Hi Brian

Why don't you just avoid using sybace_pconnect() with transactions?

I use ODBTP to connect to SQL Server and Sybase databases, and this is not
an issue.  The reason is because ODBTP involves the use of a mediating
service that pools all connections, and automatically rollbacks
transactions upon client disconnection. The odbtp extension does not
support persistent connections, since they are pooled by the ODBTP server.
And, if an odbtp script aborts, the odbtp client connection will be
terminated, which will cause the odbtp server to rollback the
transaction. However, the underlyning database connection will still be
pooled by the server for use by another client.

Persistent connections may not be the best way to implement database
connection reusage. Not only are they subject to undesirable open
transactions, they will never timeout, and a single web client can cause a
database connection to remain open for each web server child process.
Connection pooling may be a better alternative, however, it requires a
middle-man server to manage the pool.

Unfortuantely, ODBTP requires the server to be placed on a Win32 host that
can access the database with its local ODBC facilities. For some
developers, this is not always feasible. If it's feasible for you, then
your problem will be solved by using the odbtp extension. Otherwise, you
can't use persistent connections.

-- bob

On Wed, 15 Sep 2004, Brian Foddy wrote:

> On transactions, no this is my biggest concern.  Say a php script
> performs a
> "begin tran" then aborts early due to an error (user error for
> instance), but
> the script error handling fails to rollback the transaction before it exits.
> Just 1 poorly coded script.
> Now the begin tran is still open, the next  page served by that apache
> process inherits an open transaction.  Even if it is coded and perform
> perfectly with its own begin / commit trans (remember you can nest trans),
> the transaction is ultimately never commited; but there is almost no
> indication to the script.  This keeps going and going making this
> uncomitted trans bigger and bigger, locking more and more records.
> Finally one of 2 things will happen.
> 1.  The locks get so big that users may get blocked or the max locks exceed
> the sybase config.
> 2.  The apache process may actually exit if its defined for a max usage or
> is trimmed by the web server due to max running daemons.
> In case 1, application support would probably restart the web server to
> release
> the locks.  In case 2 apache itself does it.
> Either way, the transaction and all the subsequent nested ones will be
> rolled back, as if they never occurred.  It could be a big database
> headache.
> The best way I can forsee to prevent this is to always perform a
> "rollback tran" after a call to sybase_pconnect to make sure there isn't
> an unclosed trans.
> As for the other points you make, I was expecting to perform a
> sybase_pconnect each execution anyway so that is automatic.
> I'm not sure how I can reset the handle in Sybase?  If its possible I'd like
> to know how.
> The more I think about this, it seems that this whole thing is more work
> than its worth.  But if someone sees some effective ways to handle some
> of these issues I'd like to hear them.
> Brian
> Jeff Moss wrote:
> > The biggest problem I've had with persistent connections is the
> > problems that arise when the connection goes down. You have to monitor
> > the connection status anyways (and reconnect on a failure), so it was
> > usually easier to just connect every time. I don't know if this is
> > specific to sybase. You also avoid headache dealing with multiple
> > connections per process. Over a local ethernet this was usually such a
> > short delay that it didn't matter. Typically I don't care much for
> > speed, you avoid a lot of headache avoiding the persistent
> > connections, but the tradeoff is speed of course.
> >
> > It seems to make a lot more sense to me to just reset the handle to
> > drop all temp tables and that.
> >
> > As for the transactions, I think as long as you "do" the transaction
> > all at once there would be no problem right? If it was a problem in
> > the middle of a socket write, chances are the socket closed also, right?
> >
> > -Jeff
> >
> > Brian Foddy wrote:
> >
> >> I've been using PHP4/5 and Sybase for several years, using standard
> >> sybase_connect.  Today I tried playing around with pconnect to get
> >> aquainted.
> >>
> >> I expected one simple condition of a "use database" from one web page
> >> affecting another, and easilly handled that with a connection wrapper
> >> that
> >> re-uses the proper database with each reconnection.
> >>
> >> A couple other more troublesome issues also quickly came up.
> >> 1.  Any #temp database tables are not destroyed between calls.  I can
> >> probably
> >> work around this with some minor coding changes to manually drop temp
> >> tables.
> >>
> >> 2.  Any call to environmental "set" commands like "set isolation"
> >> remain in effect
> >> after the web page is complete.  Again with some work I could
> >> probably recode some
> >> pages to not change these values, or reset them when complete.
> >>
> >> 3.  The potentially most bothersome would be a page failing to call
> >> commit tran/rollback tran, especially during some error condition.
> >> Its easy to think
> >> any uncommited tran is rolled back when the page exists.  But
> >> persistent connections
> >> won't do this.  This could be a disaster by leaving open a
> >> transaction causing all
> >> subsequent calls to never be commited...  I might be able to create
> >> some wrapper
> >> that always resets the transaction state before starting??
> >>
> >> I'm curious how others have attempted to solve these problems, and
> >> others
> >> I haven't thought of yet.  Clearly many can be avoided by having
> >> clean code,
> >> but just 1 exception...
> >>
> >> Brian
> >>
> >
> --
> PHP Database Mailing List (http://www.php.net/)
> To unsubscribe, visit: http://www.php.net/unsub.php

PHP Database Mailing List (http://www.php.net/)
To unsubscribe, visit: http://www.php.net/unsub.php

Reply via email to