Skip to content

Commit 99b4a63

Browse files
michaelpqmichaelpq
committed
Add basic TAP tests for the low-level backup method
There are currently no tests for the low-level backup method where pg_backup_start() and pg_backup_stop() are involved while taking a file-system backup. The tests introduced in this commit rely on a background psql process to make sure that the backup is taken while the session doing the SQL start and stop calls remains alive. Two cases are checked here with the backup taken: - Recovery without a backup_label, leading to a corrupted state. - Recovery with a backup_label, with a consistent state reached. Both cases cross-check some patterns in the logs generated when running recovery. Author: David Steele Discussion: https://postgr.es/m/f20fcc82-dadb-478d-beb4-1e2ffb0ace76@pgmasters.net
1 parent e820db5 commit 99b4a63

File tree

1 file changed

+120
-0
lines changed

1 file changed

+120
-0
lines changed

‎src/test/recovery/t/042_low_level_backup.pl

Lines changed: 120 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,120 @@
1+
2+
# Copyright (c) 2024, PostgreSQL Global Development Group
3+
4+
# Test low-level backup method by using pg_backup_start() and pg_backup_stop()
5+
# to create backups.
6+
7+
use strict;
8+
use warnings;
9+
10+
use File::Copy qw(copy);
11+
use File::Path qw(rmtree);
12+
use PostgreSQL::Test::Cluster;
13+
use PostgreSQL::Test::Utils;
14+
use Test::More;
15+
16+
# Start primary node with archiving.
17+
my $node_primary = PostgreSQL::Test::Cluster->new('primary');
18+
$node_primary->init(has_archiving => 1, allows_streaming => 1);
19+
$node_primary->start;
20+
21+
# Start backup.
22+
my $backup_name = 'backup1';
23+
my $psql = $node_primary->background_psql('postgres');
24+
25+
$psql->query_safe("SET client_min_messages TO WARNING");
26+
$psql->set_query_timer_restart;
27+
$psql->query_safe("select pg_backup_start('test label')");
28+
29+
# Copy files.
30+
my $backup_dir = $node_primary->backup_dir . '/' . $backup_name;
31+
32+
PostgreSQL::Test::RecursiveCopy::copypath($node_primary->data_dir,
33+
$backup_dir);
34+
35+
# Cleanup some files/paths that should not be in the backup. There is no
36+
# attempt to handle all the exclusions done by pg_basebackup here, in part
37+
# because these are not required, but also to keep the test simple.
38+
#
39+
# Also remove pg_control because it needs to be copied later.
40+
unlink("$backup_dir/postmaster.pid")
41+
or BAIL_OUT("unable to unlink $backup_dir/postmaster.pid");
42+
unlink("$backup_dir/postmaster.opts")
43+
or BAIL_OUT("unable to unlink $backup_dir/postmaster.opts");
44+
unlink("$backup_dir/global/pg_control")
45+
or BAIL_OUT("unable to unlink $backup_dir/global/pg_control");
46+
47+
rmtree("$backup_dir/pg_wal")
48+
or BAIL_OUT("unable to unlink contents of $backup_dir/pg_wal");
49+
mkdir("$backup_dir/pg_wal");
50+
51+
# Create a table that will be used to verify that recovery started at the
52+
# correct location, rather than a location recorded in the control file.
53+
$psql->query_safe("create table canary (id int)");
54+
55+
# Advance the checkpoint location in pg_control past the location where the
56+
# backup started. Switch WAL to make it really clear that the location is
57+
# different and to put the checkpoint in a new WAL segment.
58+
$psql->query_safe("select pg_switch_wal()");
59+
$psql->query_safe("checkpoint");
60+
61+
# Copy pg_control last so it contains the new checkpoint.
62+
copy($node_primary->data_dir . '/global/pg_control',
63+
"$backup_dir/global/pg_control")
64+
or BAIL_OUT("unable to copy global/pg_control");
65+
66+
# Stop backup and get backup_label
67+
my $backup_label =
68+
$psql->query_safe("select labelfile from pg_backup_stop()");
69+
70+
$psql->quit;
71+
72+
# Rather than writing out backup_label, try to recover the backup without
73+
# backup_label to demonstrate that recovery will not work correctly without it,
74+
# i.e. the canary table will be missing and the cluster will be corrupted.
75+
# Provide only the WAL segment that recovery will think it needs.
76+
#
77+
# The point of this test is to explicitly demonstrate that backup_label is
78+
# being used in a later test to get the correct recovery info.
79+
my $node_replica = PostgreSQL::Test::Cluster->new('replica_fail');
80+
$node_replica->init_from_backup($node_primary, $backup_name);
81+
my $canary_query = "select count(*) from pg_class where relname = 'canary'";
82+
83+
copy(
84+
$node_primary->archive_dir . '/000000010000000000000003',
85+
$node_replica->data_dir . '/pg_wal/000000010000000000000003'
86+
) or BAIL_OUT("unable to copy 000000010000000000000003");
87+
88+
$node_replica->start;
89+
90+
ok($node_replica->safe_psql('postgres', $canary_query) == 0,
91+
'canary is missing');
92+
93+
# Check log to ensure that crash recovery was used as there is no
94+
# backup_label.
95+
ok( $node_replica->log_contains(
96+
'database system was not properly shut down; automatic recovery in progress'
97+
),
98+
'verify backup recovery performed with crash recovery');
99+
100+
$node_replica->teardown_node;
101+
$node_replica->clean_node;
102+
103+
# Save backup_label into the backup directory and recover using the primary's
104+
# archive. This time recovery will succeed and the canary table will be
105+
# present.
106+
append_to_file("$backup_dir/backup_label", $backup_label);
107+
108+
$node_replica = PostgreSQL::Test::Cluster->new('replica_success');
109+
$node_replica->init_from_backup($node_primary, $backup_name,
110+
has_restoring => 1);
111+
$node_replica->start;
112+
113+
ok($node_replica->safe_psql('postgres', $canary_query) == 1,
114+
'canary is present');
115+
116+
# Check log to ensure that backup_label was used for recovery.
117+
ok($node_replica->log_contains('starting backup recovery with redo LSN'),
118+
'verify backup recovery performed with backup_label');
119+
120+
done_testing();

0 commit comments

Comments
 (0)