]> git.ipfire.org Git - thirdparty/postgresql.git/commit
postgres_fdw: stabilize terminated-connection regression tests REL_18_STABLE github/REL_18_STABLE
authorFujii Masao <fujii@postgresql.org>
Thu, 16 Jul 2026 15:49:26 +0000 (00:49 +0900)
committerFujii Masao <fujii@postgresql.org>
Thu, 16 Jul 2026 15:50:18 +0000 (00:50 +0900)
commita00e43ac32d4f7367567791a68b2cf815f8c4d3f
tree1d53c45fea176e1fa40823afb9d55159fd89af76
parent30c0442eb308cdb9a0d1b8551758775ec4d0e75d
postgres_fdw: stabilize terminated-connection regression tests

The regression test for postgres_fdw_get_connections(true) assumed that
a terminated remote connection would still remain visible in the FDW
connection cache long enough to be reported as closed with a nonzero
remote_backend_pid.

That assumption is not always valid. postgres_fdw_get_connections()
reports only entries that are still present in ConnectionHash, while
pgfdw_inval_callback() may immediately discard an idle cached connection
(xact_depth == 0) when a relevant invalidation arrives. In CI, that can
happen between terminating the remote backend and querying
postgres_fdw_get_connections(true), causing the function to return no
rows.

Adjust the idle-connection test to accept either outcome: if the cache
entry is still present, verify that it reports the expected server name,
closed status, and nonzero remote backend PID; otherwise treat zero rows
as a legitimate result.

To preserve coverage of the terminated-backend reporting path, add a
separate check inside an explicit transaction. In that case, concurrent
invalidation may mark the connection invalid but cannot discard it
before transaction end, so postgres_fdw_get_connections(true) should
still report the terminated connection as in-use, closed, and associated
with a nonzero remote backend PID.

Backpatch to v18, where the affected postgres_fdw_get_connections(true)
test was introduced.

Reported-by: Robert Haas <robertmhaas@gmail.com>
Author: Fujii Masao <masao.fujii@gmail.com>
Reviewed-by: Robert Haas <robertmhaas@gmail.com>
Reviewed-by: Daniel Gustafsson <daniel@yesql.se>
Discussion: https://postgr.es/m/CA+Tgmoax3cHXHsm9OidN4F-xiu16y8q2W8T5dTNFic1Zoo2cOw@mail.gmail.com
Backpatch-through: 18
contrib/postgres_fdw/expected/postgres_fdw.out
contrib/postgres_fdw/sql/postgres_fdw.sql