Simon Riggs <simon@2ndQuadrant.com> writes:
> On Tue, 2009-09-15 at 09:41 +0300, Heikki Linnakangas wrote:
>> This means that the WAL replay of that record type has never been tested
>> correctly :-(.
> This must have been added after mid-Feb this year. I notice there are a
> few places where functionality is tested against temp tables, which may
> mask other non-recoverable issues in this and other rmgrs. We should
> make it standard practice to include only non-temp tables to cover
> functionality other than specific temp table commands.
I've pointed out before that the regression tests are not particularly
meant to provide an exhaustive test of WAL recovery. In this particular
case, so far as I can tell the bug is only observable with
full_page_writes turned off --- otherwise XLogInsert will invariably
decide to log the full page, because it's going to see a zeroed-out
LSN in the passed-in buffer. So the odds are good that regression
testing wouldn't have caught it anyway.
I'm in favor of trying to produce a separate set of tests that cover
WAL recovery behavior; but imposing arbitrary restrictions on the
regular regression tests is not the path to get there.
regards, tom lane