Fix rare test failure in nbtree_half_dead_pages

If auto-analyze kicks in at just the right moment, it can hold a
snapshot and prevent the VACUUM command in the test from removing the
deleted tuples. The test needs the tuples to be removed, otherwise no
half-dead page is generated. To fix, introduce a helper procedure to
wait for the removable cutoff to advance, like the one used in the
syscache-update-pruned test for similar purposes.

Thanks to Alexander Lakhin for reproducing and analyzing the test
failure, and Tom Lane for the report.

Discussion: https://www.postgresql.org/message-id/307198.1767408023@sss.pgh.pa.us

diff --git a/src/test/modules/nbtree/expected/nbtree_half_dead_pages.out b/src/test/modules/nbtree/expected/nbtree_half_dead_pages.out
index c435af7a4a5ef8b6f47d1f86039faf6cb0efa3e5..1fdd7d7e5284b93a65ea37652682125dcd1bde7e 100644 (file)
--- a/src/test/modules/nbtree/expected/nbtree_half_dead_pages.out
+++ b/src/test/modules/nbtree/expected/nbtree_half_dead_pages.out
@@ -13,6 +13,28 @@ set client_min_messages TO 'warning';
 create extension if not exists injection_points;
 create extension if not exists amcheck;
 reset client_min_messages;
+-- Wait until all recently-dead tuples on a table become fully dead
+-- and removable by vacuum. (We don't run any concurrent transactions
+-- in the test itself, but auto-analyze can kick in at any time and
+-- hold a transaction open, holding back the vacuum horizon.)
+CREATE PROCEDURE wait_prunable() LANGUAGE plpgsql AS $$
+       DECLARE
+               barrier xid8;
+               cutoff xid8;
+       BEGIN
+               barrier := pg_current_xact_id();
+               -- Pass a shared catalog rather than the table we'll
+               -- prune, to prevent the cutoff from moving
+               -- backwards. See comments at removable_cutoff()
+               LOOP
+                       ROLLBACK;  -- release MyProc->xmin, which could be the oldest
+                       cutoff := removable_cutoff('pg_database');
+                       EXIT WHEN cutoff >= barrier;
+                       RAISE LOG 'removable cutoff %; waiting for %', cutoff, barrier;
+                       PERFORM pg_sleep(.1);
+               END LOOP;
+       END
+$$;
 -- Make all injection points local to this process, for concurrency.
 SELECT injection_points_set_local();
  injection_points_set_local 
@@ -34,6 +56,7 @@ insert into nbtree_half_dead_pages SELECT g from generate_series(1, 150000) g;
 create index nbtree_half_dead_pages_id_idx on nbtree_half_dead_pages using btree (id);
 delete from nbtree_half_dead_pages where id > 100000 and id < 120000;
 -- Run VACUUM and interrupt it so that it leaves behind a half-dead page
+call wait_prunable();
 SELECT injection_points_attach('nbtree-leave-page-half-dead', 'error');
  injection_points_attach 
 -------------------------

diff --git a/src/test/modules/nbtree/sql/nbtree_half_dead_pages.sql b/src/test/modules/nbtree/sql/nbtree_half_dead_pages.sql
index b39cf275557e299ebe852aec786b1ff970857816..c9eb0e50d0efc29511ae84835817720b7f248eae 100644 (file)
--- a/src/test/modules/nbtree/sql/nbtree_half_dead_pages.sql
+++ b/src/test/modules/nbtree/sql/nbtree_half_dead_pages.sql
@@ -15,6 +15,29 @@ create extension if not exists injection_points;
 create extension if not exists amcheck;
 reset client_min_messages;
 
+-- Wait until all recently-dead tuples on a table become fully dead
+-- and removable by vacuum. (We don't run any concurrent transactions
+-- in the test itself, but auto-analyze can kick in at any time and
+-- hold a transaction open, holding back the vacuum horizon.)
+CREATE PROCEDURE wait_prunable() LANGUAGE plpgsql AS $$
+       DECLARE
+               barrier xid8;
+               cutoff xid8;
+       BEGIN
+               barrier := pg_current_xact_id();
+               -- Pass a shared catalog rather than the table we'll
+               -- prune, to prevent the cutoff from moving
+               -- backwards. See comments at removable_cutoff()
+               LOOP
+                       ROLLBACK;  -- release MyProc->xmin, which could be the oldest
+                       cutoff := removable_cutoff('pg_database');
+                       EXIT WHEN cutoff >= barrier;
+                       RAISE LOG 'removable cutoff %; waiting for %', cutoff, barrier;
+                       PERFORM pg_sleep(.1);
+               END LOOP;
+       END
+$$;
+
 -- Make all injection points local to this process, for concurrency.
 SELECT injection_points_set_local();
 
@@ -33,6 +56,7 @@ create index nbtree_half_dead_pages_id_idx on nbtree_half_dead_pages using btree
 delete from nbtree_half_dead_pages where id > 100000 and id < 120000;
 
 -- Run VACUUM and interrupt it so that it leaves behind a half-dead page
+call wait_prunable();
 SELECT injection_points_attach('nbtree-leave-page-half-dead', 'error');
 vacuum nbtree_half_dead_pages;
 SELECT injection_points_detach('nbtree-leave-page-half-dead');