Re: fixing old_snapshot_threshold's time->xid mapping

On Sat, Apr 18, 2020 at 9:27 PM Dilip Kumar <dilipbalaut@gmail.com> wrote:
> On Sat, Apr 18, 2020 at 11:47 AM Thomas Munro <thomas.munro@gmail.com> wrote:
> > I think I found another bug in MaintainOldSnapshotTimeMapping(): if
> > you make time jump by more than old_snapshot_threshold in one go, then
> > the map gets cleared and then no early pruning or snapshot-too-old
> > errors happen.  That's why in 002_too_old.pl it currently advances
> > time by 10 minutes twice, instead of 20 minutes once.  To be
> > continued.
>
> IMHO that doesn't seems to be a problem.  Because even if we jump more
> than old_snapshot_threshold in one go we don't clean complete map
> right.  The latest snapshot timestamp will become the headtimestamp.
> So in TransactionIdLimitedForOldSnapshots if (current_ts -
> old_snapshot_threshold) is still >= head_timestap then we can still do
> early pruning.

Right, thanks.  I got confused about that, and misdiagnosed something
I was seeing.

Here's a new version:

0004: Instead of writing a new kind of TAP test to demonstrate
snapshot-too-old errors, I adjusted the existing isolation tests to
use the same absolute time control technique.  Previously I had
invented a way to do isolation tester-like stuff in TAP tests, which
might be interesting but strange new perl is not necessary for this.

0005: Truncates the time map when the CLOG is truncated.  Its test is
now under src/test/module/snapshot_too_old/t/001_truncate.sql.

These apply on top of Robert's patches, but the only dependency is on
his patch 0001 "Expose oldSnapshotControl.", because now I have stuff
in src/test/module/snapshot_too_old/test_sto.c that wants to mess with
that object too.

Is this an improvement?  I realise that there is still nothing to
actually verify that early pruning has actually happened.  I haven't
thought of a good way to do that yet (stats, page inspection, ...).