On Mon, Mar 15, 2010 at 9:07 AM, Dmitry Belayev
<rumata-estor@nm.ru> wrote:
Compile-type checking can be very helpful and it is Haskell's strong side.
But what about hot code swapping? In working web application it is often convenient to update some part of application without disturbing connected users. It's strong side of Erlang, but what about Haskell?
That's definitely a good point. My current deployment uses lighttpd/CGI to server the application, along with a development system running on 3000. Here's my workflow:
1) Develop on local system using simple-server.
2) Push changes to development system. Who cares about a little downtime there.
3) When everything works, pull changes into the production repo.
4) Compile everything with -Wall -Werror.
5) Copy templates, static files and CGI binary in one fell swoop to the production directory.
So this is really a solution outside of Haskell. I believe Turbinado used some features for dynamically compiling code, but I prefer this kind of approach in general.
If I was during a standalone server approach, one possibility for production deployment might be:
1) Run the production server on port 3001 (or anything else). Set up iptables to forward port 80 to port 3001. (I do this already).
2) When I want to make a production change, start a *new* production server on port 3002.
3) Adjust iptables to port forward to 3002.
4) Next time around, switch back to 3001.
For the record, I think iptables forwarding is probably the safest way to run an application on a privileged port.
Michael Snoyman wrote:
And part 2:
Yesod is still missing a model layer. This is a project that I will most likely be doing myself, since it will need to be designed to integrate very closely with Yesod, but I'll mention some of the design ideas here. (By the way, Chris Eidhof has a package on Github called Basil which looks very interesting here.)
1) I want to use YAML quasi-quoting to define the models.
2) I don't want a super-complex "object relational mapping"; Haskell is already much closer to the relational level than an OO language.
3) Utilize IO thunks (ie, lazy IO loading) so that we don't need to load everything into memory at the beginning.
4) Not tied down to SQL *or* in-memory databases. It should support both of those, but *also* support a system that is usable even in a CGI setup. I'm contemplating now a database that uses multiple directories/files and uses append-only to avoid concurrency issues.
Compile-time checking of templates would be great. The Bravo package looks interesting, but I'm a little dubious about the approach used. I think something based on typeclasses might be a bit more nimble. Once again, tight integration with the model system I theoretically design would be nice.
So, in summary of a much-too-long e-mail, let's not bother trying to get everyone to agree on One Framework. Happs has its place, Yesod has its place, and I know others are using Loli and probably Kibro. (I'm not sure if anyone every got started with Turbinado.) However, we can raise the bar for *all* packages simultaneously by making a more solid ecosystem of Haskell web packages available that any framework can cherry-pick from.
_______________________________________________
web-devel mailing list
web-devel@haskell.org
http://www.haskell.org/mailman/listinfo/web-devel