evil. Please. Your app will talk to the network, likely run a database query which is like thousands or millions of times slower than a function call. Further, even if there is a measurable difference the cost of hardware which makes that difference go away is very likely to be smaller and much smaller at that than the cost of the engineering hours wasted during maintenance when wrestling with code which was written with a "function calls are expensive" mindset.
This doesn't mean you don't need to worry about performance and scalability but even that is going to be much easier if you have a well structured code.
Well, in my particular experience with juniors, it's astonishing the so many ways they have to make any code run slower than it has to be (besides the network and database queries), just by thinking 'nah, this is a computer and it can do millions of ops by second'. And that's why I always take the "premature optimization" dogma with a grain of salt while my life's miserable enough having to accelerate some other's stuff.
IO has a very different cost than wasting CPU cycles, so shouldn’t necessarily be lumped together when analyzing tradeoffs.
Your point still stands that fundamentally changing the design of code should limited to the hottest parts.
Probably the biggest risk to this style map/reduce code is that you end up having IO deeply nested running synchronously versus grouping operations that can be run in parallel.
His code is crap, I'm not even sorry for saying this. For the case he is working with, OOP is absolute must.
OOP is never an absolute must. It's a design choice, one of several that can be taken.
OOP is a lousy joke.