Patch for Bug 608032 - MySQL timeout and no attempt reconnect

[Geert Janssens]

On Thursday 4 February 2010, Phil Longstaff wrote:
> On Thu, 2010-02-04 at 22:26 +0100, Geert Janssens wrote:
> > On Thursday 4 February 2010, Tom Van Braeckel wrote:
> > > > On Thursday 4 February 2010, Tom Van Braeckel wrote:
> > > > > Hi guys,
> > > > >
> > > > > Here's another patch - my first *code* patch to GnuCash !
> > > > >
> > > > > Rationale: When we try to open a database transaction, and the
> > > > > database reports that the "server has gone away", we try to
> > > > > reconnect before
> > > >
> > > > failing
> > > >
> > > > > hard.
> > > >
> > > > Hi,
> > > >
> > > > Thanks you for your patch. It looks like a good start, but to my
> > > > limited knowledge of the sql backend, it seems incomplete.
> > > >
> > > > Here are my thoughts on the patch:
> > > > * you test for an error by comparing with a string. I think it would
> > > > probably
> > > > be safer and definitely be more efficient to simply test for the
> > > > error number
> > > > returned by dbi_conn_error. It's possible that the error strings
> > > > returned by
> > > > MySQL are translated into other languages on other systems, so your
> > > > string wouldn't always match.
> > >
> > > Thanks for the feedback !
> > > You're right - I've corrected this in the attached patch and cleaned it
> > > up a bit - hope you like it :-)
> > >
> > > Note that such string-based error checking is also done in other places
> > > (such as the "mysql_error_fn" function in gnc-backend-dbi.c), perhaps
> > > I'll rectify that in a future patch.
> >
> > Hmm, after giving it some more thought, I think I know why Phil used the
> > string in the first place: MySQL is not the only database supported by
> > dbi. Error code 2006 in MySQL may mean something different in SQLite or
> > Postgresql.
> >
> > So using the number only, we risk that a Postgres user gets unexpected
> > behavior if he hits a 2006 error.
> >
> > I find your reference to mysql_error_fn very interesting. This is the
> > function that is called whenever something goes wrong when accessing the
> > database, and it is specific to mysql. I wonder if you can't reuse this
> > function instead to handle the reconnection.
> 
> This is why I would have a generic "you need to reconnect" flag which is
> set by mysql_error_fn if error 2006 happens, is set by pgsql_error_fn
> when ... happens, etc.
> 
> The error function could do the reconnect, but doesn't have enough
> context to redo the operation which encountered the error.
> 
That's exactly what I was thinking as well. Thanks for clarifying this.

> Phil
> 

Geert