Thread: SSI patch version 9

I wrote:

> (3) Recent changes to index page split/combine logic are still
> being tested.

And we have a new patch now that this has been finished. Two new
lines to clean up linkages which were missed, and there was one
over-zealous Assert elsewhere which was tickled by these changes.

Dan felt that other testers were likely to run into these, so here's
a new patch with these issues fixed.

Last time we did stress tests, it uncovered some race conditions.
Those were fixed at the time, and hopefully we haven't introduced any
new ones; but it's a Very Good Thing that Dan is able to run some
more DBT-2 tests, even if that test isn't ideal for highlighting
SERIALIZABLE issues.

-Kevin

I wrote:
> Last time we did stress tests, it uncovered some race conditions.
> Those were fixed at the time, and hopefully we haven't introduced
> any new ones; but it's a Very Good Thing that Dan is able to run
> some more DBT-2 tests, even if that test isn't ideal for
> highlighting SERIALIZABLE issues.
Dan's DBT-2 run triggered an Assert that looks like it would be
caused by a race condition in the new code that uses SLRU for
graceful degradation.  I will probably have another patch some time
tomorrow morning.
Sorry about this; I guess maybe I should have waited that extra few
days before posting the patch....
-Kevin