On Wed, 2010-07-28 at 10:13 -0700, Blair Zajac wrote:
> On 07/28/2010 06:18 AM, julianfoad_at_apache.org wrote:
> > Author: julianfoad
> > Date: Wed Jul 28 13:18:28 2010
> > New Revision: 980046
> > URL: http://svn.apache.org/viewvc?rev=980046&view=rev
> > Log:
> > Add assertions in FSFS to trap an internal error that is believed to have
> > occurred in real life.
> Propose this one for a 1.6.x backport?
I don't think so. It doesn't provide an enhanced experience for the
user. Quite the opposite, in fact - as Bert pointed out, if the server
terminates because of this error, that can be worse than it failing a
commit and returning the error message to the client.
So these assertions are more to help us spot the location of the bug
more quickly if we ever manage to reproduce the problem in a test
environment, and to help us avoid writing any more silly bugs of this
Regarding Bert's concern about DOS attacks: if we can't find and fix the
bug that leads to this condition then I suppose we should add a
"friendly error message" instead of the assertion, and a comment
explaining why it's not just a simple assertion.
Now, I'm looking a bit further up in the function's callers to see if I
can see how the null txn id could happen.
Received on 2010-07-28 20:08:26 CEST