www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.announce - Release: serverino - please destroy it.

reply Andrea Fontana <nospam example.com> writes:
Hello!

I've just released serverino. It's a small & ready-to-go 
http/https server.

Every request is processed by a worker running in an isolated 
process, no fibers/threads, sorry (or thanks?)

I did some tests and the performance sounds good: on a local 
machine it can handle more than 100_000 reqs/sec for a simple 
page containing just "hello world!".Of course that's not a good 
benchmark, if you can help me with other benchmarks it would be 
much appreciated (a big thanks to Tomáš Chaloupka who did some 
tests!)

I'm trying to keep it simple and easy to compile. It has no 
external deps in its base configuration and only one external 
library (libretls) is required if you need/want to enable https.

For your first project you need just three lines of code as you 
can see here:
https://github.com/trikko/serverino/

I didn't implement a traditional router for uris as probably many 
of you expected. I use a different approach. Check out this 
example: 
https://github.com/trikko/serverino/#defining-more-than-one-endpoint

This allows you to do some interesting things giving higher or 
lower priority to each endpoint (for example you can force 
something to always running first like redirect, logging, checks 
on login...)

Instead of using a lot of different UDAs to set routing rules, 
you can simply write them in your endpoint's body and exit from 
it to pass to the next endpoint.

Please help me testing it, I'm looking forward to receiving your 
shiny new issues on github.

Dub package: https://code.dlang.org/packages/serverino

Andrea
May 08
next sibling parent reply Andrea Fontana <nospam example.com> writes:
On Sunday, 8 May 2022 at 21:32:42 UTC, Andrea Fontana wrote:
 [...]
 Andrea
Whoops, I forgot a couple of things. This was tested on linux only and it should work fine on other posix systems (macOS included!). I don't have windows, but I think you need WSL to run it, since I'm using a lot of strange posix tricks to keep performace at a good level (like sending opened file descriptors between processes thru sockets). If you can test it on windows with WSL, that would be appreciated a lot! Andrea
May 08
parent reply Guillaume Piolat <first.last gmail.com> writes:
On Sunday, 8 May 2022 at 21:45:28 UTC, Andrea Fontana wrote:
 If you can test it on windows with WSL, that would be 
 appreciated a lot!
I tried to test servrino on WSL, but dub doesn't run on WSL. => https://github.com/dlang/dub/issues/2249
May 12
parent reply Andrea Fontana <nospam example.org> writes:
On Thursday, 12 May 2022 at 10:26:28 UTC, Guillaume Piolat wrote:
 On Sunday, 8 May 2022 at 21:45:28 UTC, Andrea Fontana wrote:
 If you can test it on windows with WSL, that would be 
 appreciated a lot!
I tried to test servrino on WSL, but dub doesn't run on WSL. => https://github.com/dlang/dub/issues/2249
Hey thanks for your support! Too bad dub doesn't work with wsl, it sounds like a lost opportunity. Does dmd/rdmd work? Serverino uses std.net.curl just for running its unittests, so maybe that bug is not blocking. Andrea
May 12
next sibling parent rikki cattermole <rikki cattermole.co.nz> writes:
On 12/05/2022 11:33 PM, Andrea Fontana wrote:
 Too bad dub doesn't work with wsl, it sounds like a lost opportunity.
 
 Does dmd/rdmd work? Serverino uses std.net.curl just for running its 
 unittests, so maybe that bug is not blocking.
It doesn't look like it is dub that is failing. This is a problem in Phobos/compiler.
May 12
prev sibling parent reply Guillaume Piolat <first.last gmail.com> writes:
On Thursday, 12 May 2022 at 11:33:07 UTC, Andrea Fontana wrote:
 Does dmd/rdmd work? Serverino uses std.net.curl just for 
 running its unittests, so maybe that bug is not blocking.
Well tbh, the simple fact that I would have to use WSL is a blocker for me. AFAIK vibe or cgi.d do not require that.
May 12
parent Andrea Fontana <nospam example.org> writes:
On Thursday, 12 May 2022 at 11:46:05 UTC, Guillaume Piolat wrote:
 On Thursday, 12 May 2022 at 11:33:07 UTC, Andrea Fontana wrote:
 Does dmd/rdmd work? Serverino uses std.net.curl just for 
 running its unittests, so maybe that bug is not blocking.
Well tbh, the simple fact that I would have to use WSL is a blocker for me. AFAIK vibe or cgi.d do not require that.
Yay. I need a Windows machine (or someone with it!) to rewrite some POSIX parts. For example the part that send/receive the file descriptor (of a socket) from the master process to the worker (windows has its own API for this)
May 12
prev sibling next sibling parent reply =?UTF-8?Q?Ali_=c3=87ehreli?= <acehreli yahoo.com> writes:
Congratulations! :) Looking forward to watching your presentation at 
DConf... ;)

On 5/8/22 14:32, Andrea Fontana wrote:

 Every request is processed by a worker running in an isolated process,
 no fibers/threads, sorry (or thanks?)
That effectively uses multiple GCs. I always suspected that approach would provide better latency.
 sending opened file descriptors between processes thru sockets
Sweet! Ali
May 08
next sibling parent Andrea Fontana <nospam example.org> writes:
On Sunday, 8 May 2022 at 22:09:37 UTC, Ali Çehreli wrote:
 Congratulations! :) Looking forward to watching your 
 presentation at DConf... ;)
I wish I was able to speak publicly in English in front of an audience :)
 On 5/8/22 14:32, Andrea Fontana wrote:

 Every request is processed by a worker running in an isolated
process,
 no fibers/threads, sorry (or thanks?)
That effectively uses multiple GCs. I always suspected that approach would provide better latency.
I think it depends on what your server is doing, anyway.
 sending opened file descriptors between processes thru sockets
I sent a pull request (merged!) for druntime to make this work on macOS too!
 Sweet!

 Ali
May 08
prev sibling parent reply Adam Ruppe <destructionator gmail.com> writes:
On Sunday, 8 May 2022 at 22:09:37 UTC, Ali Çehreli wrote:
 That effectively uses multiple GCs. I always suspected that 
 approach would provide better latency.
My cgi.d has used some fork approaches for a very long time since it is a very simple way to spread this out, it works quite well.
May 08
parent reply =?UTF-8?Q?Ali_=c3=87ehreli?= <acehreli yahoo.com> writes:
On 5/8/22 16:10, Adam Ruppe wrote:
 On Sunday, 8 May 2022 at 22:09:37 UTC, Ali Çehreli wrote:
 That effectively uses multiple GCs. I always suspected that approach
 would provide better latency.
My cgi.d has used some fork approaches for a very long time since it is a very simple way to spread this out, it works quite well.
While we are on topic :) and as I finally understood what generational GC is[1], are there any fundamental issues with D to not use one? Ali [1] Translating from what I wrote in the Turkish forum, here is my current understanding: Let's not waste time checking all allocated memory at every GC cycle. Instead, let's be smarter and assume that memory that survived through this GC cycle will survive the next cycle as well. Let's put those memory blocks aside to be reconsidered only when we really have to. This effectively makes the GC only play with short-lived objects, reducing the amount of memory touched. This would make some objects live forever, but GC never promises that all finalizers will be executed.
May 08
next sibling parent reply rikki cattermole <rikki cattermole.co.nz> writes:
On 09/05/2022 11:44 AM, Ali Çehreli wrote:
 While we are on topic :) and as I finally understood what generational 
 GC is[1], are there any fundamental issues with D to not use one?
This is not a D issue, its an implementation one. We don't have write barriers, that's it. Make them opt-in and we can have more advanced GC's. Oh and book recommendation for the subject: https://www.amazon.com/Garbage-Collection-Handbook-Management-Algorithms/dp/1420082795
May 08
parent reply "H. S. Teoh" <hsteoh quickfur.ath.cx> writes:
On Mon, May 09, 2022 at 12:10:53PM +1200, rikki cattermole via
Digitalmars-d-announce wrote:
 On 09/05/2022 11:44 AM, Ali ehreli wrote:
 While we are on topic :) and as I finally understood what
 generational GC is[1], are there any fundamental issues with D to
 not use one?
This is not a D issue, its an implementation one. We don't have write barriers, that's it. Make them opt-in and we can have more advanced GC's.
[...] In the past, the argument was that write barriers represented an unacceptable performance hit to D code. But I don't think this has ever actually been measured. (Or has it?) Maybe somebody should make a dmd fork that introduces write barriers, plus a generational GC (even if it's a toy, proof-of-concept-only implementation) to see if the performance hit is really as bad as believed to be. T -- The best way to destroy a cause is to defend it poorly.
May 08
next sibling parent reply =?UTF-8?Q?Ali_=c3=87ehreli?= <acehreli yahoo.com> writes:
On 5/8/22 17:25, H. S. Teoh wrote:

 somebody should make a dmd
 fork that introduces write barriers, plus a generational GC (even if
 it's a toy, proof-of-concept-only implementation) to see if the
 performance hit is really as bad as believed to be.
Ooh! DConf is getting even more interesting. :o) Ali
May 08
parent Bruce Carneal <bcarneal gmail.com> writes:
On Monday, 9 May 2022 at 00:32:33 UTC, Ali Çehreli wrote:
 On 5/8/22 17:25, H. S. Teoh wrote:

 somebody should make a dmd
 fork that introduces write barriers, plus a generational GC
(even if
 it's a toy, proof-of-concept-only implementation) to see if
the
 performance hit is really as bad as believed to be.
Ooh! DConf is getting even more interesting. :o) Ali
A helpful paper: "Getting to Go: The Journey of Go's garbage collector". Positive highlights: 1) non-copying 2) no read barriers Less friendly: 1) write barriers 2) GC aware fiber scheduler 3) other??? Would be some (huge amount?) of work but porting/enabling an opt-in golang latency GC could be a big enabler for the casual/soft "real time" crowd. Here's a link to the paper: https://go.dev/blog/ismmkeynote
May 08
prev sibling parent reply Vladimir Panteleev <thecybershadow.lists gmail.com> writes:
On Monday, 9 May 2022 at 00:25:43 UTC, H. S. Teoh wrote:
 In the past, the argument was that write barriers represented 
 an unacceptable performance hit to D code.  But I don't think 
 this has ever actually been measured. (Or has it?)  Maybe 
 somebody should make a dmd fork that introduces write barriers, 
 plus a generational GC (even if it's a toy, 
 proof-of-concept-only implementation) to see if the performance 
 hit is really as bad as believed to be.
Implementing write barriers in the compiler (by instrumenting code) means that you're no longer allowed to copy pointers to managed memory in non-D code. This is a stricter assumption that the current ones we have; for instance, copying a struct (which has indirections) with memcpy would be forbidden.
May 08
parent "H. S. Teoh" <hsteoh quickfur.ath.cx> writes:
On Mon, May 09, 2022 at 05:55:39AM +0000, Vladimir Panteleev via
Digitalmars-d-announce wrote:
 On Monday, 9 May 2022 at 00:25:43 UTC, H. S. Teoh wrote:
 In the past, the argument was that write barriers represented an
 unacceptable performance hit to D code.  But I don't think this has
 ever actually been measured. (Or has it?)  Maybe somebody should
 make a dmd fork that introduces write barriers, plus a generational
 GC (even if it's a toy, proof-of-concept-only implementation) to see
 if the performance hit is really as bad as believed to be.
Implementing write barriers in the compiler (by instrumenting code) means that you're no longer allowed to copy pointers to managed memory in non-D code. This is a stricter assumption that the current ones we have; for instance, copying a struct (which has indirections) with memcpy would be forbidden.
Hmm, true. That puts a big damper on the possibilities... OTOH, if this could be made an optional feature, then code that we know doesn't need, e.g., passing pointers to C code, can take advantage of possibly better GC strategies. T -- English has the lovely word "defenestrate", meaning "to execute by throwing someone out a window", or more recently "to remove Windows from a computer and replace it with something useful". :-) -- John Cowan
May 09
prev sibling parent reply Vladimir Panteleev <thecybershadow.lists gmail.com> writes:
On Sunday, 8 May 2022 at 23:44:42 UTC, Ali Çehreli wrote:
 While we are on topic :) and as I finally understood what 
 generational GC is[1], are there any fundamental issues with D 
 to not use one?
I implemented one a long time ago. The only way to get write barriers with D is memory protection. It worked, but unfortunately the write barriers caused a severe performance penalty. It's possible that it might be viable with more tweaking, or in certain applications where most of the heap is not written to; I did not experiment a lot with it.
May 08
parent reply "H. S. Teoh" <hsteoh quickfur.ath.cx> writes:
On Mon, May 09, 2022 at 05:52:30AM +0000, Vladimir Panteleev via
Digitalmars-d-announce wrote:
 On Sunday, 8 May 2022 at 23:44:42 UTC, Ali ehreli wrote:
 While we are on topic :) and as I finally understood what
 generational GC is[1], are there any fundamental issues with D to
 not use one?
I implemented one a long time ago. The only way to get write barriers with D is memory protection. It worked, but unfortunately the write barriers caused a severe performance penalty.
Why is memory protection the only way to implement write barriers in D?
 It's possible that it might be viable with more tweaking, or in
 certain applications where most of the heap is not written to; I did
 not experiment a lot with it.
Interesting data point, in any case. T -- The early bird gets the worm. Moral: ewww...
May 09
parent reply Vladimir Panteleev <thecybershadow.lists gmail.com> writes:
On Monday, 9 May 2022 at 16:37:15 UTC, H. S. Teoh wrote:
 Why is memory protection the only way to implement write 
 barriers in D?
Well, it's the only way I know of without making it a major backwards-incompatible change. The main restriction in this area is that it must continue working with code written in other languages, and generally not affect the ABI drastically.
May 09
parent "H. S. Teoh" <hsteoh quickfur.ath.cx> writes:
On Mon, May 09, 2022 at 04:48:11PM +0000, Vladimir Panteleev via
Digitalmars-d-announce wrote:
 On Monday, 9 May 2022 at 16:37:15 UTC, H. S. Teoh wrote:
 Why is memory protection the only way to implement write barriers in
 D?
Well, it's the only way I know of without making it a major backwards-incompatible change. The main restriction in this area is that it must continue working with code written in other languages, and generally not affect the ABI drastically.
Ah, gotcha. Yeah, I don't think such an approach would be fruitful (it was worth a shot, though!). If D were ever to get write barriers, they'd have to be in some other form, probably more intrusive in terms of backwards-compatibility and ABI. T -- Curiosity kills the cat. Moral: don't be the cat.
May 09
prev sibling next sibling parent reply Guillaume Piolat <first.last gmail.com> writes:
On Sunday, 8 May 2022 at 21:32:42 UTC, Andrea Fontana wrote:
 Hello!

 I've just released serverino. It's a small & ready-to-go 
 http/https server.

 Dub package: https://code.dlang.org/packages/serverino

 Andrea
Looks very useful, congratulations!
May 09
parent reply Andrea Fontana <nospam example.com> writes:
On Monday, 9 May 2022 at 19:09:40 UTC, Guillaume Piolat wrote:
 On Sunday, 8 May 2022 at 21:32:42 UTC, Andrea Fontana wrote:
 Hello!

 I've just released serverino. It's a small & ready-to-go 
 http/https server.

 Dub package: https://code.dlang.org/packages/serverino

 Andrea
Looks very useful, congratulations!
Thank you. Looking forward to getting feedback, bug reports and help :) Andrea
May 09
parent reply Adam Ruppe <destructionator gmail.com> writes:
On Monday, 9 May 2022 at 19:20:27 UTC, Andrea Fontana wrote:
 Thank you. Looking forward to getting feedback, bug reports and 
 help :)
BTW I'm curious, what made you not want to use my cgi.d which has similar capabilities?
May 10
parent Andrea Fontana <nospam example.com> writes:
On Tuesday, 10 May 2022 at 15:01:43 UTC, Adam Ruppe wrote:
 On Monday, 9 May 2022 at 19:20:27 UTC, Andrea Fontana wrote:
 Thank you. Looking forward to getting feedback, bug reports 
 and help :)
BTW I'm curious, what made you not want to use my cgi.d which has similar capabilities?
I was really tempted to start from that! But it's difficult to fork and edit a 11kloc project like that :) I had yet developed fastcgi and scgi code in the past so I've reused some code and it didn't take so much time to get to serverino. Andrea
May 10
prev sibling next sibling parent reply Sebastiaan Koppe <mail skoppe.eu> writes:
On Sunday, 8 May 2022 at 21:32:42 UTC, Andrea Fontana wrote:
 Every request is processed by a worker running in an isolated 
 process, no fibers/threads, sorry (or thanks?)

 I did some tests and the performance sounds good: on a local 
 machine it can handle more than 100_000 reqs/sec for a simple 
 page containing just "hello world!".Of course that's not a good 
 benchmark, if you can help me with other benchmarks it would be 
 much appreciated (a big thanks to Tomáš Chaloupka who did some 
 tests!)
Typically server applications are IO heavy. I expect your isolated-process approach to break down with that kind of work. As an example, how many requests per second can you manage if all requests have to wait 100 msecs? For non critical workload you will probably still hit good enough performance though.
 Instead of using a lot of different UDAs to set routing rules, 
 you can simply write them in your endpoint's body and exit from 
 it to pass to the next endpoint.
My experience is that exhaustive matching is easier to reason about at larger scale.
 Please help me testing it, I'm looking forward to receiving 
 your shiny new issues on github.
I noticed it has zero unittests, that is probably a good place to start.
May 09
parent reply Andrea Fontana <nospam example.com> writes:
On Monday, 9 May 2022 at 20:08:38 UTC, Sebastiaan Koppe wrote:
 On Sunday, 8 May 2022 at 21:32:42 UTC, Andrea Fontana wrote:
 Every request is processed by a worker running in an isolated 
 process, no fibers/threads, sorry (or thanks?)

 I did some tests and the performance sounds good: on a local 
 machine it can handle more than 100_000 reqs/sec for a simple 
 page containing just "hello world!".Of course that's not a 
 good benchmark, if you can help me with other benchmarks it 
 would be much appreciated (a big thanks to Tomáš Chaloupka who 
 did some tests!)
Typically server applications are IO heavy. I expect your isolated-process approach to break down with that kind of work.
I know. We all know :) Benchmarks are just benchmarks. They are useful to understand how much overhead your server adds to the whole project. These benchmarks are made in the local machine, with almost no connection overhead. Not every application is IO heavy, anyway.
 As an example, how many requests per second can you manage if 
 all requests have to wait 100 msecs?

 For non critical workload you will probably still hit good 
 enough performance though.
Firstly, it depends on how many workers you have. Then you should consider that a lot of (most?) websites use php-fpm, that works using the same approach (but php is much slower than D). The same goes for cgi/fastcgi/scgi and so on. Let's say you have just 20 workers. 100msecs for each request (a lot of time for my standards, I would say). That means 20*10 = 200 webpages/s = 720k pages/h. I don't think your website has so much traffic... And I hope not every request will take 100msecs!
 Instead of using a lot of different UDAs to set routing rules, 
 you can simply write them in your endpoint's body and exit 
 from it to pass to the next endpoint.
My experience is that exhaustive matching is easier to reason about at larger scale.
Yes, but exactly the same thing can be done without uda. ``` endpoint void my_end(Request r, Output o) { if (r.uri == "/asd") // or whatever you want: regex, or checking another field return false; // } ``` This is just like: ``` matchuda(uri, "/asd") void my_end(....) { ... } ``` What's the difference? The first one is much more flexible, IMHO.
 Please help me testing it, I'm looking forward to receiving 
 your shiny new issues on github.
 I noticed it has zero unittests, that is probably a good place 
 to start.
Of course! They will come for sure. :) Andrea
May 09
next sibling parent reply Sebastiaan Koppe <mail skoppe.eu> writes:
On Monday, 9 May 2022 at 20:37:50 UTC, Andrea Fontana wrote:
 On Monday, 9 May 2022 at 20:08:38 UTC, Sebastiaan Koppe wrote:
 As an example, how many requests per second can you manage if 
 all requests have to wait 100 msecs?

 For non critical workload you will probably still hit good 
 enough performance though.
Firstly, it depends on how many workers you have. Then you should consider that a lot of (most?) websites use php-fpm, that works using the same approach (but php is much slower than D). The same goes for cgi/fastcgi/scgi and so on. Let's say you have just 20 workers. 100msecs for each request (a lot of time for my standards, I would say). That means 20*10 = 200 webpages/s = 720k pages/h. I don't think your website has so much traffic... And I hope not every request will take 100msecs!
100msecs is on the upper end for sure, but if you add a database, external service call, etc. it is not uncommon to reach that. The point however, is that the architecture breaks down because it is unable to do work concurrently. Every requests blocks a worker from start to finish. Unless it is CPU heavy the system will be under utilized. That is not necessarily bad though. The simplicity has something going for it, but it is definitely a tradeoff that you should consider highlighting.
 ```
      endpoint void my_end(Request r, Output o)
     {
          if (r.uri == "/asd") // or whatever you want: regex, 
 or checking another field
             return false; //
     }
 ```

 This is just like:

 ```
  matchuda(uri, "/asd") void my_end(....) { ... }
 ```

 What's the difference? The first one is much more flexible, 
 IMHO.
The difference is that with the route uda you can *only* map routes 1:1 exhaustively. With your approach it is up to the programmer to avoid errors. It is also hard to reason about the flow of requests through all those functions, and you have to look at the body of them to determine what will happen.
May 10
parent reply Andrea Fontana <nospam example.com> writes:
On Tuesday, 10 May 2022 at 08:32:15 UTC, Sebastiaan Koppe wrote:
 On Monday, 9 May 2022 at 20:37:50 UTC, Andrea Fontana wrote:
 On Monday, 9 May 2022 at 20:08:38 UTC, Sebastiaan Koppe wrote:
 As an example, how many requests per second can you manage if 
 all requests have to wait 100 msecs?

 For non critical workload you will probably still hit good 
 enough performance though.
Firstly, it depends on how many workers you have. Then you should consider that a lot of (most?) websites use php-fpm, that works using the same approach (but php is much slower than D). The same goes for cgi/fastcgi/scgi and so on. Let's say you have just 20 workers. 100msecs for each request (a lot of time for my standards, I would say). That means 20*10 = 200 webpages/s = 720k pages/h. I don't think your website has so much traffic... And I hope not every request will take 100msecs!
100msecs is on the upper end for sure, but if you add a database, external service call, etc. it is not uncommon to reach that.
And you can still handle 700k/views per hour with 20 workers!
 The point however, is that the architecture breaks down because 
 it is unable to do work concurrently. Every requests blocks a 
 worker from start to finish.

 Unless it is CPU heavy the system will be under utilized. That 
 is not necessarily bad though. The simplicity has something 
 going for it, but it is definitely a tradeoff that you should 
 consider highlighting.
Every server has its own target. BTW, I'm not developing serverino to use it as a building block of a CDN. In real-life projects, I think you can use it without any problem for not-huge projects. You can also put it under a reverse proxy (f.e. nginx), to handle just the requests you need to write in D.
 The difference is that with the route uda you can *only* map 
 routes 1:1 exhaustively. With your approach it is up to the 
 programmer to avoid errors. It is also hard to reason about the 
 flow of requests through all those functions, and you have to 
 look at the body of them to determine what will happen.
Sorry I don't follow you: I don't know which framework you're using, but if you're using UDA with matches (something like: matchUri("/main") void renderMain(...) { ... }) you still have to check all the functions if a request is not handled correctly. Or am I missing something? Using my approach if you want to check which functions escape from routing you can just add a catch-all endpoint with low priority. ``` priority(-1000) endpoint void wtf(Request r, Output o) { fatal("Request NOT HANDLED: ", r.dump()); } ``` And if a request doesn't match your UDA constraint, how do you debug what's wrong with it? I think it's easier to add a checkpoint/log on the first lines of your functions body to guess why the function is skipped. In any case if you want to use a different routing strategy it's quite easy. I really don't like libraries that force you to use their own style/way. So you can even drop my UDAs and write the app like this. It still works: ``` mixin ServerinoMain; void entry(Request r, Output o) { // Use your routing strategy here // ... // YourRouter router; // router.do(r, "/hello/world", &yourFunction); // router.do(r, "/bla", &hello); } ``` Andrea
May 10
next sibling parent reply Ola Fosheim =?UTF-8?B?R3LDuHN0YWQ=?= <ola.fosheim.grostad gmail.com> writes:
On Tuesday, 10 May 2022 at 10:49:06 UTC, Andrea Fontana wrote:
 And you can still handle 700k/views per hour with 20 workers!
Requests tend to come in bursts from the same client, thanks to clunky javascript APIs and clutters of resources (and careless web developers). For a typical D user ease-of-use is probably more important at this point, though, so good luck with your project!
May 10
parent reply Andrea Fontana <nospam example.com> writes:
On Tuesday, 10 May 2022 at 12:31:23 UTC, Ola Fosheim Grøstad 
wrote:
 On Tuesday, 10 May 2022 at 10:49:06 UTC, Andrea Fontana wrote:
 And you can still handle 700k/views per hour with 20 workers!
Requests tend to come in bursts from the same client, thanks to clunky javascript APIs and clutters of resources (and careless web developers). For a typical D user ease-of-use is probably more important at this point, though, so good luck with your project!
In my opnioni IRL that's not a big problem as it can seem. Again: that's just how nginx and apache handle php/cgi/fcgi/scgi requests. Wikipedia runs wikimedia software. Written in php. Running on apache with php-fpm (and cache!). And I'm not suggesting to run wikipedia on serverino, *for now*. If you try to open a lot of wikipedia pages at the same time in a burst, they will be served (probably using keep-alive connection) not in parallel: you're queued. And the 99.9% of users will never notice this. Is it a problem? If you need much control, you can use an http accelerator and/or you can use a reverse proxy (like nginx) to control bursts et similia. I'm running a whole website in D using fastcgi and we have no problem at all, it's blazing fast. But it's not so easy to setup as serverino :) Andrea
May 10
parent reply Ola Fosheim =?UTF-8?B?R3LDuHN0YWQ=?= <ola.fosheim.grostad gmail.com> writes:
On Tuesday, 10 May 2022 at 12:52:01 UTC, Andrea Fontana wrote:
 I'm running a whole website in D using fastcgi and we have no 
 problem at all, it's blazing fast. But it's not so easy to 
 setup as serverino :)
Easy setup is probably the number one reason people land on a specific web-tech, so it is the best initial angle, I agree. (By version 3.x you know what the practical weak spots are and can rethink the bottom layer.)
May 10
parent reply Andrea Fontana <nospam example.com> writes:
On Tuesday, 10 May 2022 at 13:15:38 UTC, Ola Fosheim Grøstad 
wrote:
 On Tuesday, 10 May 2022 at 12:52:01 UTC, Andrea Fontana wrote:
 I'm running a whole website in D using fastcgi and we have no 
 problem at all, it's blazing fast. But it's not so easy to 
 setup as serverino :)
Easy setup is probably the number one reason people land on a specific web-tech, so it is the best initial angle, I agree. (By version 3.x you know what the practical weak spots are and can rethink the bottom layer.)
Right. But it's not just marketing. I work in the R&D and every single time I even have to write a small api or a simple html interface to control some strange machine I think "omg, I have to set nginx agaaaaaain". It's pretty annoying especially if you're working on shared aws machine. (I know, docker & c. Exist, but they take a lot to setup and they are heavy for some simple api). I'm going to love serverino in the next months :)
May 10
parent reply Ola Fosheim =?UTF-8?B?R3LDuHN0YWQ=?= <ola.fosheim.grostad gmail.com> writes:
On Tuesday, 10 May 2022 at 15:00:06 UTC, Andrea Fontana wrote:
 I work in the R&D and every single time I even have to write a 
 small api or a simple html interface to control some strange 
 machine I think "omg, I have to set nginx agaaaaaain".
Good point, there are more application areas than regular websites. Embedded remote applications could be another application area where you want something simple with HTTPS (monitoring webcams, sensors, solar panels, supervising farming houses or whatever).
May 10
parent reply Andrea Fontana <nospam example.com> writes:
On Tuesday, 10 May 2022 at 15:16:22 UTC, Ola Fosheim Grøstad 
wrote:
 On Tuesday, 10 May 2022 at 15:00:06 UTC, Andrea Fontana wrote:
 I work in the R&D and every single time I even have to write a 
 small api or a simple html interface to control some strange 
 machine I think "omg, I have to set nginx agaaaaaain".
Good point, there are more application areas than regular websites. Embedded remote applications could be another application area where you want something simple with HTTPS (monitoring webcams, sensors, solar panels, supervising farming houses or whatever).
Indeed the "-ino" suffix in "serverino" stands for "small" in italian. :) Andrea
May 10
parent reply Ola Fosheim =?UTF-8?B?R3LDuHN0YWQ=?= <ola.fosheim.grostad gmail.com> writes:
On Tuesday, 10 May 2022 at 15:27:48 UTC, Andrea Fontana wrote:
 Indeed the "-ino" suffix in "serverino" stands for "small" in 
 italian. :)
Bambino > bambinello? So, the embedded-version could be «serverinello»? :O)
May 10
parent reply Andrea Fontana <nospam example.com> writes:
On Tuesday, 10 May 2022 at 15:35:35 UTC, Ola Fosheim Grøstad 
wrote:
 On Tuesday, 10 May 2022 at 15:27:48 UTC, Andrea Fontana wrote:
 Indeed the "-ino" suffix in "serverino" stands for "small" in 
 italian. :)
Bambino > bambinello? So, the embedded-version could be «serverinello»? :O)
Oh, italian is full of suffixes. -ello means a slightly different thing. It's small but sounds like a bit pejorative. -ino in bambino is not (anymore) a suffix, anyway. Andrea
May 10
next sibling parent reply Ola Fosheim =?UTF-8?B?R3LDuHN0YWQ=?= <ola.fosheim.grostad gmail.com> writes:
On Tuesday, 10 May 2022 at 16:05:11 UTC, Andrea Fontana wrote:
 Oh, italian is full of suffixes. -ello means a slightly 
 different thing. It's small but sounds like a bit pejorative.
Oh, and I loved the sound of it… suggests immaturity, perhaps? (I love the -ello and -ella endings. «Bambinella» is one of my favourite words, turns out it is a fruit too!)
May 10
parent reply Andrea Fontana <nospam example.org> writes:
On Tuesday, 10 May 2022 at 16:47:13 UTC, Ola Fosheim Grøstad 
wrote:
 On Tuesday, 10 May 2022 at 16:05:11 UTC, Andrea Fontana wrote:
 Oh, italian is full of suffixes. -ello means a slightly 
 different thing. It's small but sounds like a bit pejorative.
Oh, and I loved the sound of it… suggests immaturity, perhaps? (I love the -ello and -ella endings. «Bambinella» is one of my favourite words, turns out it is a fruit too!)
Maybe bambinetto is more about immaturity. Bambinuccio is cute. Bambinaccio is bad. Bambinone is big (an adult that behave like a child). -ello doesn't sound good with bambino, but it's very similar to -etto. Good luck :)
May 10
parent Ola Fosheim =?UTF-8?B?R3LDuHN0YWQ=?= <ola.fosheim.grostad gmail.com> writes:
On Tuesday, 10 May 2022 at 19:24:25 UTC, Andrea Fontana wrote:
 Maybe bambinetto is more about immaturity. Bambinuccio is cute. 
 Bambinaccio is bad. Bambinone is big (an adult that behave like 
 a child). -ello doesn't sound good with bambino, but it's very 
 similar to -etto.

 Good luck :)
Thanks for the explanation! <3 If only programming languages were this expressive! «Servinuccio»… ;P
May 11
prev sibling parent reply Paolo Invernizzi <paolo.invernizzi gmail.com> writes:
On Tuesday, 10 May 2022 at 16:05:11 UTC, Andrea Fontana wrote:
 On Tuesday, 10 May 2022 at 15:35:35 UTC, Ola Fosheim Grøstad 
 wrote:
 On Tuesday, 10 May 2022 at 15:27:48 UTC, Andrea Fontana wrote:
 Indeed the "-ino" suffix in "serverino" stands for "small" in 
 italian. :)
Bambino > bambinello? So, the embedded-version could be «serverinello»? :O)
Oh, italian is full of suffixes. -ello means a slightly different thing. It's small but sounds like a bit pejorative. -ino in bambino is not (anymore) a suffix, anyway. Andrea
Concordo ... (I agree!) :-P
May 10
parent reply Andrea Fontana <nospam example.com> writes:
On Tuesday, 10 May 2022 at 19:50:08 UTC, Paolo Invernizzi wrote:
 Concordo ... (I agree!)

 :-P
Wait, you have always said you're not Italian. Have you changed your mind? Andrea
May 10
parent reply Paolo Invernizzi <paolo.invernizzi gmail.com> writes:
On Tuesday, 10 May 2022 at 19:55:32 UTC, Andrea Fontana wrote:
 On Tuesday, 10 May 2022 at 19:50:08 UTC, Paolo Invernizzi wrote:
 Concordo ... (I agree!)

 :-P
Wait, you have always said you're not Italian. Have you changed your mind? Andrea
Sinceramente non ricordo di averlo scritto, ma alla mia eta ... probabilmente dimentico qualcosa ... comunque piacere! E' bello vedere altri italiani apprezzare questo magnifico linguaggio! (Frankly speaking, I don't remember to have written that, but hey, I'm getting old ... probably I'm forgetting something ... anyway nice to meet you! It's great to see Italians here enjoying this great programming language!)
May 10
parent reply Andrea Fontana <nospam example.com> writes:
On Tuesday, 10 May 2022 at 20:13:45 UTC, Paolo Invernizzi wrote:
 Sinceramente non ricordo di averlo scritto, ma alla mia eta ... 
 probabilmente dimentico qualcosa ... comunque piacere! E' bello 
 vedere altri italiani apprezzare questo magnifico linguaggio!

 (Frankly speaking, I don't remember to have written that, but 
 hey, I'm getting old ... probably  I'm forgetting something ... 
 anyway nice to meet you! It's great to see Italians here 
 enjoying this great programming language!)
I wonder if you're making a fool of me. Or maybe it's me who is getting old. I'm pretty sure that there's a user here with a really Italian name who was born somewhere in South America. Andrea
May 10
parent reply Paolo Invernizzi <paolo.invernizzi gmail.com> writes:
On Tuesday, 10 May 2022 at 20:41:17 UTC, Andrea Fontana wrote:
 On Tuesday, 10 May 2022 at 20:13:45 UTC, Paolo Invernizzi wrote:
 Sinceramente non ricordo di averlo scritto, ma alla mia eta 
 ... probabilmente dimentico qualcosa ... comunque piacere! E' 
 bello vedere altri italiani apprezzare questo magnifico 
 linguaggio!

 (Frankly speaking, I don't remember to have written that, but 
 hey, I'm getting old ... probably  I'm forgetting something 
 ... anyway nice to meet you! It's great to see Italians here 
 enjoying this great programming language!)
I wonder if you're making a fool of me. Or maybe it's me who is getting old. I'm pretty sure that there's a user here with a really Italian name who was born somewhere in South America. Andrea
Here I am ... Milanese: https://www.deepglance.com/about /Paolo
May 10
parent Andrea Fontana <nospam example.org> writes:
On Tuesday, 10 May 2022 at 21:24:46 UTC, Paolo Invernizzi wrote:
 Here I am ... Milanese: https://www.deepglance.com/about

 /Paolo
Ok it's me getting old! Andrea
May 10
prev sibling parent reply Sebastiaan Koppe <mail skoppe.eu> writes:
On Tuesday, 10 May 2022 at 10:49:06 UTC, Andrea Fontana wrote:
 On Tuesday, 10 May 2022 at 08:32:15 UTC, Sebastiaan Koppe wrote:
 The difference is that with the route uda you can *only* map 
 routes 1:1 exhaustively. With your approach it is up to the 
 programmer to avoid errors. It is also hard to reason about 
 the flow of requests through all those functions, and you have 
 to look at the body of them to determine what will happen.
Sorry I don't follow you
It is simple, since all your handler are effectively chained, any error in any one of them can cause later ones to fail or misbehave. This decreases locality and increases the things you have to reason about. There are other benefits to uda tagged endpoints too, for example they are easier to nest, or to programmatically generate them. In vibe-d I added the default option of generating OPTION handlers for every regular endpoint. This is required for CORS.
 In any case if you want to use a different routing strategy 
 it's quite easy. I really don't like libraries that force you 
 to use their own style/way.
That is good.
May 10
parent Andrea Fontana <nospam example.org> writes:
On Tuesday, 10 May 2022 at 18:33:18 UTC, Sebastiaan Koppe wrote:
 On Tuesday, 10 May 2022 at 10:49:06 UTC, Andrea Fontana wrote:
 On Tuesday, 10 May 2022 at 08:32:15 UTC, Sebastiaan Koppe 
 wrote:
 The difference is that with the route uda you can *only* map 
 routes 1:1 exhaustively. With your approach it is up to the 
 programmer to avoid errors. It is also hard to reason about 
 the flow of requests through all those functions, and you 
 have to look at the body of them to determine what will 
 happen.
Sorry I don't follow you
It is simple, since all your handler are effectively chained, any error in any one of them can cause later ones to fail or misbehave. This decreases locality and increases the things you have to reason about.
Not sure. What if your uda (regex) match is too permissive? Is that different? My code evaluates workers in order, just like yours, no? Maybe I can enable some log if set on config, to track what's happening. That could help you to debug if something goes wrong.
 There are other benefits to uda tagged endpoints too, for 
 example they are easier to nest, or to programmatically 
 generate them. In vibe-d I added the default option of 
 generating OPTION handlers for every regular endpoint. This is 
 required for CORS.
endpoint void func(...){ if(req.method == Method.OPTION){ // THIS RUN FOR EVERY ENDPOINT } }
 In any case if you want to use a different routing strategy 
 it's quite easy. I really don't like libraries that force you 
 to use their own style/way.
That is good.
Andrea
May 10
prev sibling parent reply Adam D Ruppe <destructionator gmail.com> writes:
On Monday, 9 May 2022 at 20:37:50 UTC, Andrea Fontana wrote:
 The same goes for cgi/fastcgi/scgi and so on.
Well, cgi does one process per request, so there is no worker pool (it is the original "serverless" lol). fastcgi is interesting because the Apache module for it will actually start and stop worker processes as-needed. I don't think the the nginx impl does that though. But the nicest thing about all these application models is if you write it in the right way, you can swap out the approach, either transparently adding the i/o event waits or just adding additional servers without touching the application code. That's a lot harder to do when you expect shared state etc. like other things encourage.
May 10
parent Andrea Fontana <nospam example.com> writes:
On Tuesday, 10 May 2022 at 13:34:27 UTC, Adam D Ruppe wrote:
 On Monday, 9 May 2022 at 20:37:50 UTC, Andrea Fontana wrote:
 The same goes for cgi/fastcgi/scgi and so on.
Well, cgi does one process per request, so there is no worker pool (it is the original "serverless" lol). fastcgi is interesting because the Apache module for it will actually start and stop worker processes as-needed. I don't think the the nginx impl does that though.
Some daemons can manage this by themselves (once again: check php-fpm "dynamic" setting). Serverino can do it as well. You can set in configuration the max and min number of workers. It's easy: ``` onServerInit auto setup() { ServerinoConfig sc = ServerinoConfig.create(); sc.setMinWorkers(5); sc.setMaxWorkers(100); return sc; } ``` If all workers are busy the daemon will launch a new one. You might be interested in setMaxWorkerLifetime() and sc.setMaxWorkerIdling() too!
 But the nicest thing about all these application models is if 
 you write it in the right way, you can swap out the approach, 
 either transparently adding the i/o event waits or just adding 
 additional servers without touching the application code. 
 That's a lot harder to do when you expect shared state etc. 
 like other things encourage.
I would mention that if something goes wrong and a process crash or get caught in an infinite loop, it's not a problem. Process is killed and wake up again without pull all the server down. Andrea
May 10
prev sibling next sibling parent reply Orfeo <dforum orfeo.fastmail.com> writes:
well done Andrea!

(forum begins to be too crowded with Italians :) )


---
Orfeo
May 10
parent Andrea Fontana <nospam example.com> writes:
On Wednesday, 11 May 2022 at 06:50:37 UTC, Orfeo wrote:
 well done Andrea!

 (forum begins to be too crowded with Italians :) )


 ---
 Orfeo
We all miss the good old bearophile! I think the most active italian in this forum. Andrea
May 11
prev sibling parent reply frame <frame86 live.com> writes:
On Sunday, 8 May 2022 at 21:32:42 UTC, Andrea Fontana wrote:

 Please help me testing it, I'm looking forward to receiving 
 your shiny new issues on github.

 Dub package: https://code.dlang.org/packages/serverino

 Andrea
Take care of socket exceptions - especially if you want to make a port to Windows. You should always expect one. It's not enough to test `Socket.isAlive` - a client socket may be faulty and any illegal socket operation throws and kills your loop. Even if `isAlive` works as expected, it may changes the status before you have add the socket to the set. You don't want your server to crash if a client misbehaves.
May 14
parent reply Andrea Fontana <nospam example.com> writes:
On Saturday, 14 May 2022 at 20:44:54 UTC, frame wrote:
 Take care of socket exceptions - especially if you want to make 
 a port to Windows.

 You should always expect one. It's not enough to test 
 `Socket.isAlive` - a client socket may be faulty and any 
 illegal socket operation throws and kills your loop. Even if 
 `isAlive` works as expected, it may changes the status before 
 you have add the socket to the set. You don't want your server 
 to crash if a client misbehaves.
Which kind of socket exception could be triggered by a client? Andrea
May 14
parent reply frame <frame86 live.com> writes:
On Saturday, 14 May 2022 at 23:23:47 UTC, Andrea Fontana wrote:

 Which kind of socket exception could be triggered by a client?

 Andrea
It doesn't matter if triggered by a client or not, you need to deal with the possibility. A closed/destroyed socket is an invalid resource. I recently had the scenario on Windows where a client crashed and the socket wasn't closed properly somehow. Now the server adds the socket to the set to see an update - boom! "Socket operation on non-socket" error. Also accepting sockets can throw, for eg. by a stupid network time out error - not only on Windows. Other socket operations are no exceptions either. `isAlive` is fine for properly shutdowned/closed sockets by you or peer. But it doesn't protect you from faulting ones.
May 14
parent Andrea Fontana <nospam example.com> writes:
On Sunday, 15 May 2022 at 06:37:08 UTC, frame wrote:
 On Saturday, 14 May 2022 at 23:23:47 UTC, Andrea Fontana wrote:

 Which kind of socket exception could be triggered by a client?

 Andrea
It doesn't matter if triggered by a client or not, you need to deal with the possibility. A closed/destroyed socket is an invalid resource. I recently had the scenario on Windows where a client crashed and the socket wasn't closed properly somehow. Now the server adds the socket to the set to see an update - boom! "Socket operation on non-socket" error. Also accepting sockets can throw, for eg. by a stupid network time out error - not only on Windows. Other socket operations are no exceptions either. `isAlive` is fine for properly shutdowned/closed sockets by you or peer. But it doesn't protect you from faulting ones.
Ok, added some checks on .select, .accept, .bind, .listen. Thank you. Andrea
May 15