Skip to content

Commit e2e9928

Browse files
melanieplagemanmelanieplageman
committed
Test that vacuum removes tuples older than OldestXmin
If vacuum fails to prune a tuple killed before OldestXmin, it will later find that tuple dead in lazy_scan_prune() and loop infinitely. Add a test reproducing this scenario to the recovery suite which creates a table on a primary, updates the table to generate dead tuples for vacuum, and then, during the vacuum, uses a replica to force GlobalVisState->maybe_needed on the primary to move backwards and precede the value of OldestXmin set at the beginning of vacuuming the table. This commit is separate from the fix in case there are test stability issues. Discussion of the bug: https://postgr.es/m/CAAKRu_Y_NJzF4-8gzTTeaOuUL3CcGoXPjXcAHbTTygT8AyVqag%40mail.gmail.com Discussion of the test: https://postgr.es/m/CAAKRu_apNU2MPBK96V%2BbXjTq0RiZ-%3DA4ZTaysakpx9jxbq1dbQ%40mail.gmail.com Author: Melanie Plageman Reviewed-by: Peter Geoghegan
1 parent dc6354c commit e2e9928

File tree

1 file changed

+246
-0
lines changed

1 file changed

+246
-0
lines changed

‎src/test/recovery/t/043_vacuum_horizon_floor.pl

Lines changed: 246 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,246 @@
1+
use strict;
2+
use warnings;
3+
use PostgreSQL::Test::Cluster;
4+
use Test::More;
5+
6+
# Test that vacuum prunes away all dead tuples killed before OldestXmin
7+
8+
# Set up nodes
9+
my $node_primary = PostgreSQL::Test::Cluster->new('primary');
10+
$node_primary->init(allows_streaming => 'physical');
11+
12+
$node_primary->append_conf(
13+
'postgresql.conf', qq[
14+
hot_standby_feedback = on
15+
autovacuum = off
16+
log_min_messages = INFO
17+
maintenance_work_mem = 1024
18+
]);
19+
$node_primary->start;
20+
21+
my $node_replica = PostgreSQL::Test::Cluster->new('standby');
22+
23+
$node_primary->backup('my_backup');
24+
$node_replica->init_from_backup($node_primary, 'my_backup',
25+
has_streaming => 1);
26+
27+
$node_replica->start;
28+
29+
my $test_db = "test_db";
30+
$node_primary->safe_psql('postgres', "CREATE DATABASE $test_db");
31+
32+
# Save the original connection info for later use
33+
my $orig_conninfo = $node_primary->connstr();
34+
35+
my $table1 = "vac_horizon_floor_table";
36+
37+
# Long-running Primary Session A
38+
my $psql_primaryA =
39+
$node_primary->background_psql($test_db, on_error_stop => 1);
40+
41+
# Long-running Primary Session B
42+
my $psql_primaryB =
43+
$node_primary->background_psql($test_db, on_error_stop => 1);
44+
45+
# Because vacuum's first pass, pruning, is where we use the GlobalVisState to
46+
# check tuple visibility, GlobalVisState->maybe_needed must move backwards
47+
# during pruning before checking the visibility for a tuple which would have
48+
# been considered HEAPTUPLE_DEAD prior to maybe_needed moving backwards but
49+
# HEAPTUPLE_RECENTLY_DEAD compared to the new, older value of maybe_needed.
50+
#
51+
# We must not only force the horizon on the primary to move backwards but also
52+
# force the vacuuming backend's GlobalVisState to be updated. GlobalVisState
53+
# is forced to update during index vacuuming.
54+
#
55+
# _bt_pendingfsm_finalize() calls GetOldestNonRemovableTransactionId() at the
56+
# end of a round of index vacuuming, updating the backend's GlobalVisState
57+
# and, in our case, moving maybe_needed backwards.
58+
#
59+
# Then vacuum's first (pruning) pass will continue and pruning will find our
60+
# later inserted and updated tuple HEAPTUPLE_RECENTLY_DEAD when compared to
61+
# maybe_needed but HEAPTUPLE_DEAD when compared to OldestXmin.
62+
#
63+
# Thus, we must force at least two rounds of index vacuuming to ensure that
64+
# some tuple visibility checks will happen after a round of index vacuuming.
65+
# To accomplish this, we set maintenance_work_mem to its minimum value and
66+
# insert and update enough rows that we force at least one round of index
67+
# vacuuming before getting to a dead tuple which was killed after the
68+
# standby is disconnected.
69+
$node_primary->safe_psql($test_db, qq[
70+
CREATE TABLE ${table1}(col1 int) with (autovacuum_enabled=false);
71+
INSERT INTO $table1 SELECT generate_series(1, 200000);
72+
CREATE INDEX on ${table1}(col1);
73+
UPDATE $table1 SET col1 = 0 WHERE col1 > 1;
74+
INSERT INTO $table1 VALUES(1);
75+
]);
76+
77+
# We will later move the primary forward while the standby is disconnected.
78+
# For now, however, there is no reason not to wait for the standby to catch
79+
# up.
80+
my $primary_lsn = $node_primary->lsn('flush');
81+
$node_primary->wait_for_catchup($node_replica, 'replay', $primary_lsn);
82+
83+
# Test that the WAL receiver is up and running.
84+
$node_replica->poll_query_until($test_db, qq[
85+
select exists (select * from pg_stat_wal_receiver);] , 't');
86+
87+
# Set primary_conninfo to something invalid on the replica and reload the
88+
# config. Once the config is reloaded, the startup process will force the WAL
89+
# receiver to restart and it will be unable to reconnect because of the
90+
# invalid connection information.
91+
$node_replica->safe_psql($test_db, qq[
92+
ALTER SYSTEM SET primary_conninfo = '';
93+
SELECT pg_reload_conf();
94+
]);
95+
96+
# Wait until the WAL receiver has shut down and been unable to start up again.
97+
$node_replica->poll_query_until($test_db, qq[
98+
select exists (select * from pg_stat_wal_receiver);] , 'f');
99+
100+
# Now insert and update a tuple which will be visible to the vacuum on the
101+
# primary but which will have xmax newer than the oldest xmin on the standby
102+
# that was recently disconnected.
103+
my $res = $psql_primaryA->query_safe(
104+
qq[
105+
INSERT INTO $table1 VALUES (99);
106+
UPDATE $table1 SET col1 = 100 WHERE col1 = 99;
107+
SELECT 'after_update';
108+
]
109+
);
110+
111+
# Make sure the UPDATE finished
112+
like($res, qr/^after_update$/m, "UPDATE occurred on primary session A");
113+
114+
# Open a cursor on the primary whose pin will keep VACUUM from getting a
115+
# cleanup lock on the first page of the relation. We want VACUUM to be able to
116+
# start, calculate initial values for OldestXmin and GlobalVisState and then
117+
# be unable to proceed with pruning our dead tuples. This will allow us to
118+
# reconnect the standby and push the horizon back before we start actual
119+
# pruning and vacuuming.
120+
my $primary_cursor1 = "vac_horizon_floor_cursor";
121+
122+
# The first value inserted into the table was a 1, so FETCH FORWARD should
123+
# return a 1. That's how we know the cursor has a pin.
124+
$res = $psql_primaryB->query_safe(
125+
qq[
126+
BEGIN;
127+
DECLARE $primary_cursor1 CURSOR FOR SELECT * FROM $table1 WHERE col1 = 1;
128+
FETCH FORWARD FROM $primary_cursor1;
129+
]
130+
);
131+
132+
is($res, 1, qq[Cursor query returned $res. Expected value 1.]);
133+
134+
# Get the PID of the session which will run the VACUUM FREEZE so that we can
135+
# use it to filter pg_stat_activity later.
136+
my $vacuum_pid = $psql_primaryA->query_safe("SELECT pg_backend_pid();");
137+
138+
# Now start a VACUUM FREEZE on the primary. It will call vacuum_get_cutoffs()
139+
# and establish values of OldestXmin and GlobalVisState which are newer than
140+
# all of our dead tuples. Then it will be unable to get a cleanup lock to
141+
# start pruning, so it will hang. We use VACUUM FREEZE because it will wait
142+
# for a cleanup lock instead of skipping the page pinned by the cursor.
143+
$psql_primaryA->{stdin} .= qq[
144+
VACUUM (VERBOSE, FREEZE) $table1;
145+
\\echo VACUUM
146+
];
147+
148+
# Make sure the VACUUM command makes it to the server.
149+
$psql_primaryA->{run}->pump_nb();
150+
151+
# Make sure that the VACUUM has already called vacuum_get_cutoffs() and is
152+
# just waiting on the lock to start vacuuming. We don't want the standby to
153+
# re-establish a connection to the primary and push the horizon back until
154+
# we've saved initial values in GlobalVisState and calculated OldestXmin.
155+
$node_primary->poll_query_until($test_db,
156+
qq[
157+
SELECT count(*) >= 1 FROM pg_stat_activity
158+
WHERE pid = $vacuum_pid
159+
AND wait_event = 'BufferPin';
160+
],
161+
't');
162+
163+
# Ensure the WAL receiver is still not active on the replica.
164+
$node_replica->poll_query_until($test_db, qq[
165+
select exists (select * from pg_stat_wal_receiver);] , 'f');
166+
167+
# Allow the WAL receiver connection to re-establish.
168+
$node_replica->safe_psql(
169+
$test_db, qq[
170+
ALTER SYSTEM SET primary_conninfo = '$orig_conninfo';
171+
SELECT pg_reload_conf();
172+
]);
173+
174+
# Ensure the new WAL receiver has connected.
175+
$node_replica->poll_query_until($test_db, qq[
176+
select exists (select * from pg_stat_wal_receiver);] , 't');
177+
178+
# Once the WAL sender is shown on the primary, the replica should have
179+
# connected with the primary and pushed the horizon backward. Primary Session
180+
# A won't see that until the VACUUM FREEZE proceeds and does its first round
181+
# of index vacuuming.
182+
$node_primary->poll_query_until($test_db, qq[
183+
select exists (select * from pg_stat_replication);] , 't');
184+
185+
# Move the cursor forward to the next 1. We inserted the 1 much later, so
186+
# advancing the cursor should allow vacuum to proceed vacuuming most pages of
187+
# the relation. Because we set maintanence_work_mem sufficiently low, we
188+
# expect that a round of index vacuuming has happened and that the vacuum is
189+
# now waiting for the cursor to release its pin on the last page of the
190+
# relation.
191+
$res = $psql_primaryB->query_safe("FETCH $primary_cursor1");
192+
is($res, 1,
193+
qq[Cursor query returned $res from second fetch. Expected value 1.]);
194+
195+
# Prevent the test from incorrectly passing by confirming that we did indeed
196+
# do a pass of index vacuuming.
197+
$node_primary->poll_query_until($test_db, qq[
198+
SELECT index_vacuum_count > 0
199+
FROM pg_stat_progress_vacuum
200+
WHERE datname='$test_db' AND relid::regclass = '$table1'::regclass;
201+
] , 't');
202+
203+
# Commit the transaction with the open cursor so that the VACUUM can finish.
204+
$psql_primaryB->query_until(
205+
qr/^commit$/m,
206+
qq[
207+
COMMIT;
208+
\\echo commit
209+
]
210+
);
211+
212+
# VACUUM proceeds with pruning and does a visibility check on each tuple. In
213+
# older versions of Postgres, pruning found our final dead tuple
214+
# non-removable (HEAPTUPLE_RECENTLY_DEAD) since its xmax is after the new
215+
# value of maybe_needed. Then lazy_scan_prune() would infinitely loop
216+
# because HeapTupleSatisfiesVacuum() would find the tuple HEAPTUPLE_DEAD
217+
# because its xmax preceded OldestXmin. This was fixed by passing
218+
# OldestXmin to heap_page_prune() and removing all tuples whose xmaxes
219+
# precede OldestXmin.
220+
#
221+
# With the fix, VACUUM should finish successfully, incrementing the table
222+
# vacuum_count.
223+
$node_primary->poll_query_until($test_db,
224+
qq[
225+
SELECT vacuum_count > 0
226+
FROM pg_stat_all_tables WHERE relname = '${table1}';
227+
]
228+
, 't');
229+
230+
$primary_lsn = $node_primary->lsn('flush');
231+
232+
# Make sure something causes us to flush
233+
$node_primary->safe_psql($test_db, "INSERT INTO $table1 VALUES (1);");
234+
235+
# Nothing on the replica should cause a recovery conflict, so this should
236+
# finish successfully.
237+
$node_primary->wait_for_catchup($node_replica, 'replay', $primary_lsn);
238+
239+
## Shut down psqls
240+
$psql_primaryA->quit;
241+
$psql_primaryB->quit;
242+
243+
$node_replica->stop();
244+
$node_primary->stop();
245+
246+
done_testing();

0 commit comments

Comments
 (0)