Re: [DotGNU]Execution of web services by the VRS

[Charles Shuller]

I'm still a newbie, but what about a "plugin" type of approach.  Where 
you define specific interface functions to start and stop the 
web-service, ws_start, ws_stop, ws_unload (ws = web service). Forking a 
new process to actually load, unload, and otherwise manage the plugin 
might be a good idea, so you can easily kill it if something goes wrong, 
as opposed to loading the plugin directly into the VRS process.  Is it 
difficult to determine if a .dll/.so is IL or not?  If not, determine 
the executable type on load(maby before forking the process to load it) 
and place it in the proper environment.  I know at one point I 
determined that it would be fairly straight forward to author portable 
(u*ix and win32, don't remember if I checked for VMS or not, and OS/390 
has OE (which I understand is actually a Linux Distro now) code to do 
this, and I suspect this is still the case.

--Charles

Eric Altendorf wrote:
> I need some help here, folks.
>
> So, I'm working on the VRS service manager (the module that 
> orchestrates the actual fulfillment of external web service 
> requests).  I'm at a point where I need to figure out just how the 
> service manager actually invokes a web service.  This of course ties 
> into the whole execution environment question.
>
> The service manager is a C# program.  Web services (aka executable 
> data objects / EDO's) should be able to be written in "any language". 
> These services should be able to be loaded, unloaded, and reloaded in 
> memory by the service manager.  I see two possibilities:
>
> * A service written in C# or another IL-compatible language should be 
> able to be loaded via reflection into the same process.  Nice for 
> efficiency.
>
> * A service written in a native binary format would have to execute in 
> a separate process.  Hopefully this separate process would be written 
> to load and not exit until asked to unload or reload...and somehow 
> the results would need to be efficiently communicated back to the VRS 
> process.  I'm not sure how this would work.
>
> Any ideas?  I'm willing to postpone the second and concentrate on the 
> first, which should be as easy as defining a "VRSService" or "VRSEDO" 
> or "VRSWebService" or some such interface, with methods like 
> "serve()", which web services would have to implement.
>
> Please share any thoughts...
>
> eric
>
> PS: Cross-posting to DotGNU dev maillist for greater visibility...