I am using a rather large and performance-intensive nodejs program to generate hinting data for CJK fonts (sfdhanautohint), and for some better dependency tracking I had to end up calling the nodejs program tens of thousands of times from a makefile like this.
This immediately brought me to the concern that doing such is actually putting a lot of overhead in starting and pre-heating the JIT engine, so I decided to find something like ngen.exe for nodejs. It appears that V8 already has some support for code caching, but is there anything I can do to use it in NodeJS?
Searching for kProduceCodeCache in NodeJS's GitHub repo doesn't return any non-bundled-v8 results. Perhaps it's time for a feature request…
Yes, this happens automatically. Node 5.7.0+ automatically pre-caches (pre-heats the JIT engine for your source) the first time you run your code (since PR #4845 / January 2016 here: https://github.com/nodejs/node/pull/4845).
It's important to note you can even pre-heat the pre-heat (before your code is ever even run on a machine, you can pre-cache your code and tell Node to load it).
Andres Suarez, a Facebook developer who works on Yarn, Atom and Babel created v8-compile-cache, which is a tiny little module that will JIT your code and require()s, and save your Node cache into your $TMP folder, and then use it if it's found. Check out the source for how it's done to suit other needs.
You can, if you'd like, have a little check that runs on start, and if the machine architecture is in your set of cache files, just load the cached files instead of letting Node JIT everything. This can cut your load time in half or more for a real-world large project with tons of requires, and it can do it on the very first run
Good for speeding up containers and getting them under that 500ms "microservice" boot time.
It's important to note:
Caches are binaries; they contain machine-executable code. They aren't your original JS code.
Node cache binaries are different for each target CPU you intend to run on (IA-32, IA-64, ARM etc). If you want to pre-cache pre-caches for your users, you must make cache targets for each target architecture you want to support.
Enjoy a ridiculous speed boost :)
Related
I'm packaging an application I have written into an AppImage so that it can be delivered to Linux users.
One of the key features of the GUI toolkit I'm using is that it is small and lightweight, allowing me to compile a build which is statically linked to the GUI library of around 6Mb.
However, after building the AppImage - where I do what the instructions say - use all the functionality (which basically includes only using file browser dialogues to load files) - it generates an absolutely enormous AppImage of around 200Mb!
I know that AppImages are supposed to be a "little bit" big, but this is completely mad as a proposed solution for portability when the natively compiled binary including a statically linked GUI toolkit is only 6Mb.
However, I'm not convinced at all that I need all of that 200Mb. A very similar piece of software to mine, but that additionally uses Qt (which is pretty bloated in comparison) is only about 30Mb. I actually suspect that appimage-builder is doing something very wrong - I think it is listing the files in the directory I explore when using the file browser dialogue as dependencies (they are big files). I have no real other explanation. But if so how do I stop it doing that?
Why is mine so big? What can I do about it?
For the record I am using this method for building my AppImage
Building my binary separately
Running appimage-builder --generate and completing the form
Running appimage-builder --recipe AppImageBuilder.yml --skip-tests
Edit: Removing the obviously not needed files that were being packaged have reduced the size of the appimage to just 140Mb, but this is still almost 5 times bigger than equivalent appimages I've seen. Are there some tricks/options I'm not aware of?
In few recent days got started with AppImage and faced the same problem.
Shortly: check dependencies of your app by any possible ways and configure recipe to include only concrete dependencies and avoid includings of any theme/icon/etc packages which are not realy used :)
My case is a small app, written in Dart (with Flutter). The built project itself weights about 22MB (du -sh . in output directory). My host os is Linux Mint (Cinnamon).
First time I run appimage-builder --generate it generated me the "recipe" with 17 packages to be installed and bunch of libraries to be copied from /lib/x86_64-linux-gnu/. When I generated AppImage from this recipe, result was about 105MB, which are extremely large in my opinion for small app.
My first experiments was to cleanup included files section, as I guess "all necessary" libraries should be installed from apt. I referred to a few configs from network where were marked only few libraries for include and was exclude section, which contains some DE related files (themes, fonts, icons and etc.)
Using that I got result about 50MB (which are still large enough).
Next experiments were referred to from this issue - https://github.com/AppImageCrafters/appimage-builder/issues/130#issuecomment-843288012
Shortly - after generating an AppImage file, there appeared file .bundle.yml file inside AppDir folder, which contains deployed libraries. Advice is to try exclude something from that. May be it's a good enough advice, but it takes too long time to check for each package/library if it breaks resulted AppImage file at least with official tests of appimage-builder (docker containers). I faced more broken results than any sane size reduction.
My next experiment was to reduce dependencies which should be installed from package manager and use files from host system. I deleted AppDir and appimage-builder-cache folders and regenerated the recipe. At next step I commented all packages which should be installed from package manager and leaved only included files. Result was fail, because of needing one package, but after adding it I got AppImage result in 36MB. That sounds much better than starting 105MB or even previous result of 50MB.
Here I got small "boost" - Flutter project built into AOT binaries, without runtime. So I checked output of ldd for my app, and then mapped list of required libraries to list of library files which were detected by appimage-builder. Finally some of them was correct, some not found in ldd output and some was in ldd output, but were not detected by appimage-builder. I added all undetected, removed all unused. My final result is 26MB and it passed all appimage-builder tests (running in docker images of fedora, cent, debian, ubuntu and arch)
I understand that it's bad enough for continuous building, because it will require to always check for used libraries and adapt config if something changed, but for rare enough builds I guess it's has some kind of balance between good and bad.
I am creating a simple linux kernel with buildroot and I am adding a small driver I've done myself, I created the Config.in file and drivername.mk to be able to select the driver in make menuconfig succesfully.
When executing make to build the image, the compilation goes correctly until my driver starts to compile, it looks to compile and create the image right but I get loooots of warnings saying that different files in ./lib/gcc/arm-buildroot-linux-uclibcgnueabihf/ are touched by more than one package: [u'host-gcc-initial', u'host-gcc-final'].
Anyone can explain me a bit about this issue and what is causing it? Do you need any more info to know what is happening? Is it safe to ignore them?
Thanks beforehand
Actually, doing a search on 'touched by more than one package', I found http://lists.busybox.net/pipermail/buildroot/2017-October/205602.html, where we find that this warning can safely be ignored if you're not doing a parallel build and aren't a kernel maintainer.
That said, if you're submitting code for inclusion in the Linux kernel, please be a good citizen and make sure you identify all of the things your code is dependent upon. (I'm not actually an active kernel hacker, so I don't know what method they're using for this right now.)
The basic idea is that there are a bunch of steps in compiling things that need to be done in a logical order. In a small project, we simply use dependencies that we know to put in because we also coded in that dependency. But with a project the size of the kernel, you can guarantee that not everyone does this. Some of them instead just specify dependencies if they're needed for things to build properly - if the default order works, things could go years before someone figures out that there was a missing dependency, causing them grief when they were trying to update just the one thing that was a missing dependency, and the other code not getting updated as a result.
When you're doing things in parallel, on the other hand, it becomes a lot more complicated. Now you really need to have every dependency specified, because there is no longer any inherent dependable order. Some people will probably still build serially, while others use two processing threads. I'll use 8. I've worked in groups that would be inclined to do 30, because they're on a 32 processor machine, and don't really need all of those during the off hours. Suddenly the fact that the file you needed from a directory that normally got processed 30 directories before yours is now getting processed at the same time as your file that needed it, because you didn't list the dependency and everything in those 30 directories that hasn't already been processed and isn't being processed has a dependency that's not yet finished its processing.
I have a complicated workflow of TS compilation and I want to make my watchers faster (and still smarts). I currently have 3 different TS compilation that are executed when Grunt starts but also on watch changes.
grunt-ts configuration:
https://gist.github.com/Vadorequest/f1fb95ab4bbc786f420b
grunt-watch configuration:
https://gist.github.com/Vadorequest/eaa82c292a5d3e1ee51f
It currently works. But it takes too much time to recompile every file each time a change is made in any TS file that belong to a set of files. I'm looking for a way to compile only what needs to be compiled, in a smart way. (Meaning that if A.ts inherits of B.ts, if B is changed then A should be recompiled too, it should be possible since WebStorm IDE is able to do that using its Files Watchers)
I read something about fast-compile on https://github.com/TypeStrong/grunt-ts#fast but it doesn't seem like I could use it, but I'm confused about it. (see https://github.com/TypeStrong/grunt-ts/issues/293)
I'm looking for a solution, and also for advices because I think my setup can be improved. It's great to have server side TS files and even shared TS files across the server and the client, but it adds a lot of compilation workflow which is hard to understand and to maintain. Maybe using the recent feature tsconfig.json would help? Any advice would be appreciated.
More details:
serverCommonJs: The server uses TS files that are compiled before to start the application, like controllers and models for instance.
clientCommonJs: Most of the client scripts are in CommonJs rather than in AMD because they're all concatened and minified and it's way easier to work with commonJS than AMD which requires a tons of setup.
amd: Some files are compiled in AMD, whether they're used in the server or the client, or both.
On my computer it takes about 1.5s to 2.5s to compile one set of file. Once compiled they're all copied into a temporary folder which is served to the browser (assets). So it takes easily 5 up to 10 seconds and it could be much much faster if only the changed files were compiled and copied.
I also have a similar issue with LESS files, but that's another story and it should be much simpler to fix since I only have one set of LESS files.
If you look at cached js files that get reloaded, you can see in the network panel that it literally takes NO time at all to reload them.
Why hassle with requireJS when you basically can load 3 MB of js out of the main memory in less than 1 microsecond
Why bother?
On one hand it's all about lazy loading of modules that isn't using frequently. On the other hand it's alway good practice to build amd modules in packages for production until http 2.0 will go in masses.
looking at jquery parse times which typically take 10-100 ms, depending on the device I would conclude that when there are actually many libraries, it is worth the hassle.
Probably in future, this need will degrade though (top end devices like iphone 5 have 10 ms parse time.)
Take the same code that sits on nodejs.org home page. Serve a static file that is 1.8Mb. And do the same with Nginx, and watch the difference.
Code : http://pastie.org/3730760
Screencast : http://screencast.com/t/Or44Xie11Fnp
Please share if you know anything that'd prevent this from happening, so we don't need to deploy nginx servers and complicate our lives.
ps1. this test is done with node 0.6.12. out of curiosity, i downgraded to 0.4.12 just to check if it's a regression, on the contrary, it was worse. same file used 25% twice.
ps2. this post is not a nodejs hate - we use nodejs, and we love it, except this glitch which actually delayed our launch (made us really sad), and seemed quite serious to me. i've never read, heard, seen or expected to come across.
The problem with your node benchmark is that you store the static file in a variable inside the V8 heap. Due to the way how V8 handles memory it can't directly send data contained in javascript variables to the network, because addresses of allocated objects may change during runtime, therefore V8 has to make a copy of your 1.8MB string on every request, sure that kills performance.
What you could do is to use a Buffer:
replace: longAssString = fs.readFileSync(pathToABigFile, 'utf8');
with: longAssString = fs.readFileSync(pathToABigFile);
that way you have your static file in a buffer, buffers are stored outside of V8s heap and require no copy when sent to the network and should therefore be much faster.