[yugabyte#15660] YSQL: ANALYZE catalog tables as part of initdb

Summary:
With this diff, we analyze catalog tables as part of initdb.

ANALYZE is run only once for the template1 database.
Collected stats are copied from template1 to template0, postgres, and yugabyte databases because they use template1 as their template database.

Since PG system cache is not invalidated (stale cache) after direct UPDATE on catalog tables if the cache uses OID as part of its hash value calculation,
now ANALYZE reads tuples directly from pg_class table instead of RELOID system cache during initdb as a workaround for this stale cache issue.

A YSQL upgrade script is not provided for this initdb change because we don't know the state (number of tables, etc.) of existing clusters.
Thus, TestYsqlUpgrade is changed accordingly to skip checking the state of `pg_statistic`.

In addition, a few regression tests are updated because of the initial catalog statistic change.

Jira: DB-5026

Test Plan:
After `./yb_build.sh reinitdb`, manually verified catalog stats in `pg_class` and `pg_statistic` is collected.

./yb_build.sh --java-test 'org.yb.pgsql.TestPgRegressYbStat#testYbStat'
./yb_build.sh --java-test 'org.yb.pgsql.TestPgRegressYbBitmapScans#testPgRegressYbBitmapScans'
./yb_build.sh --java-test 'org.yb.pgsql.TestPgRegressPushdownKey#testPgRegressPushdownKey'
./yb_build.sh --java-test 'org.yb.pgsql.TestPgRegressPgStat#testPgStat'
./yb_build.sh --java-test 'org.yb.pgsql.TestPgRegressMisc#testPgRegressMiscSerial3'
./yb_build.sh --java-test 'org.yb.pgsql.TestPgRegressJson#testPgRegressJson'
./yb_build.sh --cxx-test pgwrapper_pg_catalog_version-test --gtest_filter PgCatalogVersionTest.FixCatalogVersionTable
./yb_build.sh --java-test 'org.yb.pgsql.TestPgRegressTablegroup#testPgRegressTablegroup'
./yb_build.sh --java-test 'org.yb.pgsql.TestPgRegressIndex#testPgRegressIndex'
./yb_build.sh --java-test 'org.yb.pgsql.TestYsqlDump'
./yb_build.sh --java-test 'org.yb.pgsql.TestYsqlUpgrade#migratingIsEquivalentToReinitdb'
./yb_build.sh --java-test 'org.yb.pgsql.TestYsqlUpgrade#upgradeIsIdempotentSingleConn'
./yb_build.sh --java-test 'org.yb.pgsql.TestYsqlUpgrade#upgradeIsIdempotent'

Reviewers: mihnea, kfranz, telgersma, amartsinchyk

Reviewed By: kfranz, amartsinchyk

Subscribers: jason, yql

Tags: #jenkins-ready

Differential Revision: https://phorge.dev.yugabyte.com/D34576

java/yb-client/src/test/java/org/yb/util/SideBySideDiff.java

@@ -33,7 +33,7 @@ public static String generate(File f1, File f2) throws IOException {
     // you're in a terminal with 8-column tab stops, this causes the RHS to begin at a tab stop
     // which will make the LHS and the RHS identical when the input files have embedded tabs.
     // Embedded tabs aren't expanded even with the --expand-tabs flag.
-    String diffCmd = String.format("sdiff --width=157 --expand-tabs '%s' '%s'", f1, f2);
+    String diffCmd = String.format("sdiff -Z --width=157 --expand-tabs '%s' '%s'", f1, f2);
     CommandResult commandResult = CommandUtil.runShellCommand(diffCmd);
     List<String> stdoutLines = commandResult.getStdoutLines();
     return StringUtil.joinLinesForLoggingNoPrefix(stdoutLines);

java/yb-pgsql/src/test/java/org/yb/pgsql/TestYsqlDump.java

@@ -18,7 +18,6 @@
 import java.io.File;
 import java.io.IOException;
 import java.nio.charset.StandardCharsets;
-import java.sql.Statement;
 import java.util.ArrayList;
 import java.util.Arrays;
 import java.util.Collections;

java/yb-pgsql/src/test/java/org/yb/pgsql/TestYsqlUpgrade.java

@@ -1496,6 +1496,13 @@ private SysCatalogSnapshot takeSysCatalogSnapshot(Statement stmt) throws Excepti
 
     for (Row tableInfoRow : tablesInfo) {
       String tableName = tableInfoRow.getString(0);
+      // Different runs of ANALYZE on catalog tables can result in different statistics,
+      // and we don't know the state of existing clusters,
+      // so we don't provide YSQL migration scripts for catalog statistics.
+      // Thus, exclude capturing pg_statistic catalog from "snapshot".
+      if (tableName.equals("pg_statistic")) {
+        continue;
+      }
       String query;
       // Filter out stuff created for shared entities.
       switch (tableName) {

src/postgres/src/backend/commands/vacuum.c

@@ -52,6 +52,8 @@
 #include "utils/syscache.h"
 #include "utils/tqual.h"
 
+/* YB includes. */
+#include "access/sysattr.h"
 #include "pg_yb_utils.h"
 
 /*
@@ -894,8 +896,31 @@ vac_update_relstats(Relation relation,
 
 	rd = heap_open(RelationRelationId, RowExclusiveLock);
 
-	/* Fetch a copy of the tuple to scribble on */
-	ctup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relid));
+	if (IsYugaByteEnabled() && YBIsInitDbModeEnvVarSet())
+	{
+		/*
+		 * YB: workaround for stale cache issue #13500 during initdb.
+		 * Instead of fetching a tuple from sys cache,
+		 * read the tuple from pg_class directly.
+		 */
+		HeapScanDesc pg_class_scan;
+		ScanKeyData key[1];
+
+		ScanKeyInit(&key[0],
+					ObjectIdAttributeNumber,
+					BTEqualStrategyNumber, F_OIDEQ,
+					ObjectIdGetDatum(relid));
+
+		pg_class_scan = heap_beginscan_catalog(rd, 1, key);
+		ctup = heap_getnext(pg_class_scan, ForwardScanDirection);
+		ctup = heap_copytuple(ctup);
+		heap_endscan(pg_class_scan);
+	}
+	else
+	{
+		/* Fetch a copy of the tuple to scribble on */
+		ctup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relid));
+	}
 	if (!HeapTupleIsValid(ctup))
 		elog(ERROR, "pg_class entry for relid %u vanished during vacuuming",
 			 relid);

src/postgres/src/bin/initdb/initdb.c

@@ -3234,11 +3234,18 @@ initialize_data_directory(void)
 	/* Enable pg_stat_statements */
 	enable_pg_stat_statements(cmdfd);
 
-	if (!IsYugaByteGlobalClusterInitdb())
-	{
-		/* Do not need to vacuum in YB */
-		vacuum_db(cmdfd);
-	}
+	/*
+	 * YB: we used to skip the call of vacuum_db() because we don't need
+	 * to vacuum in YB. As of 04/30/2024, we want to ANALYZE catalog tables
+	 * to get initial stats.
+	 * vacuum_db() runs both ANALYZE and VACUUM. Since VACUUM is a no-op in
+	 * YB. It is safe to just call vacuum_db().
+	 * All other databases: template0, postgres, yugabyte use template1 as the
+	 * template database. The initial stats of their catalog tables is copied
+	 * from template1 stats, so we don't need to do extra work to run ANALYZE
+	 * for each of them.
+	 */
+	vacuum_db(cmdfd);
 
 	make_template0(cmdfd);
 

src/postgres/src/test/regress/expected/yb_bitmap_scans_system.out

@@ -39,7 +39,7 @@ SELECT * FROM pg_authid WHERE rolname LIKE 'pg_%' OR rolname LIKE 'yb_%' ORDER B
  yb_fdw                    | f        | t          | f             | f           | f           | f              | f            |           -1 |             |
 (11 rows)
 
-/*+ BitmapScan(pg_roles) */ EXPLAIN (ANALYZE, COSTS OFF, SUMMARY OFF) SELECT spcname FROM pg_tablespace WHERE spcowner NOT IN (
+/*+ BitmapScan(pg_authid) */ EXPLAIN (ANALYZE, COSTS OFF, SUMMARY OFF) SELECT spcname FROM pg_tablespace WHERE spcowner NOT IN (
     SELECT oid FROM pg_roles WHERE rolname = 'postgres' OR rolname LIKE 'pg_%' OR rolname LIKE 'yb_%');
                                                               QUERY PLAN
 --------------------------------------------------------------------------------------------------------------------------------------
@@ -60,14 +60,14 @@ SELECT * FROM pg_authid WHERE rolname LIKE 'pg_%' OR rolname LIKE 'yb_%' ORDER B
                        Storage Index Filter: ((rolname = 'postgres'::name) OR (rolname ~~ 'pg_%'::text) OR (rolname ~~ 'yb_%'::text))
 (15 rows)
 
-/*+ BitmapScan(pg_roles) */ SELECT spcname FROM pg_tablespace WHERE spcowner NOT IN (
+/*+ BitmapScan(pg_authid) */ SELECT spcname FROM pg_tablespace WHERE spcowner NOT IN (
     SELECT oid FROM pg_roles WHERE rolname = 'postgres' OR rolname LIKE 'pg_%' OR rolname LIKE 'yb_%');
  spcname
 ---------
 (0 rows)
 
 SET yb_enable_expression_pushdown = false;
-/*+ BitmapScan(pg_roles) */ EXPLAIN (ANALYZE, COSTS OFF, SUMMARY OFF) SELECT spcname FROM pg_tablespace WHERE spcowner NOT IN (
+/*+ BitmapScan(pg_authid) */ EXPLAIN (ANALYZE, COSTS OFF, SUMMARY OFF) SELECT spcname FROM pg_tablespace WHERE spcowner NOT IN (
     SELECT oid FROM pg_roles WHERE rolname = 'postgres' OR rolname LIKE 'pg_%' OR rolname LIKE 'yb_%');
                                                  QUERY PLAN
 ------------------------------------------------------------------------------------------------------------
@@ -86,7 +86,7 @@ SET yb_enable_expression_pushdown = false;
                        Index Cond: ((rolname >= 'yb'::name) AND (rolname < 'yc'::name))
 (13 rows)
 
-/*+ BitmapScan(pg_roles) */ SELECT spcname FROM pg_tablespace WHERE spcowner NOT IN (
+/*+ BitmapScan(pg_authid) */ SELECT spcname FROM pg_tablespace WHERE spcowner NOT IN (
     SELECT oid FROM pg_roles WHERE rolname = 'postgres' OR rolname LIKE 'pg_%' OR rolname LIKE 'yb_%');
  spcname
 ---------

src/postgres/src/test/regress/expected/yb_create_table_like.out

@@ -190,7 +190,7 @@ UNION
 SELECT 'index' AS object_type, relname AS obj_name, obj_description(oid) AS comments FROM pg_class WHERE oid IN (SELECT indexrelid FROM pg_index WHERE indrelid=$1::regclass::oid)
 UNION
 SELECT 'constraint' AS object_type, conname AS obj_name, obj_description(oid) AS comments FROM pg_constraint WHERE conrelid=$1::regclass::oid
-ORDER BY obj_name
+ORDER BY obj_name, object_type
 $$ LANGUAGE SQL;
 -- Without specifying INCLUDING COMMENTS, comments are not copied.
 CREATE TABLE comments1 (LIKE liketest1 INCLUDING INDEXES, LIKE liketest2 INCLUDING CONSTRAINTS);
@@ -202,8 +202,8 @@ SELECT * FROM get_comments('comments1');
  column      | c                  |
  index       | comments1_b_idx    |
  index       | comments1_expr_idx |
- index       | comments1_pkey     |
  constraint  | comments1_pkey     |
+ index       | comments1_pkey     |
  column      | d                  |
  constraint  | liketest2_c_check  |
 (9 rows)
@@ -232,8 +232,8 @@ SELECT * FROM get_comments('comments3');
  column      | b                  | B
  index       | comments3_b_idx    | index b
  index       | comments3_expr_idx |
- index       | comments3_pkey     | index pkey
  constraint  | comments3_pkey     |
+ index       | comments3_pkey     | index pkey
  constraint  | liketest1_a_check  | a_check
 (7 rows)
 

src/postgres/src/test/regress/expected/yb_dml_systable_scan.out

@@ -121,40 +121,40 @@ return null;
 END;
 $$ LANGUAGE plpgsql STABLE;
 -- Expect pushdown in all cases.
-EXPLAIN SELECT * FROM pg_database WHERE datname = test_null_pushdown();
+EXPLAIN /*+IndexScan(pg_database)*/ SELECT * FROM pg_database WHERE datname = test_null_pushdown();
                                           QUERY PLAN
 -----------------------------------------------------------------------------------------------
  Index Scan using pg_database_datname_index on pg_database  (cost=0.00..4.37 rows=1 width=254)
    Index Cond: (datname = test_null_pushdown())
 (2 rows)
 
-EXPLAIN SELECT * FROM pg_database WHERE datname IN (test_null_pushdown());
+EXPLAIN /*+IndexScan(pg_database)*/ SELECT * FROM pg_database WHERE datname IN (test_null_pushdown());
                                           QUERY PLAN
 -----------------------------------------------------------------------------------------------
  Index Scan using pg_database_datname_index on pg_database  (cost=0.00..4.37 rows=1 width=254)
    Index Cond: (datname = test_null_pushdown())
 (2 rows)
 
-EXPLAIN SELECT * FROM pg_database WHERE datname IN ('template1', test_null_pushdown(), 'template0');
+EXPLAIN /*+IndexScan(pg_database)*/ SELECT * FROM pg_database WHERE datname IN ('template1', test_null_pushdown(), 'template0');
                                             QUERY PLAN
 ---------------------------------------------------------------------------------------------------
  Index Scan using pg_database_datname_index on pg_database  (cost=0.00..4.37 rows=1 width=254)
    Index Cond: (datname = ANY (ARRAY['template1'::name, test_null_pushdown(), 'template0'::name]))
 (2 rows)
 
 -- Test execution.
-SELECT * FROM pg_database WHERE datname = test_null_pushdown();
+/*+IndexScan(pg_database)*/ SELECT * FROM pg_database WHERE datname = test_null_pushdown();
  datname | datdba | encoding | datcollate | datctype | datistemplate | datallowconn | datconnlimit | datlastsysoid | datfrozenxid | datminmxid | dattablespace | datacl
 ---------+--------+----------+------------+----------+---------------+--------------+--------------+---------------+--------------+------------+---------------+--------
 (0 rows)
 
-SELECT * FROM pg_database WHERE datname IN (test_null_pushdown());
+/*+IndexScan(pg_database)*/ SELECT * FROM pg_database WHERE datname IN (test_null_pushdown());
  datname | datdba | encoding | datcollate | datctype | datistemplate | datallowconn | datconnlimit | datlastsysoid | datfrozenxid | datminmxid | dattablespace | datacl
 ---------+--------+----------+------------+----------+---------------+--------------+--------------+---------------+--------------+------------+---------------+--------
 (0 rows)
 
 -- Test null mixed with valid (existing) options.
-SELECT * FROM pg_database WHERE datname IN ('template1', test_null_pushdown(), 'template0');
+/*+IndexScan(pg_database)*/ SELECT * FROM pg_database WHERE datname IN ('template1', test_null_pushdown(), 'template0');
   datname  | datdba | encoding | datcollate |  datctype   | datistemplate | datallowconn | datconnlimit | datlastsysoid | datfrozenxid | datminmxid | dattablespace |               datacl
 -----------+--------+----------+------------+-------------+---------------+--------------+--------------+---------------+--------------+------------+---------------+-------------------------------------
  template0 |     10 |        6 | C          | en_US.UTF-8 | t             | f            |           -1 |             0 |            0 |          1 |          1663 | {=c/postgres,postgres=CTc/postgres}

src/postgres/src/test/regress/expected/yb_pg_json.out

@@ -387,9 +387,10 @@ SELECT row_to_json(row((select array_agg(x) as d from generate_series(5,10) x)),
 select to_json(histogram_bounds) histogram_bounds
 from pg_stats
 where attname = 'tmplname' and tablename = 'pg_pltemplate';
- histogram_bounds
-------------------
-(0 rows)
+                                   histogram_bounds
+---------------------------------------------------------------------------------------
+ ["plperl","plperlu","plpgsql","plpython2u","plpython3u","plpythonu","pltcl","pltclu"]
+(1 row)
 
 -- to_json, timestamps
 select to_json(timestamp '2014-05-28 12:22:35.614298');

src/postgres/src/test/regress/expected/yb_pg_stat_backend.out

@@ -8,12 +8,12 @@
 -- this connection, and the java test.
 SELECT datname, usename, state, query, backend_type,
        catalog_version IS NOT null AS has_catalog_snapshot
-    FROM pg_stat_activity;
+    FROM pg_stat_activity ORDER BY usename;
  datname  |    usename    | state  |                           query                            |  backend_type  | has_catalog_snapshot
 ----------+---------------+--------+------------------------------------------------------------+----------------+----------------------
  yugabyte | yugabyte      | active | SELECT datname, usename, state, query, backend_type,      +| client backend | t
           |               |        |        catalog_version IS NOT null AS has_catalog_snapshot+|                |
-          |               |        |     FROM pg_stat_activity;                                 |                |
+          |               |        |     FROM pg_stat_activity ORDER BY usename;                |                |
  yugabyte | yugabyte_test | idle   | COMMIT                                                     | client backend | f
           |               |        |                                                            | checkpointer   | f
 (3 rows)

src/postgres/src/test/regress/expected/yb_reindex.out

@@ -485,7 +485,7 @@ INFO:  index "yb_j_idx" was reindexed
 REINDEX INDEX pg_database_datname_index; -- fail
 ERROR:  cannot reindex shared system indexes
 -- make sure index isn't broken after failure
-/*+IndexOnlyScan(pg_database_datname_index)*/
+/*+IndexOnlyScan(pg_database pg_database_datname_index)*/
 SELECT datname from pg_database WHERE datname LIKE 'template%';
   datname
 -----------

src/postgres/src/test/regress/expected/yb_stat.out

@@ -211,17 +211,17 @@ show temp_file_limit;
 
 SELECT * FROM generate_series(0, 1234567);
 ERROR:  temporary file size exceeds temp_file_limit (0kB)
-SELECT databasename, termination_reason, query_text FROM yb_terminated_queries;
+SELECT databasename, termination_reason, query_text FROM yb_terminated_queries ORDER BY databasename;
     databasename    |                   termination_reason                    |                                                         query_text
 --------------------+---------------------------------------------------------+----------------------------------------------------------------------------------------------------------------------------
- yugabyte           | temporary file size exceeds temp_file_limit (0kB)       | SELECT * FROM generate_series(0, 1000000);
- yugabyte           | temporary file size exceeds temp_file_limit (0kB)       | SELECT 'bob' FROM generate_series(0, 1000000);
- yugabyte           | temporary file size exceeds temp_file_limit (1048576kB) | SELECT * FROM generate_series(0, 100000002);
  db2                | temporary file size exceeds temp_file_limit (0kB)       | SELECT * FROM generate_series(0, 1000001);
  db2                | temporary file size exceeds temp_file_limit (0kB)       | SELECT 'We were taught in this modern age that science and mathematics is the pinnacle of human achievement.'             +
                     |                                                         | 'Yet, in our complacency, we began to neglect the very thing which our ancestors had once done: to challenge the process.'+
                     |                                                         | 'We need to stand back
  test_user_database | temporary file size exceeds temp_file_limit (0kB)       | SELECT * FROM generate_series(0, 1234567);
+ yugabyte           | temporary file size exceeds temp_file_limit (0kB)       | SELECT * FROM generate_series(0, 1000000);
+ yugabyte           | temporary file size exceeds temp_file_limit (0kB)       | SELECT 'bob' FROM generate_series(0, 1000000);
+ yugabyte           | temporary file size exceeds temp_file_limit (1048576kB) | SELECT * FROM generate_series(0, 100000002);
 (6 rows)
 
 -- Drop the superuser privilege as we want to see if we would only see the terminated query

src/postgres/src/test/regress/expected/yb_tablegroup.out

@@ -409,13 +409,13 @@ CREATE TABLE tgroup_test6 (col1 int, col2 int) TABLEGROUP grp3;
     Group Name    | Group Owner | Access privileges | Group Description | Group Tablespace | Group Options |           Name           | Type  |  Owner   | Rel Description | Size
 ------------------+-------------+-------------------+-------------------+------------------+---------------+--------------------------+-------+----------+-----------------+------
  grp3             | yugabyte    |                   |                   | tblspc           |               | tgroup_test6             | table | yugabyte |                 |
- tgroup2          | yugabyte    |                   |                   |                  |               | tgroup_test3             | table | yugabyte |                 |
  tgroup2          | yugabyte    |                   |                   |                  |               | tgroup_test3_col1_idx    | index | yugabyte |                 |
  tgroup2          | yugabyte    |                   |                   |                  |               | tgroup_test4             | table | yugabyte |                 |
  tgroup2          | yugabyte    |                   |                   |                  |               | tgroup_test4_col1_idx    | index | yugabyte |                 |
+ tgroup2          | yugabyte    |                   |                   |                  |               | tgroup_test3             | table | yugabyte |                 |
  tgroup2          | yugabyte    |                   |                   |                  |               | tgroup_test4_col2_idx    | index | yugabyte |                 |
- tgroup_describe1 | yugabyte    |                   |                   |                  |               | tgroup_describe          | table | yugabyte |                 |
  tgroup_describe1 | yugabyte    |                   |                   |                  |               | tgroup_describe_col1_idx | index | yugabyte |                 |
+ tgroup_describe1 | yugabyte    |                   |                   |                  |               | tgroup_describe          | table | yugabyte |                 |
 (8 rows)
 
 --