Re: Deleting older versions in unique indexes to avoid page splits

пн, 26 окт. 2020 г. в 22:15, Peter Geoghegan <pg(at)bowt(dot)ie>:

> Attached is v5, which has changes that are focused on two important
> high level goals:
>

And some more comments after another round of reading the patch.

1. Looks like UNIQUE_CHECK_NO_WITH_UNCHANGED is used for HOT updates,
should we use UNIQUE_CHECK_NO_HOT here? It is better understood like
this.

2. You're modifying the table_tuple_update() function on 1311 line of
include/access/tableam.h,
adding modified_attrs_hint. There's a large comment right before it
describing parameters,
I think there should be a note about modified_attrs_hint parameter in
that comment, 'cos
it is referenced from other places in tableam.h and also from
backend/access/heap/heapam.c

3. Can you elaborate on the scoring model you're using?
Why do we expect a score of 25, what's the rationale behind this number?
And should it be #define-d ?

4. heap_compute_xid_horizon_for_tuples contains duplicate logic. Is it
possible to avoid this?

5. In this comment

+ * heap_index_batch_check() helper function. Sorts deltids array in the
+ * order needed for useful processing.

perhaps it is better to replace "useful" with more details? Or point to
the place
where "useful processing" is described.

6. In this snippet in _bt_dedup_delete_pass()

+ else if (_bt_keep_natts_fast(rel, state->base, itup) > nkeyatts &&
+ _bt_dedup_save_htid(state, itup))
+ {
+
+ }

I would rather add a comment, explaining that the empty body of the
clause is actually expected.

7. In the _bt_dedup_delete_finish_pending() you're setting ispromising to
false for both
posting and non-posting tuples. This contradicts comments before
function.

--
Victor Yegorov