Webware for Python

I'd certainly take no offence if a reST one is used :) That little wiki I
whacked together is admitedly a little messy and probably useless without
the auth the rest of my software provides. :) Whats the PyDiddy port like?

--
Shayne O'Neill
http://perth.indymedia.org
I know how hard it is for you to put food on your family."
 ----George W. Bush

On Tue, 6 Apr 2004, Ian Bicking wrote:

> Now that we have a server for Webware, we should think about what Wiki
> to use.  Several options have been presented -- Shayne's recent Wiki,
> the Wiki I wrote in the Sandbox, Frank's PyDiddy port (which I had
> quite forgotten about), and maybe a few others (e.g., I have yet
> another Wiki that I didn't put in the Sandbox, that I had been working
> on a couple years ago; no doubt other Webware wikis are out there).
>
> I *really* want reST input, because I think it's good for writing about
> programming, and that's what the Wiki will be about.  So writing code
> blocks and the sort should be easy, which it is in reST.  At the moment
> this means a patched version of MoinMoin, or my sandbox wiki.
>
> I'd also like to move the Cheetah Wiki over, since it's hosted with the
> Webware Wiki right now.  And there was a request for setting up a
> SQLObject Wiki, which I'd also like to put up, because by that time it
> should be easy to add yet another ;)  And they're all related projects.
>   But I think it's okay if they are all separate Wikis in separate Wiki
> namespaces.  They could each be a domain -- wiki.webwareforpython.org,
> wiki.cheetahtemplate.org, wiki.sqlobject.org.
>
> Any thoughts?  Right now I'm inclined to flesh mine out a bit more, and
> keep the pages in plain text files to make the content easy to
> manipulate if we change implementations.
>
> --
> Ian Bicking | ianb@... | http://blog.ianbicking.org
>
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by: IBM Linux Tutorials
> Free Linux tutorial presented by Daniel Robbins, President and CEO of
> GenToo technologies. Learn everything from fundamentals to system
> administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
> _______________________________________________
> Webware-discuss mailing list
> Webware-discuss@...
> https://lists.sourceforge.net/lists/listinfo/webware-discuss
>

Hallo,
Shayne ONeill hat gesagt: // Shayne ONeill wrote:

> I'd certainly take no offence if a reST one is used :) That little wiki I
> whacked together is admitedly a little messy and probably useless without
> the auth the rest of my software provides. :) Whats the PyDiddy port like?

PyDiddy is my port of PikiPiki to be a Webware-WikiWiki. (Do I sound
like a Teletubby...? ) 

You can see it in action on my site: http://footils.org/cms/wiki/ and
get the code at webware-sandbox.sf.net 

It is a very simple, but well working Wiki, that we also use in our
intranet for several time now. 

But it isn't a very well structured piece of software, the renderer is
hard to extend and it's not template-based. But it is very stable and
generally JustWorks (tm) all the time.

It even has a crude user authentication! But I wouldn't recommend it
for the Webware-site.

ciao
-- 
 Frank Barknecht                               _ ______footils.org__

For those of you not familiar with the latest MoinMoin release, version
1.2.1 has been restructured and now runs under Twisted, FastCGI, and
ModPython. 

The Twisted specific code is pretty small. I have intentions of taking
a stab at trying to get MoinMoin to run under Webware, but I don't have
the time at the moment and my skills are pretty lightweight.

The M$ approach to this would be to adapt MoinMoin to Webware and then
layer Webware specific features on top of it. For my own Webware
application, I have been trying to integrate a wiki into the
application. While it presently works well (just running MoinMoin with
the old CGI interface), it would be very handy to have access to
Webware session variables when writing MoinMoin macros to insert data
into a wiki page.

Roger Haase


__________________________________
Do you Yahoo!?
Yahoo! Small Business $15K Web Design Giveaway 
http://promotions.yahoo.com/design_giveaway/

It just struck me that whereas the php people fall over themselves to
build a million and one "yet another groupware package" thingees, the
webware folks fall over each other to make wiki's.

This tells me a lot of positive things about the psychology round here!

Nice :)

--
Shayne O'Neill
http://perth.indymedia.org
I know how hard it is for you to put food on your family."
 ----George W. Bush

On Tue, 6 Apr 2004, Roger Haase wrote:

> For those of you not familiar with the latest MoinMoin release, version
> 1.2.1 has been restructured and now runs under Twisted, FastCGI, and
> ModPython.
>
> The Twisted specific code is pretty small. I have intentions of taking
> a stab at trying to get MoinMoin to run under Webware, but I don't have
> the time at the moment and my skills are pretty lightweight.
>
> The M$ approach to this would be to adapt MoinMoin to Webware and then
> layer Webware specific features on top of it. For my own Webware
> application, I have been trying to integrate a wiki into the
> application. While it presently works well (just running MoinMoin with
> the old CGI interface), it would be very handy to have access to
> Webware session variables when writing MoinMoin macros to insert data
> into a wiki page.
>
> Roger Haase
>
>
> __________________________________
> Do you Yahoo!?
> Yahoo! Small Business $15K Web Design Giveaway
> http://promotions.yahoo.com/design_giveaway/
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by: IBM Linux Tutorials
> Free Linux tutorial presented by Daniel Robbins, President and CEO of
> GenToo technologies. Learn everything from fundamentals to system
> administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
> _______________________________________________
> Webware-discuss mailing list
> Webware-discuss@...
> https://lists.sourceforge.net/lists/listinfo/webware-discuss
>

Shayne ONeill wrote:

>It just struck me that whereas the php people fall over themselves to
>build a million and one "yet another groupware package" thingees, the
>webware folks fall over each other to make wiki's.
>  
>
Call it the "hello world" effect.  New scheme programmers MUST write a
fibonacci or factorial function.   It is a way to cut your teeth on a
low-risk project.

Hallo,
Chuck Esterbrook hat gesagt: // Chuck Esterbrook wrote:

> On Tue, 6 Apr 2004 07:49:47 +1000, Oliver Bock wrote:
> > The most difficult part of the installation was stripping WebWare
> > down to fit in 10MB with the rest of my things.  (It would be nice
> > to have a doc-free and perhaps sample-free installation option.)
> 
> You might look into building the C version of the adapter. So you
> would still end up with a CGI program, but as native code program,
> it provides a much snappier response and more throughput.

Actually I always run the C-adapter, and I never used mod_webkit.
Strange, huh?

Is mod_webkit that much faster than the C-CGI?

ciao
-- 
 Frank Barknecht                               _ ______footils.org__

Frank Barknecht wrote:
> Actually I always run the C-adapter, and I never used mod_webkit.
> Strange, huh?
> 
> Is mod_webkit that much faster than the C-CGI?

In my limited testing, the C adapter (wkcgi) was as fast (faster?) than 
the mod_python adapter (which admittedly is a pretty silly adapter). 
Though it's been a long time since I checked, so my memory is unclear. 
mod_webkit was a bit faster, but that would be easily lost in the noise 
of a real application -- either option is quite fast.  WebKit.cgi was 
noticeably slower, as one would expect.

   Ian

Hallo,
Ian Bicking hat gesagt: // Ian Bicking wrote:

> In my limited testing, the C adapter (wkcgi) was as fast (faster?) than 
> the mod_python adapter (which admittedly is a pretty silly adapter). 
> Though it's been a long time since I checked, so my memory is unclear. 
> mod_webkit was a bit faster, but that would be easily lost in the noise 
> of a real application -- either option is quite fast.  

Thanks for this info. In my applications, the adapter surely isn't the
bottleneck so I'll just stay with the C one at least until I find the
time to read the mod_webkit docs. ;)

ciao
-- 
 Frank Barknecht                               _ ______footils.org__

Ian Bicking wrote:

> Terrel Shumway wrote:
>
>>> If it doesn't, a WebKit instance per-user does not sound appealing.
>>
>> Instance-per-user is not a problem unless you want to host 50+ users on
>> one box.  But dedicated servers are cheap enough that you shouldn't need
>> to put more than 10 webware users (max) on a box anyway.
>
>
> I can't say this from direct experience, but I don't really see any 
> reason 50 users would be a problem.  

I can't say that from experience either -- its just a conservative 
estimate knowing that
 1) my boxes are small, commodity servers
 2) I expect my clients to be getting some decent traffic
 3) I would rather underestimate and overdeliver than die when
    the first publicity event hits.
 4) the enlightened elite who want to run webware have enough experience
    to recognize that "low-price" is not equivalent to "low-cost",
    so paying $20+/mo for quality hosting is not going to throw them
    for a loop.

> My only concern would be memory consumption.  I'd be curious what kind 
> of memory use different people see on their sites.  

Python does eat a lot of non-sharable memory.  I try to reduce that by 
using the
shared-library build, but the compiled bytecode is still a bigger chunk.

Very quick test:
  Webware python process has RSS=6168K or about 1.2% of a 512M server 
after responding to
  ab -n 50 http://****/wk/Examples/Colors
  using Python-2.3.3 w/ shared lib, Webware CVS on linux, mod_webkit2

This bears out the hypothesis that memory is the limiting resource. I 
would plan on 10-20
webware processes per 512MB of physical RAM, based on this 
not-so-real-world test.

More testing to come.

> Mostly, I think Webware would scale decently alongside other products. 

I don't see why it wouldn't.

I would be interested to see how PHP running as a SUEXEC CGI binary 
compares with Webware
(both OneShot and WebKit) on a moderately-busy site.

> And separate instances give built-in clustering potential ;)  

I don't see much of a win for clustering unless you are dealing with a 
cluster of
physical machines.  In that scenario, memory and CPU resource issues 
take a back
seat to the complexity of managing the cluster.

Clustering is something less than 2% of web sites need to worry about 
anyway... By the
time *I* get there, I will have enough money to pay someone else to 
worry about it. 8-)

IOW, for the simple and common domain we are dealing with (i.e. "the 
hosting factor...")
"one instance per user per box for security and isolation purposes" does 
the job[1].

I guess I am becoming an "underachiever" -- Do the simplest thing that 
could possibly work.

-- Terrel

___________
[1] and does it *far* better than the 99% of the shared hosts out there 
who use mod_php with
world writable files and directories, FTP, and 1000 low-traffic sites 
per box. (I could do
that for $1/mo/site too if I wanted to.)

Now who wants to port Squirrelmail to Python?  Anyone working on a 
webmail server?

Terrel Shumway wrote:
> I can't say that from experience either -- its just a conservative estimate knowing that
> 1) my boxes are small, commodity servers
> 2) I expect my clients to be getting some decent traffic

A lot of the stuff I've worked on is more along the lines of intranet 
applications -- so the traffic is low, but the complexity high.  Or even 
outside of an intranet that can occur -- a good example might be a CMS 
that publishes static content.  I might want each client to have their 
own separate instance of the CMS (manageable with MakeAppWorkDir, 
version control, and other tools).  There the performance requirements 
can be really low -- some instances may be entirely dormant for weeks at 
a time.

At least, that's the usage scenario I could imagine where you'd have 50 
instances.

> 3) I would rather underestimate and overdeliver than die when
>    the first publicity event hits.
> 4) the enlightened elite who want to run webware have enough experience
>    to recognize that "low-price" is not equivalent to "low-cost",
>    so paying $20+/mo for quality hosting is not going to throw them
>    for a loop. 


>> And separate instances give built-in clustering potential ;)
> 
> I don't see much of a win for clustering unless you are dealing with
> a cluster of physical machines.  In that scenario, memory and CPU
> resource issues take a back seat to the complexity of managing the
> cluster.

Hmmm... I said that wrong.  It's not clustering potential I'm thinking 
of, more clustering avoidance, which you obviously can appreciate. 
That's one advantage to having out-of-process data stores (i.e., 
client/server RDBMS), and any other services you can separate across 
process boundaries can probably be moved to other machines.

   Ian

Frank Barknecht wrote:

>Thanks for this info. In my applications, the adapter surely isn't the
>bottleneck so I'll just stay with the C one at least until I find the
>time to read the mod_webkit docs. ;)
>  
>
Behold, the most concise and marvelous mod_webkit docs:

1) edit makefile to point APXS to your binary
2) make
3) edit whatever.conf
------------whatever.conf----
    LoadModule webkit_module modules/mod_webkit.so

    <Location /wk>
    WKServer localhost 8086
    SetHandler webkit-handler
    </Location>
----------
4) restart apache

Amazing, but true -- it really is that simple.
Easier than editing WebKit.cgi, IMOMHO.

-- Terrel