On 2016-07-27 16:53:47 -0700, Andres Freund wrote:
> On 2016-07-07 22:43:59 +0300, Oskari Saarenmaa wrote:
> > Something like the attached patch? It
> > still modifies the toast_delete API, but
> > we could also implement a new
> > speculative-tuple-aware version of it if
> > that's a concern.
>
> Maybe I'm missing something, but toasted columns aren't actually marked
> as speculative right now, no? Cf.
>
> HeapTuple
> toast_insert_or_update(Relation rel, HeapTuple newtup, HeapTuple oldtup,
> int options)
> ..
> /*
> * Ignore the INSERT_SPECULATIVE option. Speculative insertions/super
> * deletions just normally insert/delete the toast values. It seems
> * easiest to deal with that here, instead on, potentially, multiple
> * callers.
> */
> options &= ~HEAP_INSERT_SPECULATIVE;
>
> so going through heap_abort_speculative()/heap_delete_speculative()
> doesn't look right to me. We could change thing so toasted rows also
> get inserted speculatively - but I don't see much of a point.
>
> > +void
> > +heap_delete_speculative(Relation relation, ItemPointer tid, bool is_toast)
> > +{
>
> If we go this way, it seems easier to get is_toast from
> IsToastRelation(relation), rather than hand it through painstakingly.
Oh, and this desperately needs a minimal test. Shouldn't be hard to do
with isolationtester.