On 09/12/2023 23:48, Peter Geoghegan wrote:
On Tue, Mar 21, 2023 at 3:37 PM Peter Geoghegan <p...@bowt.ie> wrote:
I think that we should do something like the attached, to completely
avoid this ambiguity. This patch adds a new XLOG_HEAP2 bit that's
similar to XLOG_HEAP_INIT_PAGE -- XLOG_HEAP2_BYVACUUM. This allows all
XLOG_HEAP2 record types to indicate that they took place during
VACUUM, by XOR'ing the flag with the record type/info when
XLogInsert() is called. For now this is only used by PRUNE records.
Tools like pg_walinspect will report a separate "Heap2/PRUNE+BYVACUUM"
record_type, as well as the unadorned Heap2/PRUNE record_type, which
we'll now know must have been opportunistic pruning.
The approach of using a bit in the style of the heapam init bit makes
sense to me, because the bit is available, and works in a way that is
minimally invasive. Also, one can imagine needing to resolve a similar
ambiguity in the future, when (say) opportunistic freezing is added.
Starting a new, dedicated thread to keep track of this in the CF app.
This patch bitrot. Attached is v2, rebased on top of HEAD.
I included changes like this in commit f83d709760 ("Merge prune, freeze
and vacuum WAL record formats"). Marking this as Committed in the
commitfest.
--
Heikki Linnakangas
Neon (https://neon.tech)
