On 12/15/15 3:15 PM, Tom Lane wrote:
> Peter Eisentraut <peter_e@gmx.net> writes:
>> On 12/12/15 6:08 AM, Michael Paquier wrote:
>>> And is caused by the fact that master checks that the source node has
>>> been stopped with DB_SHUTDOWNED and not DB_SHUTDOWNED_IN_RECOVERY.
>
>> It seems that the/a problem is actually that the test setup tries a
>> rewind using the standby as the source and the standby also as the
>> target. The source should be the primary, and then it doesn't matter
>> whether we allow DB_SHUTDOWNED_IN_RECOVERY. Need to check that test setup.
>
> While the test setup might have been a bit weird, I think that it
> uncovered an actual bug in pg_rewind, or at least a very undesirable
> and unnecessary restriction. Why should people not be allowed to use
> it to sync in that direction?
That issue was known and was fixed in 9.6. I don't mind that it was
fixed in 9.5 as well.