[Ur] Moving channel/client ID generation from the runtime into the database and/or compiler?

[austin seipp]

The Ur/Web runtime system represents asynchronous channels for message
passing between clients with an ID that is generated by the when a new
channel is allocated, and uses this for example when you insert a
'channel' into the database. Would it be possible to move this channel
generation into the database itself, perhaps with a sequence or
something similar?

My motivation is this: I have this fancy little perl script here that
uploads and runs multiple Ur/Web applications across several hosts
transparently (via ssh) and load balances them with an HTTP proxy
automatically. So it basically just allows you to shoot a
compiler-generated .exe to a few places and the main system takes care
of balancing in a very simplistic manner. It's reminiscent of a script
in the Opa distribution that does something similar (although perl >
bash.)

This all works out very well in fact! It works well when using regular
SQL tables, postgres or mysql, and ajax and all the stuff that
*should* work, does work. But because of the ID generation in the
runtime system (read: per process,) I can't make applications that use
message passing work at all, because multiple executables will step
over each others channel IDs when clients connect.

Chronologically:

1. Start two processes, on port 8081 and 8082. Both connect to the
same database.
2. Client 1 visits HTTP proxy. Redirects them to process on port 8081.
3. Port 8081 process generates a new channel for the client connection
and stuffs it in the database for further use. This is the first
client, so the runtime system increments a counter and gives this
channel an ID of 1.
4. Client 2 visits HTTP proxy. Directs them to process on port 8082.
5. Port 8082 process generates a new channel ID for the client
connection, but it *also* generates 1 and tries to insert it into the
*same* database.
6. Insert error (Channel ID already in use.) Process goes bye-bye.

Now perhaps it's just the structure of my application (a table with
one field of type 'client int' which is a PRIMARY KEY,) that makes
this so obvious, but I think the general problem remains that multiple
instances will step on each others toes one way or another with IDs
which they attempt to insert as each process will generate the same
IDs independently. I believe the same will basically apply to client
IDs as well since they're generated in a similar manner.

So is it feasible/possible to move this ID generation for
clients/channels into the DB and/or compiler, so multiple connected
instances can be coherent? I guess it's worth asking if this is even
considered a valid use case or if it's feasible at all given the
current design. Channels/client IDs are probably only really useful
when you're using a database engine to store them anyway. I just threw
the script together over the course of the day to see if the idea
would actually work. It mostly does, but this sorta makes it
non-workable at the moment for anything beyond non-trivial/not aware
of it. It'd be pretty neat if it could work 100%, though.

-- 
Regards,
Austin