Webware for Python

Yes. I've checked in all the changes.  I was only working on the
communications side of things.  HTTPRequest will still receive a dictionary
with input as one of the keys.  But the input is not sent from the Adapter
as part of the dictionary, it is added to the dictionary later.

I understand what you want, you want input to be a buffered file object to
read from.  We will get there, but that is not a simple, clean change.

Ideally, we want to have the sending stream that servlets write to be a
buffered object that is written to, and optionally the data can be sent as
it is written. The same technique could be used for receiving input data, as
you have asked.  This will not be terribly difficult to do, I just haven't
gotten to it yet.  

The first step is to get the communications protocol set up in a way that
will allow this type of system.  That's what I've done.


Jay


---------------------------
Vladimir Kralik wrote:

DATE: 01/29/2001 23:38:32
SUBJECT: RE: [Webware-discuss] AppServer Communications
Jay Love wrote:
> 
> This is all technical.  If you don't care about the internal workings of
> WebKit, you won't miss anything if you stop reading now.
> 
> I have implemented the new AppServer <-> Adapter communications protocol
> in each of the AppServer / Adapters.  (with the exception of mod_snake).
> 
> The new protocol goes like this.
> 
> The request from the Adapter(Webserver) to the AppServer:
> 
> A marshalled integer representing the length of the marshalled
> dictionary (next)
> A marshalled dictionary containing the records:  format, time and environ.
> An unmarshalled string containing POST data.
Do you check in all changes ? In HTTPRequest ( rev 1.17 ) you still read
INPUT as dictionary :-(



----------------------------------------------------------------------------

This e-mail and any attachments may be confidential or legally privileged.
If you received this message in error or are not the intended recipient, you
should destroy the e-mail message and any attachments or copies, and you are
prohibited from retaining, distributing, disclosing or using any information
contained herein.  Please inform us of the erroneous delivery by return
e-mail. 

Thank you for your cooperation.

----------------------------------------------------------------------------

Re: [Webware-discuss] AppServer Communications From: Terrel Shumway <tshumway@tr...> - 2001-01-30 14:57 "Love, Jay" wrote: > Ideally, we want to have the sending stream that servlets write to be a > buffered object that is written to, and optionally the data can be sent as > it is written. The same technique could be used for receiving input data, as > you have asked. This will not be terribly difficult to do, I just haven't > gotten to it yet. > How difficult would it be to expose all of the medusa tools (producers, continuations, et. al.) and make things asynchronous throughout? I would like to be able to say: "here is what I want done, I don't care how or when you do it." I realize that it is a model most people are not familiar with. Maybe we could create a class such as AsyncServlet.... servlet.write would map to channel.push. Hmm... how does one do an asynchronous SQL query? My ignorance is showing. Sendfile is an obvious application, but anything that requires computation requires a lot of work to break it up. I need to play with this some more. -- Terrel

Re: [Webware-discuss] AppServer Communications From: Geoff Talvola <gtalvola@na...> - 2001-01-30 15:37 Terrel Shumway wrote: > "Love, Jay" wrote: > > > Ideally, we want to have the sending stream that servlets write to be a > > buffered object that is written to, and optionally the data can be sent as > > it is written. The same technique could be used for receiving input data, as > > you have asked. This will not be terribly difficult to do, I just haven't > > gotten to it yet. > > > > How difficult would it be to expose all of the medusa tools (producers, > continuations, et. al.) and make things asynchronous throughout? I > would like to be able to say: "here is what I want done, I don't care > how or when you do it." I realize that it is a model most people are > not familiar with. Maybe we could create a class such as > AsyncServlet.... servlet.write would map to channel.push. > Hmm... how does one do an asynchronous SQL query? > > My ignorance is showing. Sendfile is an obvious application, but > anything that requires computation requires a lot of work to break > it up. I need to play with this some more. > > -- Terrel The asynchronous model is nice for I/O-bound operations, such as reading and writing the sockets. But I think it's an inappropriate model for the servlets because it would force you to code your servlets in small chunks of activity, which is very unnatural compared to simply writing whatever code you need and letting the OS's threads handle switching between servlets. AsyncThreadedAppServer does about the best you can do, I think -- it uses the async model for socket communications, but uses a thread pool to actually execute the servlets. The async model for socket communications should really help if the sockets are slow, although in the typical situation where you're using a web server and WebKit on the same machine, it doesn't seem like it would make much difference. (That's why I wrote the experimental HTTP-serving version of WebKit based on AsyncThreadedAppServer rather than ThreadedAppServer -- because I assumed that if we're serving HTTP directly, the sockets could be going over the internet and could therefore be very slow, so you'd rather let the async loop handle the sockets instead of tying up the threads.) -- - Geoff Talvola Parlance Corporation gtalvola@...

Re: [Webware-discuss] AppServer Communications From: Vladimir Kralik <vkralik@so...> - 2001-01-30 16:08 "Love, Jay" wrote: > > Yes. I've checked in all the changes. I was only working on the > communications side of things. HTTPRequest will still receive a dictionary > with input as one of the keys. But the input is not sent from the Adapter > as part of the dictionary, it is added to the dictionary later. Sorry, I'm wrong, I don't look in ThreadedAppServer :-(. > I understand what you want, you want input to be a buffered file object to > read from. We will get there, but that is not a simple, clean change. In AppServer add into dictionary instead string-input, opened file for input. In HTTPRequest, instead making StringIO from file, simply send this file. Know problems : * HTTPRequest._rawRequest cannot be simply copied :-(, who need it ? Why cannot copy parsed input after cgi module was called ? * in AsyncAppServer socket from non-blocking to blocking mode must be set ( file-like access with limmited buffer is needed ) > > Ideally, we want to have the sending stream that servlets write to be a > buffered object that is written to, and optionally the data can be sent as > it is written. The same technique could be used for receiving input data, as > you have asked. This will not be terribly difficult to do, I just haven't > gotten to it yet. If you want, you can see in my patch https://sourceforge.net/patch/?func=detailpatch&patch_id=102597&group_id=4866 for classes AsyncRead and AsyncWrite, this implemets file-like access with limited buffer. I use it only for AsyncAppServer, in other servers this isn't necessary. But marshall cannot directly read/write for this buffered file :-(. vlk

Re: [Webware-discuss] AppServer Communications From: Vladimir Kralik <vkralik@so...> - 2001-01-30 17:04 Terrel Shumway wrote: > My ignorance is showing. Sendfile is an obvious application, but > anything that requires computation requires a lot of work to break > it up. I need to play with this some more. In servlet you can make new thread which compute resuls, and servlet return to browser header "Refresh: 10" with text "wait please". After ten seconds browser call your servlet again and if thread done all work, return result else next page "wait please". Servlet with thread can communicate over Session. You need write some managmet for long computations threads. vlk

Re: [Webware-discuss] AppServer Communications From: ender <kthangavelu@ea...> - 2001-01-31 21:07 On Tuesday 30 January 2001 07:42, Geoff Talvola wrote: > Terrel Shumway wrote: > > "Love, Jay" wrote: > > > Ideally, we want to have the sending stream that servlets write to be a > > > buffered object that is written to, and optionally the data can be sent > > > as it is written. The same technique could be used for receiving input > > > data, as you have asked. This will not be terribly difficult to do, I > > > just haven't gotten to it yet. > > > > How difficult would it be to expose all of the medusa tools (producers, > > continuations, et. al.) and make things asynchronous throughout? I > > would like to be able to say: "here is what I want done, I don't care > > how or when you do it." I realize that it is a model most people are > > not familiar with. Maybe we could create a class such as > > AsyncServlet.... servlet.write would map to channel.push. > > Hmm... how does one do an asynchronous SQL query? > > > > My ignorance is showing. Sendfile is an obvious application, but > > anything that requires computation requires a lot of work to break > > it up. I need to play with this some more. > > > > -- Terrel > > The asynchronous model is nice for I/O-bound operations, such as reading > and writing the sockets. But I think it's an inappropriate model for the > servlets because it would force you to code your servlets in small chunks > of activity, which is very unnatural compared to simply writing whatever > code you need and letting the OS's threads handle switching between > servlets. this how zserver from zope works. ref: Zserver/medusa/select_trigger.py however there was recently a patch to medusa posted to the medusa mailing list which allows for stalled producers. the author used it to make a caching ftp server from medusa to use on a slow fs. also it should be possible to with this patch to create a proactor based system (POSA II Schmidt et al) vs the standard medusa reactor style. http://www.cs.wustl.edu/~schmidt/PDF/proactor.pdf kapil thangavelu > AsyncThreadedAppServer does about the best you can do, I think -- it uses > the async model for socket communications, but uses a thread pool to > actually execute the servlets. The async model for socket communications > should really help if the sockets are slow, although in the typical > situation where you're using a web server and WebKit on the same machine, > it doesn't seem like it would make much difference. > > (That's why I wrote the experimental HTTP-serving version of WebKit based > on AsyncThreadedAppServer rather than ThreadedAppServer -- because I > assumed that if we're serving HTTP directly, the sockets could be going > over the internet and could therefore be very slow, so you'd rather let the > async loop handle the sockets instead of tying up the threads.) > > -- > > > - Geoff Talvola > Parlance Corporation > gtalvola@... > > > _______________________________________________ > Webware-discuss mailing list > Webware-discuss@... > http://lists.sourceforge.net/lists/listinfo/webware-discuss

1 message has been excluded from this view by a project administrator.