On 8/23/2013 2:55 PM, Bergquist, Brett wrote:
Mike I would love to open a Jira but having a reproducible case is important as
I realize. Right now, the reproducible database is 2Gb compressed which is
really not practical to upload.
My goal right now is to fix the problem. The problem is in production, a large
telecomm company, and the real database is about 200Gb. The database has been
hard upgraded and has been in use for about 2 weeks so rolling back to backup
before the hard upgrade is not possible. Only going forward is possible. So
I have the debugger connected up right now and am going through the code. I am
also comparing 10.8.2.2 version 10.9.1.0 to look for any suspicious
differences. I have to fix this now, not months from now.
What would be really useful would be to:
1. understand how to turn on the optimizer tracing facility that I see in the
10.9.1.0 source. For example:
optimizerTrace = lcc.getOptimizerTrace();
I can figure this out, but if anyone can point to a wiki page or some other
resource to enable the magic, sharing it will be most welcome.
I have not used the following but have been pointed at it by those who do:
http://wiki.apache.org/db-derby/OptimizerTracing
2. understand if it would be possible to access my test database that has been
hard upgraded to 10.9.1.0 with 10.8.2.2. I know that when I boot the database
it will complain and not boot but if I am just doing a read-only query on the
database with 10.8.2.2 and the query works, this will give me real confidence
that the problem is in the code changes from 10.8.2.2 to 10.9.1.1 and not
something related to the structure now stored in the database.
So I am working on this, am looking for pointers of where to look in the code,
and will open a Jira also supply a patch when I have this fixed.
this one is hard, definitely nothing is supported to to do this, and no
magic property is coming to mind. If it does not take days to load your
test db, I would just build a 10.8 db and load it.
Mostly catalog issues are what happen in hard upgrade, so doing a query
with old code on a new hard upgrade db "probably" works. But even
booting might write stuff that would then be a problem using it later,
even if your intent is only to run read-only queries.
You could build your own 10.8 and just change the code to skip the
checks that don't allow it to boot. This might work or it might corrupt
the db during the boot process. If you are trying this I would make
a copy of the test db to try i on.
