my Java developers are telling me, the recent JDBC drivers are throwing
an exception if they inadvertantly issue a commit() on an autocommit
connection. this is causing a bunch of problems for us, we have a
huge code base of java jdbc software which was originally written for
Oracle, and it assumes that read only operations are NOT starting a
transaction block (apparently Oracle only starts a transaction block on
DML like insert/update/delete/select for update/etc.). We have long
running threads which just do SELECTs and others that do transactions,
some which can do either of these depending on external events. We've
tried to make the SELECT only connections 'autocommit' so they don't
start transaction blocks, and we've tried to enforce .Commit() on other
situations, but the code is complex enough that sometimes the wires get
crossed, and a Commit() s done on an autocommit connection. All of
this is to prevent gigenormous data bloating from week-long <IDLE> in
transactions.
in the JDBC releases for PG 8.x, this was quietly ignored. On the new
9.x JDBC releases, this crashes with an ugly exception.
My very-oracle-centric lead SQL developer is trying to use this as yet
another excuse not to use Postgres as in his opinion, Oracle just 'does
what you want'. ok, it does what HE wants, meh.
Meanwhile, these random crashes that only show up under full production
volume workloads (at our deployment sites in Asia) are freaking out the
operations people, who ALSO are becoming afraid of Postgres.
--
john r pierce N 37, W 122
santa cruz ca mid-left coast