* Wilson Bilkovich ([EMAIL PROTECTED]) [051209 11:20]:
> Sorry to resurrect an old thread, but what about just defining a
> "no-op" query for each adapter.  In Oracle, that would be, say, 
> "select * from dual".
> Many Java connection pool implementations have such a thing, and
> execute it when a connection is assigned.  If it fails, the connection
> is closed and reopened.  If it works, it's safe to run your actual
> code.
> 
> Retries are scary.

Not only that, but with 0.14.4 we've turned up a problem on Oracle with
transaction isolation in tests.  Basically, if we have a test that calls
some AR functionality which happens to use a transaction internally,
using use_transactional_fixtures, with 0.14.3 the changes in that test
would be isolated, with 0.14.4 (on Oracle, but not on Postgres) the
changes in that test are visible outside the test.  We've gotten it down
to a fairly small test pair and it's definitely a result of changes in
oci_adapter.rb (which are almost entirely related to the new recoverable
connections).  When I have something clear enough to share I'll pass it
along.

Rick
-- 
 http://www.rickbradley.com    MUPRN: 562
                       |  Joke: do not send
   random email haiku  |  me your vlb hardware I can't
                       |  believe it either.
_______________________________________________
Rails-core mailing list
[email protected]
http://lists.rubyonrails.org/mailman/listinfo/rails-core

Reply via email to