On Thu, 27 Nov 2025 at 18:48, David Rowley <dgrowleyml@gmail.com> wrote:
> This seems to have caused issues on skink [1] under Valgrind. The
> problem seems to be that initscan() does not always initialise
> rs_startblock. I'm now trying to figure out if there's some reason for
> that, or if that's been overlooked at some point.
I've written the attached patch to address the uninitialised
rs_startblock field.
The patch basically adds:
if (!keep_startblock)
scan->rs_startblock = InvalidBlockNumber;
to initscan() when in parallel mode. The rest of the patch is a small
refactor to make it clearer which parts are for parallel and which are
for serial. I also added a comment to mention that the syncscan start
location is figured out in table_block_parallelscan_startblock_init()
for parallel scans.
David
> [1] https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=skink&dt=2025-11-27%2002%3A17%3A35
