Hi,
On 2020-03-18 16:54:19 -0300, Alvaro Herrera wrote:
> On 2020-Mar-18, Andres Freund wrote:
> > On 2020-03-18 16:46:23 +0100, Peter Eisentraut wrote:
> > > When SaveSlotToPath() is called with elevel=LOG, the early exits don't
> > > release the slot's io_in_progress_lock. Fix attached.
> >
> > I'm a bit confused as to why we we ever call it with elevel = LOG
> > (i.e. why we have the elevel parameter at all). That seems to have been
> > there from the start, so it's either me or Robert that's to blame. But I
> > can't immediately see a reason for it?
>
> I guess you didn't want failure to save a slot be a reason to abort a
> checkpoint.
I don't see a valid reason for that though - if anything it's dangerous,
because we're not persistently saving the slot. It should fail the
checkpoint imo. Robert, do you have an idea?
Greetings,
Andres Freund