"Simon Riggs" <simon@2ndquadrant.com> writes:
> On Wed, 2007-07-25 at 22:20 +0100, Gregory Stark wrote:
>> Also, it might let us set up a standby database in the regression
>> suite which would give us some xlog coverage as well which is a
>> critical portion of the system the regression suite doesn't test at
>> all.
> I like that general thought and will help where I can.
One pretty crude test I use a lot is to "kill -9" the bgwriter or some
other subprocess immediately after the regression tests finish.
Assuming you have checkpoints far enough apart, this forces a replay
from WAL of the entire regression sequence. Doesn't exercise archiving
though.
regards, tom lane