Thread: updating to 8.3.x

updating to 8.3.x

From
"Scott Marlowe"
Date:
In another thread, Ben mentioned the issues he's running into with
upgrading to 8.3 and lack of some of the old implicit (but wrong) type
conversion.

> Remember 8.3 also gives you the "opportunity" to fix all the sloppy code in
> your application that depends upon implicit casting. If that's as expensive
> for you to fix as it is for us, then you won't be moving to 8.3 any time
> soon.

What we did to fix our code was to upgrade to 8.3.3 in development
first, and fix all the bugs (about 40 or 50) we could see that popped
up in the sql logs regarding type mismatches.  then, in production, we
replaced our 8.1 slony slave with an 8.3 slony slave and set it to
handle half the read load.  We found another dozen or so bad sql
statements doing that, and fixed and released new code that night.

Then after that we swapped the 8.3 for the 8.1 server and then added
the other new server, also running 8.3 to replace the 8.1 server.

Re: updating to 8.3.x

From
Ben
Date:
On Wed, 17 Sep 2008, Scott Marlowe wrote:

> In another thread, Ben mentioned the issues he's running into with
> upgrading to 8.3 and lack of some of the old implicit (but wrong) type
> conversion.
>
>> Remember 8.3 also gives you the "opportunity" to fix all the sloppy code in
>> your application that depends upon implicit casting. If that's as expensive
>> for you to fix as it is for us, then you won't be moving to 8.3 any time
>> soon.
>
> What we did to fix our code was to upgrade to 8.3.3 in development
> first, and fix all the bugs (about 40 or 50) we could see that popped
> up in the sql logs regarding type mismatches.  then, in production, we
> replaced our 8.1 slony slave with an 8.3 slony slave and set it to
> handle half the read load.  We found another dozen or so bad sql
> statements doing that, and fixed and released new code that night.
>
> Then after that we swapped the 8.3 for the 8.1 server and then added
> the other new server, also running 8.3 to replace the 8.1 server.

....and just to be clear, our problem is not so much in finding where the
casting problems are or in deploying the fixes, but rather scope of the QA
cycle after fixing them, because they're spread quite liberally throughout
everything. 8.3 isn't enough of a win over 8.2 for our usage to justify
that effort at the moment.