Change a whole-database VACUUM into doing just pg_attribute, which is
the portion that verifies what we want it to do. The original
formulation wastes a lot of CPU time, which leads the test to fail when
runtime exceeds isolationtester timeout when it's super-slow, such as
under CLOBBER_CACHE_ALWAYS. Per buildfarm member friarbird.
It turns out that the previous shape of the test doesn't always detect
the condition it is supposed to detect (on unpatched reorderbuffer
code): the reason is that there is a good chance of encountering a
xl_running_xacts record (logged every 15 seconds) before the checkpoint
-- and because we advance the xmin when we receive that WAL record, and
we *don't* advance the xmin twice consecutively without receiving a
client message in between, that means the xmin is not advanced enough
for the tuple to be pruned from pg_attribute by VACUUM. So the test
would spuriously pass.
The reason this test deficiency wasn't detected earlier is that HOT
pruning removes the tuple anyway, even if vacuum leaves it in place, so
the test correctly fails (detecting the coding mistake), but for the
wrong reason.
To fix this mess, run the s0_get_changes step twice before vacuum
instead of once: this seems to cause the xmin to be advanced reliably,
wreaking havoc with more certainty.
Author: Arseny Sher
Discussion: https://postgr.es/m/87h8lkuxoa.fsf@ars-thinkpad
Two closely related bugs are fixed. First, xmin of logical slots was
advanced too early. During xl_running_xacts processing, xmin of the
slot was set to the oldest running xid in the record, but that's wrong:
actually, snapshots which will be used for not-yet-replayed transactions
might consider older txns as running too, so we need to keep xmin back
for them. The problem wasn't noticed earlier because DDL which allows
to delete tuple (set xmax) while some another not-yet-committed
transaction looks at it is pretty rare, if not unique: e.g. all forms of
ALTER TABLE which change schema acquire ACCESS EXCLUSIVE lock
conflicting with any inserts. The included test case (test_decoding's
oldest_xmin) uses ALTER of a composite type, which doesn't have such
interlocking.
To deal with this, we must be able to quickly retrieve oldest xmin
(oldest running xid among all assigned snapshots) from ReorderBuffer. To
fix, add another list of ReorderBufferTXNs to the reorderbuffer, where
transactions are sorted by base-snapshot-LSN. This is slightly
different from the existing (sorted by first-LSN) list, because a
transaction can have an earlier LSN but a later Xmin, if its first
record does not obtain an xmin (eg. xl_xact_assignment). Note this new
list doesn't fully replace the existing txn list: we still need that one
to prevent WAL recycling.
The second issue concerns SnapBuilder snapshots and subtransactions.
SnapBuildDistributeNewCatalogSnapshot never assigned a snapshot to a
transaction that is known to be a subtxn, which is good in the common
case that the top-level transaction already has one (no point in doing
so), but a bug otherwise. To fix, arrange to transfer the snapshot from
the subtxn to its top-level txn as soon as the kinship gets known.
test_decoding's snapshot_transfer verifies this.
Also, fix a minor memory leak: refcount of toplevel's old base snapshot
was not decremented when the snapshot is transferred from child.
Liberally sprinkle code comments, and rewrite a few existing ones. This
part is my (Álvaro's) contribution to this commit, as I had to write all
those comments in order to understand the existing code and Arseny's
patch.
Reported-by: Arseny Sher <a.sher@postgrespro.ru>
Diagnosed-by: Arseny Sher <a.sher@postgrespro.ru>
Co-authored-by: Arseny Sher <a.sher@postgrespro.ru>
Co-authored-by: Álvaro Herrera <alvherre@alvh.no-ip.org>
Reviewed-by: Antonin Houska <ah@cybertec.at>
Discussion: https://postgr.es/m/87lgdyz1wj.fsf@ars-thinkpad
Alvaro Herrera