[erlang-questions] Erlang http servers

Michael Turner <>
Wed Oct 3 04:25:40 CEST 2012


I recently received a copy of Building Web Applications with Erlang,
which is almost entirely about yaws. The discussion of inets on this
thread spurred me to look in the index under "i" -- except, oops, the
book has no index. It does have an appendix about other webservers and
frameworks in Erlang -- but inets is not listed there.

"Well-hidden" indeed.

-michael turner


On Wed, Oct 3, 2012 at 12:52 AM, Serge Aleynikov <> wrote:
> I agree with Joe that having a design of the HTTP server architecture
> included in OTP that would offer standardized internal API and allow
> developers to either replace selected functionality via providing
> callback modules or swap the HTTP servers easily would be a great plus.
>
> Concerning inets, what's really strange is that whenever we see some
> benchmarks posted comparing performance of different HTTP
> implementations, inets is never included there, though I would think, it
> would make more sense in terms of showing how a custom implementation
> compares to what's included in OTP.  Does it have to do something with
> what Loïc referred to as inets being well hidden in OTP?  I doubt, but
> if not that, then what?  There are not even benchmarks comparing yaws to
> inets, and despite the fact that both offer a rich set of functionality
> make one wonder was was the idea of writing yaws in the first place
> given existence of inets' httpd.
>
> Diversity is good, but at some point it creates a burden for the user
> who needs to make a selection of an HTTP server for his project.
> Instead of solving the application-related problem, now the task becomes
> studying several implementations to see which one offers a feature-set
> best suited for the problem being solved, and the answer to that
> question is not always straightforward.
>
> Perhaps a good initiative from the OTP team would be to spec out the
> open architecture of an HTTP server and welcome feedback from authors of
> current http libraries, so that the successor of inets would be flexible
> enough to accommodate building light-weight and heavy-weight HTTP
> servers based on the supported code base included in OTP?  This would
> also open a possibility of diversity for library writers
> replacing/enhancing parts to suit their needs.
>
> Serge
>
> On 10/1/2012 3:52 PM, Loïc Hoguin wrote:
>> On 10/01/2012 08:11 PM, Kenneth Lundin wrote:
>>>  > * OTP has a long release process. Many companies need their fixes to
>>> be included quickly to avoid maintaining huge branches for their
>>> changes. They generally agree in return to keep track of changes that
>>> might happen during the product's development.
>>> I think we are pretty fast with service releases about every 3rd month
>>> and updates on the maint branch in git between that.
>>
>> But doesn't that imply upgrading the whole of OTP? If it does, that
>> means more testing and more chances for new issues. If you can build a
>> release upgrade with one app from one OTP version and the rest from a
>> previous version, I'd like to know how.
>>
>>>  > Note that while not everyone uses inets, most servers do use
>>> erlang:decode_packet/3 which does most of the job. Cowboy doesn't
>>> anymore, because it needed backward incompatible changes to
>>> erlang:decode_packet/3. I didn't ask your thoughts, simply because the
>>> experiment I did hinted that doing the parsing in full Erlang was a
>>> better idea: this change results in improved HiPE optimizations and
>>> better reduction management which results in better latency under load.
>>>
>>> Interesting, would be good if you could give us more info on this. The
>>> http parsing in
>>> decode_packet is there to improve performance. If it is not ( without
>>> using HiPE) we should consider removing it!
>>
>> Between now and then binaries appeared, got performance improvements,
>> and considering Cowboy is full binary (no conversion to atom for methods
>> or header names or anything) and only receives requests, not responses,
>> that means it can skip many steps that decode_packet/3 can't.
>>
>> Binaries also mean you can parse the request line and all the headers
>> using a single match context which seems to be very fast (good job!) and
>> I'm guessing is the main reason for improved performance under HiPE (I'm
>> about to start reading the HiPE papers to understand that better).
>>
>>>  >> In Erlang/OTP we have keep strict compatibility with many of our APIs
>>>  >> which is both a strength and sometimes a burden (when we are not
>>> happy
>>>  >> with the old API) but there are mostly ways around that if there is
>>>  >> strong case for a change.
>>>  >
>>>  >
>>>  > It's a big strength for people looking for stability and a big burden
>>> for people looking for reactivity. Can't really blame the second group
>>> for finding solutions to their problems.
>>>
>>> I am not blaming people
>>
>> I used a word too strong there. :)
>>
>>> for implementing their own solution, it is good
>>> to have a lot to choose from. But I still think the stability is very
>>> important for the majority of users making systems with a longer life
>>> cycle than a year.
>>> Our experience is that we have to be very compatible in order to
>>> convince our
>>> users (often big projects) to step up to our latest release.
>>> By this we can also stop supporting older versions.
>>
>> Yes and I wouldn't want that to stop, that's a strong point. In fact
>> having both ways is great, it means you can make sure the important
>> parts of your system are stable while the more bleeding-edge parts are
>> kept, well, bleeding-edge.
>>
>>>  >> I think it is of great value to have a relative complete
>>> functionality
>>>  >> in Erlang/OTP so that
>>>  >> the user can implement basic functionality without having to search
>>>  >> for separate implementations of, with unknown quality and
>>> completeness
>>>  >> at least for a beginner.
>>>  >
>>>  >
>>>  > I would prefer Erlang/OTP to focus on the core features and leave the
>>> satellite features to third parties. OTP has an HTTP server but doesn't
>>> have proper Unicode support. OTP has OpenGL but doesn't have proper
>>> timezones support. These 2 examples missing from OTP are some of many
>>> things that are too complex for most people to implement, and that would
>>> not benefit from having alternatives.
>>>
>>> What is missing when it comes to timezones? Has it been mentioned in
>>> another thread?
>>> I think we rely on the OS configuration here.
>>
>> Date/time conversion from one timezone to another. For example if your
>> data is saved with UTC time but the user is in Paris you want to convert
>> it to Europe/Paris.
>>
>> Perhaps it exists, but I haven't found it.
>>
>> I think there's a lot of work to be done with regards to localization
>> and internationalization support, but I haven't had enough time to think
>> about it in details yet. Unicode (not just Unicode though, but a native
>> string type abstracting encodings, EEP coming in a few weeks),
>> timezones, graphical drawing libraries are pretty much what I find lacks
>> the most for the web side of things. Graphical drawing libraries are
>> fine outside of OTP for a while (until there's one that can do near
>> everything and do it well) though.
>>
>>
>>
>> Thanks for the info on package management and bug tracker. Hope we'll be
>> involved in the process of creating the former. And thanks for your nice
>> reply. :)
>>
> _______________________________________________
> erlang-questions mailing list
> 
> http://erlang.org/mailman/listinfo/erlang-questions



More information about the erlang-questions mailing list