Webware for Python

I haven't quite figured it, but there's something weird with
Mozilla/Galeon and reloading pages created from POST.  It just doesn't
seem to do it.  I usually go back and resubmit.

There's still something about Netscape 4.x, despite all it's lameness,
that keeps me coming back from time to time.  It's just so much more
responsive than all the modern browsers, even IE (though Opera could
probably win me over if I gave it a chance).

On Thu, 2001-11-29 at 15:55, Chuck Esterbrook wrote:
> On Windows, I develop WebKit sites with IE & OneShot.cgi and simply hit 
> reload after I make changes. Everything works fine.
> 
> However, on Linux, both Mozilla and Galeon always give back the same page 
> when I reload even though they seem to churn through the CGI script.
> 
> Konqueror gives me back a new page, but loses the last set of POSTed 
> variables thereby causing other problems.
> 
> In all cases, completely quitting the brower and restarting works, but 
> that's obviously painful.
> 
> I tried adding this to SitePage, but it didn't help:
> 
> 	def awake(self, trans):
> 		Page.awake(self, trans)
> 		setHeader = self.response().setHeader
> 		setHeader('Cache-Control', 'no-cache')
> 		setHeader('Expires', '0')
> 		setHeader('Pragma', 'no-cache')
> 
> What browser do you guys use for developing WebKit sites on Linux? Do you 
> have this reload problem? Do you use OneShot?
> 
> 
> -Chuck
> 
> _______________________________________________
> Webware-discuss mailing list
> Webware-discuss@...
> https://lists.sourceforge.net/lists/listinfo/webware-discuss
>

> However, on Linux, both Mozilla and Galeon always give back the same page 
> when I reload even though they seem to churn through the CGI script.

I suppose you already tried forcing mozilla to not look at it's page
cache?  You can tell it to always fetch a fresh copy everytime.

I haven't really had this problem, and I use mozilla most of the time. 

jack.

Chuck and Geoff,
you guys seem to have heavy workloads these days, so thanks for 
the work you're putting in getting 0.6 released!  
Tavis

We actually have a hardware load balancer that is collecting dust
because I had already written oodles of cold fusion using sessions and
getting primary keys from MS SQL Server by "select max(primaryKey) + 1".


The load balancer doesn't have the option to use cookies to keep users
on the same server.  That option cost $4,000 :).  Cold Fusion does
support database storage of session variables, called client variables.
That would solve the session problem. =20

The other problem is that SQL Server doesn't have Sequences.  Hal Helms,
a CF guru, writes that to overcome this, just use GetUUID() to create
primary keys that are almost guaranteed to be unique across the entire
world.  The downside is that those primary keys are 35 bytes long.
Yuck.  I'm curious how others simulate sequences on MS SQL Server and
similar databases.  Server side locking of tables during inserts?

I'm still reading that etoys clustering article.  Good stuff!

-Jeff

At 10:23 AM 11/29/01 -0500, Jeff Johnson wrote:
>The other problem is that SQL Server doesn't have Sequences.  Hal Helms,
>a CF guru, writes that to overcome this, just use GetUUID() to create
>primary keys that are almost guaranteed to be unique across the entire
>world.  The downside is that those primary keys are 35 bytes long.
>Yuck.  I'm curious how others simulate sequences on MS SQL Server and
>similar databases.  Server side locking of tables during inserts?

I'm not sure what Sequences are, but SQL Server most definitely does 
support unique primary keys -- just make sure Identity is turned on for the 
primary key column.  Then you can insert a new row and retrieve its 
auto-generated unique primary key with something like this:

SET NOCOUNT ON
INSERT INTO MyTable (MyValue1, MyValue2) VALUES ('foo', 'bar')
SELECT @@IDENTITY AS NewPrimaryKey
SET NOCOUNT OFF

In other words, @@IDENTITY is the last Identity value that was 
auto-generated on the given connection.

The business with "SET NOCOUNT ON" and "SET NOCOUNT OFF" is needed due to a 
bug that I read about in the Microsoft Knowledge Base.


--

- Geoff Talvola
   gtalvola@...

At 10:23 AM 11/29/01 -0500, Jeff Johnson wrote:
>We actually have a hardware load balancer that is collecting dust
>because I had already written oodles of cold fusion using sessions and
>getting primary keys from MS SQL Server by "select max(primaryKey) + 1".
>
>The load balancer doesn't have the option to use cookies to keep users
>on the same server.  That option cost $4,000 :).  Cold Fusion does
>support database storage of session variables, called client variables.
>That would solve the session problem.

What do you think of the option we sort of outlined yesterday:

- Software or hardware load balancer distributes incoming requests randomly 
to a pool of machines running Apache (call them APACHE1, APACHE2, etc).
- APACHE1, APACHE2, etc have a modified version of the WebKit adapter 
running on them.  The modification is: if no session ID is found in the 
request, it sends the request to a random machine, otherwise it parses the 
session ID to find which particular machine to send the request to, from a 
pool of available WebKit machines (WEBKIT1, WEBKIT2, etc).  The pool of 
machines would be configurable in the adapter's config file.
- WEBKIT1, WEBKIT2, etc have a modified version of Session.py that encodes 
the machine name into the session key. This would be configured in 
Application.config.  The session key would be something like 
"WEBKIT1_237480927".

The idea being to keep requests for the same session on the same 
machine.  Would this work for you?  I'm interested in helping to write the 
WebKit modifications needed.

The other option is, as you said, to write a SessionSQLStore.py to store 
sessions in a database.  You have to be a little bit careful about locking 
here, since multiple requests for the same session may be coming from 
different appservers at the same time.  I'm not sure how to avoid wiping 
out one appserver's changes with another appserver's changes in that case.


--

- Geoff Talvola
   gtalvola@...

> What do you think of the option we sort of outlined yesterday:
>=20
> - Software or hardware load balancer distributes incoming=20
> requests randomly=20
> to a pool of machines running Apache (call them APACHE1,=20
> APACHE2, etc).
> - APACHE1, APACHE2, etc have a modified version of the WebKit adapter=20
> running on them.  The modification is: if no session ID is=20
> found in the=20
> request, it sends the request to a random machine, otherwise=20
> it parses the=20
> session ID to find which particular machine to send the=20
> request to, from a=20
> pool of available WebKit machines (WEBKIT1, WEBKIT2, etc). =20
> The pool of=20
> machines would be configurable in the adapter's config file.
> - WEBKIT1, WEBKIT2, etc have a modified version of Session.py=20
> that encodes=20
> the machine name into the session key. This would be configured in=20
> Application.config.  The session key would be something like=20
> "WEBKIT1_237480927".
>=20
> The idea being to keep requests for the same session on the same=20
> machine.  Would this work for you?  I'm interested in helping=20
> to write the=20
> WebKit modifications needed.

I'm a fan of cookies (or too lazy to do it the other way) but some
people have said they can't count on them.  Would this support those
folks too?

One of the suggestions for multi-domains bothers me, not only would
bookmarks be server specific but SSL certs would have to exist for each
domain.

>=20
> The other option is, as you said, to write a=20
> SessionSQLStore.py to store=20
> sessions in a database.  You have to be a little bit careful=20
> about locking=20
> here, since multiple requests for the same session may be coming from=20
> different appservers at the same time.  I'm not sure how to=20
> avoid wiping=20
> out one appserver's changes with another appserver's changes=20
> in that case.

I started thinking about those issues and no longer think it's trivial
:).  All reads and writes would probably have to go through a method
that locked the SQL table.  Luckily I've never had to lock tables.  I
try to avoid db vendor specific code but I think you'd have to do it for
this.

At 12:07 PM 11/29/01 -0500, Jeff Johnson wrote:
> > What do you think of the option we sort of outlined yesterday:
> >
> > - Software or hardware load balancer distributes incoming
> > requests randomly
> > to a pool of machines running Apache (call them APACHE1,
> > APACHE2, etc).
> > - APACHE1, APACHE2, etc have a modified version of the WebKit adapter
> > running on them.  The modification is: if no session ID is
> > found in the
> > request, it sends the request to a random machine, otherwise
> > it parses the
> > session ID to find which particular machine to send the
> > request to, from a
> > pool of available WebKit machines (WEBKIT1, WEBKIT2, etc).
> > The pool of
> > machines would be configurable in the adapter's config file.
> > - WEBKIT1, WEBKIT2, etc have a modified version of Session.py
> > that encodes
> > the machine name into the session key. This would be configured in
> > Application.config.  The session key would be something like
> > "WEBKIT1_237480927".
> >
> > The idea being to keep requests for the same session on the same
> > machine.  Would this work for you?  I'm interested in helping
> > to write the
> > WebKit modifications needed.
>
>I'm a fan of cookies (or too lazy to do it the other way) but some
>people have said they can't count on them.  Would this support those
>folks too?

It could be made to support them -- you just have to make sure the the 
logic in the adapter knows how to find the session ID, whatever method is 
being used to encode it.

>One of the suggestions for multi-domains bothers me, not only would
>bookmarks be server specific but SSL certs would have to exist for each
>domain.

But wouldn't the hardware load balancer make it look to the user like 
there's only a single box, even though it's actually distributing the 
requests to multiple back ends machines running Apache?  I don't know about 
the multiple SSL certs; could you just use a single cert and place it onto 
all of your boxes?

> >
> > The other option is, as you said, to write a
> > SessionSQLStore.py to store
> > sessions in a database.  You have to be a little bit careful
> > about locking
> > here, since multiple requests for the same session may be coming from
> > different appservers at the same time.  I'm not sure how to
> > avoid wiping
> > out one appserver's changes with another appserver's changes
> > in that case.
>
>I started thinking about those issues and no longer think it's trivial
>:).  All reads and writes would probably have to go through a method
>that locked the SQL table.  Luckily I've never had to lock tables.  I
>try to avoid db vendor specific code but I think you'd have to do it for
>this.

--

- Geoff Talvola
   gtalvola@...

On Wed, Nov 28, 2001 at 08:42:52PM -0600, Darryl wrote:
> Just throwing idear's around to see what came up

Gotta watch out for those flying idears.  They can hit like a
boomerang.

-- 
-Mike (Iron) Orr, iron@...  (if mail problems: mso@...)
   http://iron.cx/     English * Esperanto * Russkiy * Deutsch * Espan~ol