| 1 |
# Copyright (c) 2021-2025, PostgreSQL Global Development Group |
| 2 |
|
| 3 |
# Tests statistics handling around restarts, including handling of crashes and |
| 4 |
# invalid stats files, as well as restoring stats after "normal" restarts. |
| 5 |
|
| 6 |
use strict; |
| 7 |
use warnings FATAL => 'all'; |
| 8 |
use PostgreSQL::Test::Cluster; |
| 9 |
use PostgreSQL::Test::Utils; |
| 10 |
use Test::More; |
| 11 |
use File::Copy; |
| 12 |
|
| 13 |
my $node = PostgreSQL::Test::Cluster->new('primary'); |
| 14 |
$node->init(allows_streaming => 1); |
| 15 |
$node->append_conf('postgresql.conf', "track_functions = 'all'"); |
| 16 |
$node->start; |
| 17 |
|
| 18 |
my $connect_db = 'postgres'; |
| 19 |
my $db_under_test = 'test'; |
| 20 |
|
| 21 |
my $sect = "startup"; |
| 22 |
|
| 23 |
# Check some WAL statistics after a fresh startup. The startup process |
| 24 |
# should have done WAL reads, and initialization some WAL writes. |
| 25 |
my $standalone_io_stats = io_stats('init', 'wal', 'standalone backend'); |
| 26 |
my $startup_io_stats = io_stats('normal', 'wal', 'startup'); |
| 27 |
cmp_ok( |
| 28 |
'0', '<', |
| 29 |
$standalone_io_stats->{writes}, |
| 30 |
"$sect: increased standalone backend IO writes"); |
| 31 |
cmp_ok( |
| 32 |
'0', '<', |
| 33 |
$startup_io_stats->{reads}, |
| 34 |
"$sect: increased startup IO reads"); |
| 35 |
|
| 36 |
# create test objects |
| 37 |
$node->safe_psql($connect_db, "CREATE DATABASE $db_under_test"); |
| 38 |
$node->safe_psql($db_under_test, |
| 39 |
"CREATE TABLE tab_stats_crash_discard_test1 AS SELECT generate_series(1,100) AS a" |
| 40 |
); |
| 41 |
$node->safe_psql($db_under_test, |
| 42 |
"CREATE FUNCTION func_stats_crash_discard1() RETURNS VOID AS 'select 2;' LANGUAGE SQL IMMUTABLE" |
| 43 |
); |
| 44 |
|
| 45 |
# collect object oids |
| 46 |
my $dboid = $node->safe_psql($db_under_test, |
| 47 |
"SELECT oid FROM pg_database WHERE datname = '$db_under_test'"); |
| 48 |
my $funcoid = $node->safe_psql($db_under_test, |
| 49 |
"SELECT 'func_stats_crash_discard1()'::regprocedure::oid"); |
| 50 |
my $tableoid = $node->safe_psql($db_under_test, |
| 51 |
"SELECT 'tab_stats_crash_discard_test1'::regclass::oid"); |
| 52 |
|
| 53 |
# generate stats and flush them |
| 54 |
trigger_funcrel_stat(); |
| 55 |
|
| 56 |
# verify stats objects exist |
| 57 |
$sect = "initial"; |
| 58 |
is(have_stats('database', $dboid, 0), 't', "$sect: db stats do exist"); |
| 59 |
is(have_stats('function', $dboid, $funcoid), |
| 60 |
't', "$sect: function stats do exist"); |
| 61 |
is(have_stats('relation', $dboid, $tableoid), |
| 62 |
't', "$sect: relation stats do exist"); |
| 63 |
|
| 64 |
# regular shutdown |
| 65 |
$node->stop(); |
| 66 |
|
| 67 |
# backup stats files |
| 68 |
my $statsfile = $PostgreSQL::Test::Utils::tmp_check . '/' . "discard_stats1"; |
| 69 |
ok(!-f "$statsfile", "backup statsfile cannot already exist"); |
| 70 |
|
| 71 |
my $datadir = $node->data_dir(); |
| 72 |
my $og_stats = "$datadir/pg_stat/pgstat.stat"; |
| 73 |
ok(-f "$og_stats", "origin stats file must exist"); |
| 74 |
copy($og_stats, $statsfile) or die "Copy failed: $!"; |
| 75 |
|
| 76 |
|
| 77 |
## test discarding of stats file after crash etc |
| 78 |
|
| 79 |
$node->start; |
| 80 |
|
| 81 |
$sect = "copy"; |
| 82 |
is(have_stats('database', $dboid, 0), 't', "$sect: db stats do exist"); |
| 83 |
is(have_stats('function', $dboid, $funcoid), |
| 84 |
't', "$sect: function stats do exist"); |
| 85 |
is(have_stats('relation', $dboid, $tableoid), |
| 86 |
't', "$sect: relation stats do exist"); |
| 87 |
|
| 88 |
$node->stop('immediate'); |
| 89 |
|
| 90 |
ok(!-f "$og_stats", "no stats file should exist after immediate shutdown"); |
| 91 |
|
| 92 |
# copy the old stats back to test we discard stats after crash restart |
| 93 |
copy($statsfile, $og_stats) or die "Copy failed: $!"; |
| 94 |
|
| 95 |
$node->start; |
| 96 |
|
| 97 |
# stats should have been discarded |
| 98 |
$sect = "post immediate"; |
| 99 |
is(have_stats('database', $dboid, 0), 'f', "$sect: db stats do not exist"); |
| 100 |
is(have_stats('function', $dboid, $funcoid), |
| 101 |
'f', "$sect: function stats do exist"); |
| 102 |
is(have_stats('relation', $dboid, $tableoid), |
| 103 |
'f', "$sect: relation stats do not exist"); |
| 104 |
|
| 105 |
# get rid of backup statsfile |
| 106 |
unlink $statsfile or die "cannot unlink $statsfile $!"; |
| 107 |
|
| 108 |
|
| 109 |
# generate new stats and flush them |
| 110 |
trigger_funcrel_stat(); |
| 111 |
|
| 112 |
$sect = "post immediate, new"; |
| 113 |
is(have_stats('database', $dboid, 0), 't', "$sect: db stats do exist"); |
| 114 |
is(have_stats('function', $dboid, $funcoid), |
| 115 |
't', "$sect: function stats do exist"); |
| 116 |
is(have_stats('relation', $dboid, $tableoid), |
| 117 |
't', "$sect: relation stats do exist"); |
| 118 |
|
| 119 |
# regular shutdown |
| 120 |
$node->stop(); |
| 121 |
|
| 122 |
|
| 123 |
## check an invalid stats file is handled |
| 124 |
|
| 125 |
overwrite_file($og_stats, "ZZZZZZZZZZZZZ"); |
| 126 |
|
| 127 |
# normal startup and no issues despite invalid stats file |
| 128 |
$node->start; |
| 129 |
|
| 130 |
# no stats present due to invalid stats file |
| 131 |
$sect = "invalid_overwrite"; |
| 132 |
is(have_stats('database', $dboid, 0), 'f', "$sect: db stats do not exist"); |
| 133 |
is(have_stats('function', $dboid, $funcoid), |
| 134 |
'f', "$sect: function stats do not exist"); |
| 135 |
is(have_stats('relation', $dboid, $tableoid), |
| 136 |
'f', "$sect: relation stats do not exist"); |
| 137 |
|
| 138 |
|
| 139 |
## check invalid stats file starting with valid contents, but followed by |
| 140 |
## invalid content is handled. |
| 141 |
|
| 142 |
trigger_funcrel_stat(); |
| 143 |
$node->stop; |
| 144 |
append_file($og_stats, "XYZ"); |
| 145 |
$node->start; |
| 146 |
|
| 147 |
$sect = "invalid_append"; |
| 148 |
is(have_stats('database', $dboid, 0), 'f', "$sect: db stats do not exist"); |
| 149 |
is(have_stats('function', $dboid, $funcoid), |
| 150 |
'f', "$sect: function stats do not exist"); |
| 151 |
is(have_stats('relation', $dboid, $tableoid), |
| 152 |
'f', "$sect: relation stats do not exist"); |
| 153 |
|
| 154 |
|
| 155 |
## checks related to stats persistency around restarts and resets |
| 156 |
|
| 157 |
# Ensure enough checkpoints to protect against races for test after reset, |
| 158 |
# even on very slow machines. |
| 159 |
$node->safe_psql($connect_db, "CHECKPOINT; CHECKPOINT;"); |
| 160 |
|
| 161 |
|
| 162 |
## check checkpoint and wal stats are incremented due to restart |
| 163 |
|
| 164 |
my $ckpt_start = checkpoint_stats(); |
| 165 |
my $wal_start = wal_stats(); |
| 166 |
$node->restart; |
| 167 |
|
| 168 |
$sect = "post restart"; |
| 169 |
my $ckpt_restart = checkpoint_stats(); |
| 170 |
my $wal_restart = wal_stats(); |
| 171 |
|
| 172 |
cmp_ok( |
| 173 |
$ckpt_start->{count}, '<', |
| 174 |
$ckpt_restart->{count}, |
| 175 |
"$sect: increased checkpoint count"); |
| 176 |
cmp_ok( |
| 177 |
$wal_start->{records}, '<', |
| 178 |
$wal_restart->{records}, |
| 179 |
"$sect: increased wal record count"); |
| 180 |
cmp_ok($wal_start->{bytes}, '<', $wal_restart->{bytes}, |
| 181 |
"$sect: increased wal bytes"); |
| 182 |
is( $ckpt_start->{reset}, |
| 183 |
$ckpt_restart->{reset}, |
| 184 |
"$sect: checkpoint stats_reset equal"); |
| 185 |
is($wal_start->{reset}, $wal_restart->{reset}, |
| 186 |
"$sect: wal stats_reset equal"); |
| 187 |
|
| 188 |
|
| 189 |
## Check that checkpoint stats are reset, WAL stats aren't affected |
| 190 |
|
| 191 |
$node->safe_psql($connect_db, "SELECT pg_stat_reset_shared('checkpointer')"); |
| 192 |
|
| 193 |
$sect = "post ckpt reset"; |
| 194 |
my $ckpt_reset = checkpoint_stats(); |
| 195 |
my $wal_ckpt_reset = wal_stats(); |
| 196 |
|
| 197 |
cmp_ok($ckpt_restart->{count}, |
| 198 |
'>', $ckpt_reset->{count}, "$sect: checkpoint count smaller"); |
| 199 |
cmp_ok($ckpt_start->{reset}, 'lt', $ckpt_reset->{reset}, |
| 200 |
"$sect: stats_reset newer"); |
| 201 |
|
| 202 |
cmp_ok( |
| 203 |
$wal_restart->{records}, |
| 204 |
'<=', |
| 205 |
$wal_ckpt_reset->{records}, |
| 206 |
"$sect: wal record count not affected by reset"); |
| 207 |
is( $wal_start->{reset}, |
| 208 |
$wal_ckpt_reset->{reset}, |
| 209 |
"$sect: wal stats_reset equal"); |
| 210 |
|
| 211 |
|
| 212 |
## check that checkpoint stats stay reset after restart |
| 213 |
|
| 214 |
$node->restart; |
| 215 |
|
| 216 |
$sect = "post ckpt reset & restart"; |
| 217 |
my $ckpt_restart_reset = checkpoint_stats(); |
| 218 |
my $wal_restart2 = wal_stats(); |
| 219 |
|
| 220 |
# made sure above there's enough checkpoints that this will be stable even on slow machines |
| 221 |
cmp_ok( |
| 222 |
$ckpt_restart_reset->{count}, |
| 223 |
'<', |
| 224 |
$ckpt_restart->{count}, |
| 225 |
"$sect: checkpoint still reset"); |
| 226 |
is($ckpt_restart_reset->{reset}, |
| 227 |
$ckpt_reset->{reset}, "$sect: stats_reset same"); |
| 228 |
|
| 229 |
cmp_ok( |
| 230 |
$wal_ckpt_reset->{records}, |
| 231 |
'<', |
| 232 |
$wal_restart2->{records}, |
| 233 |
"$sect: increased wal record count"); |
| 234 |
cmp_ok( |
| 235 |
$wal_ckpt_reset->{bytes}, |
| 236 |
'<', |
| 237 |
$wal_restart2->{bytes}, |
| 238 |
"$sect: increased wal bytes"); |
| 239 |
is( $wal_start->{reset}, |
| 240 |
$wal_restart2->{reset}, |
| 241 |
"$sect: wal stats_reset equal"); |
| 242 |
|
| 243 |
|
| 244 |
## check WAL stats stay reset |
| 245 |
|
| 246 |
$node->safe_psql($connect_db, "SELECT pg_stat_reset_shared('wal')"); |
| 247 |
|
| 248 |
$sect = "post wal reset"; |
| 249 |
my $wal_reset = wal_stats(); |
| 250 |
|
| 251 |
cmp_ok( |
| 252 |
$wal_reset->{records}, '<', |
| 253 |
$wal_restart2->{records}, |
| 254 |
"$sect: smaller record count"); |
| 255 |
cmp_ok( |
| 256 |
$wal_reset->{bytes}, '<', |
| 257 |
$wal_restart2->{bytes}, |
| 258 |
"$sect: smaller bytes"); |
| 259 |
cmp_ok( |
| 260 |
$wal_reset->{reset}, 'gt', |
| 261 |
$wal_restart2->{reset}, |
| 262 |
"$sect: newer stats_reset"); |
| 263 |
|
| 264 |
$node->restart; |
| 265 |
|
| 266 |
$sect = "post wal reset & restart"; |
| 267 |
my $wal_reset_restart = wal_stats(); |
| 268 |
|
| 269 |
# enough WAL generated during prior tests and initdb to make this not racy |
| 270 |
cmp_ok( |
| 271 |
$wal_reset_restart->{records}, |
| 272 |
'<', |
| 273 |
$wal_restart2->{records}, |
| 274 |
"$sect: smaller record count"); |
| 275 |
cmp_ok( |
| 276 |
$wal_reset->{bytes}, '<', |
| 277 |
$wal_restart2->{bytes}, |
| 278 |
"$sect: smaller bytes"); |
| 279 |
cmp_ok( |
| 280 |
$wal_reset->{reset}, 'gt', |
| 281 |
$wal_restart2->{reset}, |
| 282 |
"$sect: newer stats_reset"); |
| 283 |
|
| 284 |
$node->stop('immediate'); |
| 285 |
$node->start; |
| 286 |
|
| 287 |
$sect = "post immediate restart"; |
| 288 |
my $wal_restart_immediate = wal_stats(); |
| 289 |
|
| 290 |
cmp_ok( |
| 291 |
$wal_reset_restart->{reset}, |
| 292 |
'lt', |
| 293 |
$wal_restart_immediate->{reset}, |
| 294 |
"$sect: reset timestamp is new"); |
| 295 |
|
| 296 |
$node->stop; |
| 297 |
done_testing(); |
| 298 |
|
| 299 |
sub trigger_funcrel_stat |
| 300 |
{ |
| 301 |
$node->safe_psql( |
| 302 |
$db_under_test, q[ |
| 303 |
SELECT * FROM tab_stats_crash_discard_test1; |
| 304 |
SELECT func_stats_crash_discard1(); |
| 305 |
SELECT pg_stat_force_next_flush();]); |
| 306 |
} |
| 307 |
|
| 308 |
sub have_stats |
| 309 |
{ |
| 310 |
my ($kind, $dboid, $objid) = @_; |
| 311 |
|
| 312 |
return $node->safe_psql($connect_db, |
| 313 |
"SELECT pg_stat_have_stats('$kind', $dboid, $objid)"); |
| 314 |
} |
| 315 |
|
| 316 |
sub overwrite_file |
| 317 |
{ |
| 318 |
my ($filename, $str) = @_; |
| 319 |
open my $fh, ">", $filename |
| 320 |
or die "could not overwrite \"$filename\": $!"; |
| 321 |
print $fh $str; |
| 322 |
close $fh; |
| 323 |
return; |
| 324 |
} |
| 325 |
|
| 326 |
sub append_file |
| 327 |
{ |
| 328 |
my ($filename, $str) = @_; |
| 329 |
open my $fh, ">>", $filename |
| 330 |
or die "could not append to \"$filename\": $!"; |
| 331 |
print $fh $str; |
| 332 |
close $fh; |
| 333 |
return; |
| 334 |
} |
| 335 |
|
| 336 |
sub checkpoint_stats |
| 337 |
{ |
| 338 |
my %results; |
| 339 |
|
| 340 |
$results{count} = $node->safe_psql($connect_db, |
| 341 |
"SELECT num_timed + num_requested FROM pg_stat_checkpointer"); |
| 342 |
$results{reset} = $node->safe_psql($connect_db, |
| 343 |
"SELECT stats_reset FROM pg_stat_checkpointer"); |
| 344 |
|
| 345 |
return \%results; |
| 346 |
} |
| 347 |
|
| 348 |
sub wal_stats |
| 349 |
{ |
| 350 |
my %results; |
| 351 |
$results{records} = |
| 352 |
$node->safe_psql($connect_db, "SELECT wal_records FROM pg_stat_wal"); |
| 353 |
$results{bytes} = |
| 354 |
$node->safe_psql($connect_db, "SELECT wal_bytes FROM pg_stat_wal"); |
| 355 |
$results{reset} = |
| 356 |
$node->safe_psql($connect_db, "SELECT stats_reset FROM pg_stat_wal"); |
| 357 |
|
| 358 |
return \%results; |
| 359 |
} |
| 360 |
|
| 361 |
sub io_stats |
| 362 |
{ |
| 363 |
my ($context, $object, $backend_type) = @_; |
| 364 |
my %results; |
| 365 |
|
| 366 |
$results{writes} = $node->safe_psql( |
| 367 |
$connect_db, qq{SELECT writes FROM pg_stat_io |
| 368 |
WHERE context = '$context' AND object = '$object' AND |
| 369 |
backend_type = '$backend_type'}); |
| 370 |
$results{reads} = $node->safe_psql( |
| 371 |
$connect_db, qq{SELECT reads FROM pg_stat_io |
| 372 |
WHERE context = '$context' AND object = '$object' AND |
| 373 |
backend_type = '$backend_type'}); |
| 374 |
|
| 375 |
return \%results; |
| 376 |
} |