Alex,
On 3/15/21 23:48, My Subs wrote:
---- On Mon, 15 Mar 2021 11:25:17 -0500 Christopher Schultz
<ch...@christopherschultz.net> wrote ----
> Alex,
>
> On 3/13/21 10:20, My Subs wrote:
> > ---- On Fri, 12 Mar 2021 20:24:48 -0500 Christopher Schultz
<ch...@christopherschultz.net> wrote ----
> >
> > >
> > > Alex,
> > >
> > > On 3/12/21 16:32, My Subs wrote:
> > > > ---- On Fri, 12 Mar 2021 02:35:27 -0500 Mark Thomas
<ma...@apache.org> wrote ----
> > > >
> > > > > On 12/03/2021 03:57, My Subs wrote:
> > > > > > Hello,
> > > > > >
> > > > > > I'm using Tomcat 10.0.0. Suppose I call setAutoCommit(false)
on a connection obtained from a Tomcat JDBC Connection Pool. Then I do some stuff with the
connection, call commit() or rollback() and finally call close() on it without ever calling
setAutocommit(true).
> > > > > >
> > > > > > What will the autocommit state of that connection be, the
next time the pool gives it to my code? Will it be in the same state I left it, that is, with
autocommit set to false, or will it be reverted back to the default state (autocommit set to
true)?
> > > > >
> > > > > The simplest way to be sure is to run a test with a pool size
of 1.
> > > > >
> > > > > A quick look at the code (I might have missed something)
suggests that
> > > > > if defaultAutoCommit is configured then it will be reset to
that default
> > > > > value. Otherwise it will be unchanged from when the connection
was
> > > > > returned to the pool.
> > > > >
> > > > > Mark
> > > > >
> > > > >
---------------------------------------------------------------------
> > > > > To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
> > > > > For additional commands, e-mail: users-h...@tomcat.apache.org
> > > >
> > > > Hi Mark,
> > > >
> > > > Thanks for your comment. I found this on
https://tomcat.apache.org/tomcat-10.0-doc/jdbc-pool.html:
> > > >
> > > > "The only state the pool itself inserts are defaultAutoCommit,
defaultReadOnly, defaultTransactionIsolation, defaultCatalog if these are set. These 4 properties are
only set upon connection creation. Should these properties be modified during the usage of the
connection, the pool itself will not reset them."
> > > >
> > > > It seems to say that connections will not be reverted back to the
default auto-commit state even if defaultAutoCommit is set. Given your reading of the
source code, might it be that the docs are wrong?
> > >
> > > I think the docs can easily be misunderstood.
> > >
> > > I don't use the tomcat-pool but I do use the *other* pool provided by
> > > Tomcat and they are generally expected to be (mostly) interchangeable.
> > >
> > > A connection pool which doesn't reset the auto-commit state on
> > > connection-return operation would IMO be considered very badly broken.
> > >
> > > tomcat-pool allows you to NOT specify a default in which case
> > > Connection.setAutoCommit() will never be called *at all* by the pool.
> > > That seems like reasonable behavior (don't mess with my stuff!), but
it
> > > means that the application will *always* have to reset the auto-commit
> > > state to a known value *every single time* a connection is borrowed.
> > >
> > > Since the whole point of the pool is to manage this kind of thing, I
> > > would argue that having defaultAutoCommit NOT set to anything would be
> > > considered very bad practice.
> > >
> > > If you set defaultAutoCommit="true", you should expect that, when your
> > > connections are returned to the pool, that setAutoCommit(true) will be
> > > called every single time the connection is returned to the pool --
> > > usually by the application calling Connection.close().
> > >
> > > Are you worried about a particular use-case or are you just being
> > > extra-vigilant? Or are you observing some unexpected behavior?
> > >
> > > -chris
> > >
> > > ---------------------------------------------------------------------
> > > To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
> > > For additional commands, e-mail: users-h...@tomcat.apache.org
> > >
> > >
> >
> > Hi Chris,
> >
> > I’m trying to wrap my head around on whether I can end transaction blocks
in a try clause with a simple call to commit(). That would be the case if the pool
resets the auto-commit state to the value of defaultAutoCommit. Try blocks would end
with commit(), catch blocks with rollback(), and the pool takes care of setting
auto-commit back to true.
> >
> > On the other hand, if as the docs say, “the pool itself will not reset”
the state of connection attributes like auto-commit, then I’d need to tack to every
try/catch a finally clause having setAutoCommit(true). But this would be
unnecessary boilerplate if, as Mark says —and you believe should be the case—, the
pool does effect the reset.
> >
> > I’d rather not bloat the code, hence my question. I guess that puts me
in the extra-vigilant camp.
>
> It should work like any other connection pool you have used in the past.
>
> Just be sure to set defaultAutoCommit="true" in your configuration. Then
> it will *always* set autocommit=true when you return your connections to
> the pool.
>
> Just some more food for thought:
>
https://blog.christopherschultz.net/2009/03/16/properly-handling-pooled-jdbc-connections/
>
> -chris
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
> For additional commands, e-mail: users-h...@tomcat.apache.org
>
>
Hi Chris,
Thank you for the advice and the great article. The example it gives
on how to prevent partial commits using multiple catch clauses that
roll back on every possible exception scenario has me thinking
whether all that is necessary when the pool attribute
rollbackOnReturn is set to true.
That post was written before rollbackOnReturn was a thing, and before
try-with-resources was a thing. Probably before multi-catch was a thing,
too. I also like giving complete code examples and sometimes brevity
must be sacrificed in favor of clarity.
Also, rollbackOnReturn will always issue a rollback() whether it's
necessary or not. Most drivers will implement that as a real ROLLBACK
sent to the server which could represent a (small) reduction in performance.
I understand that in such case any exception thrown from within a
TWR's try block will make TWR to call close() on the connection
declared on the TWR header. And given that rollbackOnReturn is set
to true, this will make the pool to call rollback() on the connection
before returning it to the pool. By the time control reaches any
catch clause, the transaction has already been rolled back.
I'd have to read the code, but I'd assume that return-object-handling
looks something like this:
if(autoRollbackOnReturn) {
conn.rollback(); // with try/catch obviously
}
if(null != defaultAutoCommit) {
conn.setAutoCommit(defaultAutoCommit);
}
If that's correct, then you can probably get away with
try-with-resources. You will end up calling rollback after every
transaction, though, which might be a (bit of a) waste.
Am I missing something and rolling back on the catch clauses is
nevertheless necessary?
I'm not sure it is necessary, unless I've missed something.
As I said before, I tend to favor clarity in my code for improved
maintainability, and in the case of try-with-resources plus an
elsewhere-configured-and-documented rollback behavior, I would
personally stay away from it. One change to that configuration and
literally all of your code breaks.
-chris
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org
