fork vs. posix_spawn (vfork)

[Petar]

On Sunday, 14 October 2018 at 11:36:28 UTC, notna wrote:
> Hi D gurus.
>
> Did read an interesting post form GitLab [1] how they improved 
> performance by 30x by just going to go_v1.9... because they 
> again went from "fork" to "posix_spawn"...
>
> [1] 
> https://about.gitlab.com/2018/01/23/how-a-fix-in-go-19-sped-up-our-gitaly-service-by-30x/

Hi notna,

Thanks for the interesting article.

> I've searched the GitHub DLANG org for "posix_spawn" and didn't 
> find a hit...

Indeed currently (as of DMD 2.082.1) druntime does not include 
bindings of posix_spawn [1]. The nice thing about D is that if 
you need to use a C function that's not yet declared in druntime 
it's very easy to add the declaration to your app and use it.

I looked at glibc's implementation and translated the 
declarations. You can find them along with the example from the 
man page here: 
https://run.dlang.io/gist/ZombineDev/536eb8b456cc365b7ee05133ec452627?compiler=dmd
(if you want to use posix_spawn in your app/library copy the code 
from line 133 till the end of the file).

> so asking myself and you.... is DLANG still on "fork" and could 
> there be some performance improvement potential?

One cannot say whether D is "on fork" or not because unlike Go, 
everyone is free to use whatever system function they need, 
without explicit cooperation from upstream. Though probably many 
apps or libraries use the high-level functions declared in 
https://dlang.org/phobos/std_process instead of using the 
low-level system functions directly. In particular, spawnProcess 
[2] is currently implemented via fork/exec. I found an open issue 
about optimizing std.process: 
https://issues.dlang.org/show_bug.cgi?id=14770.

I'll look into making a pull request to druntime about 
posix_spawn, though no promises for phobos, as there the code is 
a bit more involved and would need more careful analysis.

[1]: http://man7.org/linux/man-pages/man3/posix_spawn.3.html
[2]: https://dlang.org/phobos/std_process#.spawnProcess