Re: 2 connections 1 transaction

Hi Daniel,

> 3.App-Server reads the database and makes changes. Problem : the changes
> the client does are not commited -> the server can't see the changes or
> the case more bad the server waits for the client connection.
> (transaction isolation and table / record locking)

The app server CAN see the changes if you set the transaction level to
serializable.  If using multiple servers is a requirement (and Dima's
comments are right - why not just use the one connection to pg and pass
everything through the app server?) just rethink your post-pg processing so
that the app server doesn't update rows modified by pg.  If you process all
the changed rows in a different table then everything should work.  If not
roll back pg and the app server.

Or get rid of the app server.  Tricky one..

Cheers,

Jason


On Fri, 18 Jul 2003 12:47 am, Daniel Schuchardt wrote:
> Hi Jason,
>
> Thats not what I mean. Both, the Client and the App-Server connects to
> the same Postgres-Database. The problem a procedure like this:
>
> 1.Client starts Transaction and does some changes.
> 2.Now the client notices that very huge operations are nescesarry and
> starts a procedure @ App-Server
> 3.App-Server reads the database and makes changes. Problem : the changes
> the client does are not commited -> the server can't see the changes or
> the case more bad the server waits for the client connection.
> (transaction isolation and table / record locking)
> 4.If everything works well both changes (done by the client and the
> server) should commit or rollback now
>
> So both Postgres-Connections has to be @ the same TransAction-OID in
> Postgres.
>
> greets
>
> Daniel
>
> -----Ursprüngliche Nachricht-----
> Von: pgsql-general-owner+M45575@postgresql.org
> [mailto:pgsql-general-owner+M45575@postgresql.org] Im Auftrag von Jason
> Godden
> Gesendet: Donnerstag, 17. Juli 2003 14:23
> An: Daniel Schuchardt; pgsql-general@postgresql.org
> Betreff: Re: [GENERAL] 2 connections 1 transaction
>
>
> Hi Daniel,
>
> Maybe make procedural wrappers around all events undertaken and simulate
> it?
>
> ie:
>
> 1. Client connects to MS SQL Server (Application Server) and PG Server
> 2. SQL Server connects to PG Server aswell 3. Client begins new record
> process 4. New record process starts by putting PG into serializable
> transactions (so
> App Server can respect the changes).
> 5. If one server transaction fails roll back the other and vice versa
> and
> start again.
> 6. If all good, commit both at the client as the last item in the new
> record
> process.
>
> Meanwhile the Application Server may make it's own calls against Pg
> through
> strored procedures.  If any part of this PG transaction fails, error
> comes
> back to App Server and App Server transaction roll backs, which tells
> the
> client and the client rolls back it's own call to PG.
>
> I don't know if this is a viable method (v. messy) and given the
> concurrent
> update issue this may play havoc with the client logic if both the
> client and
> the app server attempt to modify the same data at the PG Server.  Can
> all
> connections simply go through the App Server leaving it to manage the PG
>
> stuff in it's own transactions (still, concurrent update problem can
> occur if
> dealing with same records)?
>
> Rgds,
>
> Jason
>
> On Thu, 17 Jul 2003 09:45 pm, Daniel Schuchardt wrote:
> > Hi @ all,
> >
> > Our software consists of a Client-Side App and a Application Server.
> > Every client owns a direct connection to the PSql-Server and for every
> >
> > Client the Application-Server also creates a connection to the
> > PSql-Server. The problem is that it is nescesary that the Client and
> > the Application-Server are in the same transaction. But how say
> > connection x to be in the same transaction like connection y?
> >
> > Thanks for help,
> >
> > Daniel
>
> ---------------------------(end of broadcast)---------------------------
> TIP 4: Don't 'kill -9' the postmaster
>
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 4: Don't 'kill -9' the postmaster