This technique works, but involves a lot of coding and testing.
I have used this approach in a number of different applications, both in
WebSphere, and pure RPG batch based applications.
The downfalls I found were two fold. The first was the time spent getting a
production quality application working reliably, and the second was large
transaction throughput causing the data queues to fill up because the server
programs could not process the transactions fast enough in busy situations.
In short, the server programs become a bottleneck.
If you are expecting a large number of users you are likely to need many
server jobs waiting for responses from many data queues. You will need to
build into the design some means of knowing which data queue your session is
using. I recommend creating your own subsystem to run the desired number of
server jobs.
To avoid this problem in my Fortress/400 product, I use a cache of data held
in user spaces. Access to user spaces using pointers is very fast and avoids
file open/close transactions. I keep the data in the cache current using
database trigger programs. This approach avoids data queues and therefore
has no bottlenecks.
This sounds like a very useful piece of generalised infrastructure to build,
but it could take a number of weeks to develop.
Personally, I prefer products where the infrastructure already exists (I use
IceBreak) so I don't need to re-invent wheels. Perhaps it would be worth a
web search to see if any product already exists to do this job for you.
Syd Nicholson
Castlehill Computer Services Ltd.
-----Original Message-----
From: web400-bounces@xxxxxxxxxxxx [mailto:web400-bounces@xxxxxxxxxxxx] On
Behalf Of Aaron Bartell
Sent: 06 January 2009 15:31
To: Web Enabling the AS400 / iSeries
Subject: Re: [WEB400] Persistent CGIDEV2 problem
As far as I know there is no way to associate a server job with a
specific client.
But you could do this with some coding. Create a main "doorway" app
that reads in the necessary stdin and puts it into a "accessible place"
(maybe a User Space). Then evaluate a session id sent up from the
client (i.e. hidden form variable) to determine what key to use on the
subsequent keyed data queue send/write. Then you would have a variety
of single user jobs listening on the keyed data queue for requests they
are listening for (i.e. they are listening for their specific key to
enter the data queue). These single user programs receiving entries
from the keyed data queue can operate similar to a traditional
5250-green-screen-programming-apprach, you just change out the EXFMT
with a rcvdtaq API call. You would also retrieve your input from a
different location and send your output to a different location vs.
occupying DSPF fields (though you could simulate writing DSPF records
with SPECIAL files). Note a lot of this is theory as I haven't done it
entirely in the RPG space but instead worked with Joe Pluta to have the
front end be Java (more specifically JSP).
An approach like the above would also make it easier to retain SQL
cursor positions and I find I use a lot more embedded SQL when I am
doing RPG+CGI.
The best part about it is the entire framework behind the data queue is
"safe" from the chosen front end which could be a variety of
technologies like .NET thick client, HTML+CSS+Javascript, Silverlight,
JavaFX, Flex.
I am just itching to try something like this and even get a rough
proof-of-concept on the net for people to play with - I just keep
getting tied up in billable work and that seems to take precedence
(butts need diapers - no, not my butt ;-)
Aaron Bartell
http://mowyourlawn.com
Brandon Peterson wrote:
Hi Rob,
I doubt this has anything to do with activation groups, but even if it
does you have a larger problem.
CGI is different than 5250. When you are logged in via 5250 you have a
single job that services your session. But with CGI the web server has
a pool of jobs that service users and every time you send a request you
could potentially use a different job.
Basically you could think of it as every time you send a new request
your programs start over from scratch. As far as I know there is no way
to associate a server job with a specific client.
HTH,
Brandon
-----Original Message-----
From: web400-bounces@xxxxxxxxxxxx [mailto:web400-bounces@xxxxxxxxxxxx]
On Behalf Of Rob Dixon
Sent: Tuesday, January 06, 2009 7:44 AM
To: web400@xxxxxxxxxxxx
Subject: [WEB400] Persistent CGIDEV2 problem
I am using CGIDEV2 (in persistent mode) for the first time and thought
that
I was making reasonable progress but have now run against a problem for
which I need help, although I am sure that it has been solved many times
before.
My cgi program is RPGLE and is a request processing program (I will call
it
RPP) that handles 5250 or HTML/Javascript output. It does not have any
direct database access but calls a second program (I will call it MAIN).
This handles business logic and reads records and passes them back to
the
RPP which displays them using 5250 or CGI. In CGI mode, the first call
by
RPP of MAIN works OK and the records are displayed by RPP using
wrtsection.
When I type something on my HTML display, the RPP is reactivated
correctly
so my Handle is presumably OK and I can read what I typed using
ZhbGetInput. However, when RPP calls MAIN again so that that program
can
retrieve the required records, a new instance of MAIN is called. I am
sure
that in the first call of MAIN I did not set on LR and the mechanism
works
fine in 5250 mode I imagine that this has something to do with
activation
groups.
Can anyone help please? I have tried compiling MAIN with the same named
activation group and binding directory as RPP but this did not help.
Many thanks
midrange.com
