Merge branch 'master' into stable

Author: za-arthur

src/restore.c

@@ -14,7 +14,6 @@
 #include <sys/stat.h>
 #include <sys/types.h>
 #include <unistd.h>
-#include <pthread.h>
 
 #include "catalog/pg_control.h"
 #include "utils/logger.h"

tests/false_positive.py

@@ -13,7 +13,9 @@ class FalsePositive(ProbackupTest, unittest.TestCase):
     # @unittest.skip("skip")
     @unittest.expectedFailure
     def test_validate_wal_lost_segment(self):
-        """Loose segment located between backups. ExpectedFailure. This is BUG """
+        """
+        Loose segment located between backups. ExpectedFailure. This is BUG
+        """
         fname = self.id().split('.')[3]
         node = self.make_simple_node(
             base_dir="{0}/{1}/node".format(module_name, fname),
@@ -31,14 +33,7 @@ def test_validate_wal_lost_segment(self):
         self.backup_node(backup_dir, 'node', node)
 
         # make some wals
-        node.pgbench_init(scale=2)
-        pgbench = node.pgbench(
-            stdout=subprocess.PIPE,
-            stderr=subprocess.STDOUT,
-            options=["-c", "4", "-T", "10"]
-        )
-        pgbench.wait()
-        pgbench.stdout.close()
+        node.pgbench_init(scale=5)
 
         # delete last wal segment
         wals_dir = os.path.join(backup_dir, "wal", 'node')

tests/helpers/ptrack_helpers.py

@@ -614,15 +614,16 @@ def backup_node(
 
         return self.run_pb(cmd_list + options, async, gdb)
 
-    def merge_backup(self, backup_dir, instance, backup_id):
+    def merge_backup(
+            self, backup_dir, instance, backup_id, async=False, gdb=False):
         cmd_list = [
             "merge",
             "-B", backup_dir,
             "--instance={0}".format(instance),
             "-i", backup_id
         ]
 
-        return self.run_pb(cmd_list)
+        return self.run_pb(cmd_list, async, gdb)
 
     def restore_node(
             self, backup_dir, instance, node=False,

tests/merge.py

@@ -458,7 +458,8 @@ def test_merge_ptrack_truncate(self):
     def test_merge_delta_delete(self):
         """
         Make node, create tablespace with table, take full backup,
-        alter tablespace location, take delta backup, restore database.
+        alter tablespace location, take delta backup, merge full and delta,
+        restore database.
         """
         fname = self.id().split('.')[3]
         backup_dir = os.path.join(self.tmp_path, module_name, fname, 'backup')
@@ -542,3 +543,74 @@ def test_merge_delta_delete(self):
 
         # Clean after yourself
         self.del_test_dir(module_name, fname)
+
+    # @unittest.skip("skip")
+    def test_continue_failed_merge(self):
+        """
+        Check that failed MERGE can be continued
+        """
+        fname = self.id().split('.')[3]
+        backup_dir = os.path.join(self.tmp_path, module_name, fname, 'backup')
+        node = self.make_simple_node(
+            base_dir="{0}/{1}/node".format(module_name, fname),
+            set_replication=True, initdb_params=['--data-checksums'],
+            pg_options={
+                'wal_level': 'replica'
+            }
+        )
+
+        self.init_pb(backup_dir)
+        self.add_instance(backup_dir, 'node', node)
+        self.set_archiving(backup_dir, 'node', node)
+        node.start()
+
+        # FULL backup
+        self.backup_node(backup_dir, 'node', node)
+
+        node.safe_psql(
+            "postgres",
+            "create table t_heap as select i as id,"
+            " md5(i::text) as text, md5(i::text)::tsvector as tsvector"
+            " from generate_series(0,1000) i"
+        )
+
+        # DELTA BACKUP
+        self.backup_node(
+            backup_dir, 'node', node, backup_type='delta'
+        )
+
+        node.safe_psql(
+            "postgres",
+            "delete from t_heap"
+        )
+
+        node.safe_psql(
+            "postgres",
+            "vacuum t_heap"
+        )
+
+        # DELTA BACKUP
+        self.backup_node(
+            backup_dir, 'node', node, backup_type='delta'
+        )
+
+        if self.paranoia:
+            pgdata = self.pgdata_content(node.data_dir)
+
+        backup_id = self.show_pb(backup_dir, "node")[2]["id"]
+
+        gdb = self.merge_backup(backup_dir, "node", backup_id, gdb=True)
+
+        gdb.set_breakpoint('move_file')
+        gdb.run_until_break()
+
+        if gdb.continue_execution_until_break(20) != 'breakpoint-hit':
+            print('Failed to hit breakpoint')
+            exit(1)
+
+        gdb._execute('signal SIGKILL')
+
+        print(self.show_pb(backup_dir, as_text=True, as_json=False))
+
+        # Try to continue failed MERGE
+        self.merge_backup(backup_dir, "node", backup_id)

tests/page.py

@@ -166,11 +166,6 @@ def test_page_stream(self):
             '\n Unexpected Error Message: {0}\n'
             ' CMD: {1}'.format(repr(self.output), self.cmd))
 
-        # GET RESTORED PGDATA AND COMPARE
-        if self.paranoia:
-            pgdata_restored = self.pgdata_content(node.data_dir)
-            self.compare_pgdata(pgdata, pgdata_restored)
-
         node.slow_start()
         full_result_new = node.execute("postgres", "SELECT * FROM t_heap")
         self.assertEqual(full_result, full_result_new)
@@ -184,6 +179,12 @@ def test_page_stream(self):
                 backup_id=page_backup_id, options=["-j", "4"]),
             '\n Unexpected Error Message: {0}\n'
             ' CMD: {1}'.format(repr(self.output), self.cmd))
+
+        # GET RESTORED PGDATA AND COMPARE
+        if self.paranoia:
+            pgdata_restored = self.pgdata_content(node.data_dir)
+            self.compare_pgdata(pgdata, pgdata_restored)
+
         node.slow_start()
         page_result_new = node.execute("postgres", "SELECT * FROM t_heap")
         self.assertEqual(page_result, page_result_new)
@@ -659,3 +660,92 @@ def test_parallel_pagemap_1(self):
         # Clean after yourself
         node.cleanup()
         self.del_test_dir(module_name, fname)
+
+    # @unittest.skip("skip")
+    def test_page_backup_with_lost_wal_segment(self):
+        """
+        make node with archiving
+        make archive backup, then generate some wals with pgbench,
+        delete latest archived wal segment
+        run page backup, expecting error because of missing wal segment
+        make sure that backup status is 'ERROR'
+        """
+        fname = self.id().split('.')[3]
+        node = self.make_simple_node(
+            base_dir="{0}/{1}/node".format(module_name, fname),
+            initdb_params=['--data-checksums'],
+            pg_options={'wal_level': 'replica'}
+            )
+        backup_dir = os.path.join(self.tmp_path, module_name, fname, 'backup')
+        self.init_pb(backup_dir)
+        self.add_instance(backup_dir, 'node', node)
+        self.set_archiving(backup_dir, 'node', node)
+        node.start()
+
+        self.backup_node(backup_dir, 'node', node)
+
+        # make some wals
+        node.pgbench_init(scale=3)
+
+        # delete last wal segment
+        wals_dir = os.path.join(backup_dir, 'wal', 'node')
+        wals = [f for f in os.listdir(wals_dir) if os.path.isfile(os.path.join(
+            wals_dir, f)) and not f.endswith('.backup')]
+        wals = map(str, wals)
+        file = os.path.join(wals_dir, max(wals))
+        os.remove(file)
+        if self.archive_compress:
+            file = file[:-3]
+
+        # Single-thread PAGE backup
+        try:
+            self.backup_node(
+                backup_dir, 'node', node,
+                backup_type='page')
+            self.assertEqual(
+                1, 0,
+                "Expecting Error because of wal segment disappearance.\n "
+                "Output: {0} \n CMD: {1}".format(
+                    self.output, self.cmd))
+        except ProbackupException as e:
+            self.assertTrue(
+                'INFO: Wait for LSN' in e.message and
+                'in archived WAL segment' in e.message and
+                'WARNING: could not read WAL record at' in e.message and
+                'ERROR: WAL segment "{0}" is absent\n'.format(
+                    file) in e.message,
+                '\n Unexpected Error Message: {0}\n CMD: {1}'.format(
+                    repr(e.message), self.cmd))
+
+        self.assertEqual(
+            'ERROR',
+            self.show_pb(backup_dir, 'node')[1]['status'],
+            'Backup {0} should have STATUS "ERROR"')
+
+        # Multi-thread PAGE backup
+        try:
+            self.backup_node(
+                backup_dir, 'node', node,
+                backup_type='page', options=["-j", "4"])
+            self.assertEqual(
+                1, 0,
+                "Expecting Error because of wal segment disappearance.\n "
+                "Output: {0} \n CMD: {1}".format(
+                    self.output, self.cmd))
+        except ProbackupException as e:
+            self.assertTrue(
+                'INFO: Wait for LSN' in e.message and
+                'in archived WAL segment' in e.message and
+                'WARNING: could not read WAL record at' in e.message and
+                'ERROR: WAL segment "{0}" is absent\n'.format(
+                    file) in e.message,
+                '\n Unexpected Error Message: {0}\n CMD: {1}'.format(
+                    repr(e.message), self.cmd))
+
+        self.assertEqual(
+            'ERROR',
+            self.show_pb(backup_dir, 'node')[2]['status'],
+            'Backup {0} should have STATUS "ERROR"')
+
+        # Clean after yourself
+        self.del_test_dir(module_name, fname)

tests/validate_test.py

@@ -1237,78 +1237,12 @@ def test_validate_corrupt_wal_between_backups(self):
         self.del_test_dir(module_name, fname)
 
     # @unittest.skip("skip")
-    def test_validate_wal_lost_segment_2(self):
+    def test_pgpro702_688(self):
         """
-        make node with archiving
-        make archive backup
-        delete from archive wal segment which DO NOT belong to this backup
-        run validate, expecting error because of missing wal segment
-        make sure that backup status is 'ERROR'
+        make node without archiving, make stream backup,
+        get Recovery Time, validate to Recovery Time
         """
         fname = self.id().split('.')[3]
-        node = self.make_simple_node(base_dir="{0}/{1}/node".format(module_name, fname),
-            initdb_params=['--data-checksums'],
-            pg_options={'wal_level': 'replica'}
-            )
-        backup_dir = os.path.join(self.tmp_path, module_name, fname, 'backup')
-        self.init_pb(backup_dir)
-        self.add_instance(backup_dir, 'node', node)
-        self.set_archiving(backup_dir, 'node', node)
-        node.start()
-
-        self.backup_node(backup_dir, 'node', node)
-
-        # make some wals
-        node.pgbench_init(scale=2)
-        pgbench = node.pgbench(
-            stdout=subprocess.PIPE,
-            stderr=subprocess.STDOUT,
-            options=["-c", "4", "-T", "10"]
-        )
-        pgbench.wait()
-        pgbench.stdout.close()
-
-        # delete last wal segment
-        wals_dir = os.path.join(backup_dir, 'wal', 'node')
-        wals = [f for f in os.listdir(wals_dir) if os.path.isfile(os.path.join(
-            wals_dir, f)) and not f.endswith('.backup')]
-        wals = map(str, wals)
-        file = os.path.join(wals_dir, max(wals))
-        os.remove(file)
-        if self.archive_compress:
-            file = file[:-3]
-
-        # Try to restore
-        try:
-            backup_id = self.backup_node(
-                backup_dir, 'node', node, backup_type='page')
-            self.assertEqual(
-                1, 0,
-                "Expecting Error because of wal segment disappearance.\n "
-                "Output: {0} \n CMD: {1}".format(
-                    self.output, self.cmd))
-        except ProbackupException as e:
-            self.assertTrue(
-                'INFO: Wait for LSN' in e.message and
-                'in archived WAL segment' in e.message and
-                'WARNING: could not read WAL record at' in e.message and
-                'ERROR: WAL segment "{0}" is absent\n'.format(
-                    file) in e.message,
-                '\n Unexpected Error Message: {0}\n CMD: {1}'.format(
-                    repr(e.message), self.cmd))
-
-        self.assertEqual(
-            'ERROR',
-            self.show_pb(backup_dir, 'node')[1]['status'],
-            'Backup {0} should have STATUS "ERROR"')
-
-        # Clean after yourself
-        self.del_test_dir(module_name, fname)
-
-    # @unittest.skip("skip")
-    def test_pgpro702_688(self):
-        """make node without archiving, make stream backup, get Recovery Time, validate to Recovery Time"""
-        fname = self.id().split('.')[3]
         node = self.make_simple_node(
             base_dir="{0}/{1}/node".format(module_name, fname),
             set_replication=True,
@@ -1346,7 +1280,10 @@ def test_pgpro702_688(self):
 
     # @unittest.skip("skip")
     def test_pgpro688(self):
-        """make node with archiving, make backup, get Recovery Time, validate to Recovery Time. Waiting PGPRO-688. RESOLVED"""
+        """
+        make node with archiving, make backup, get Recovery Time,
+        validate to Recovery Time. Waiting PGPRO-688. RESOLVED
+        """
         fname = self.id().split('.')[3]
         node = self.make_simple_node(
             base_dir="{0}/{1}/node".format(module_name, fname),
@@ -1361,9 +1298,11 @@ def test_pgpro688(self):
         node.start()
 
         backup_id = self.backup_node(backup_dir, 'node', node)
-        recovery_time = self.show_pb(backup_dir, 'node', backup_id)['recovery-time']
+        recovery_time = self.show_pb(
+            backup_dir, 'node', backup_id)['recovery-time']
 
-        self.validate_pb(backup_dir, 'node', options=["--time={0}".format(recovery_time)])
+        self.validate_pb(
+            backup_dir, 'node', options=["--time={0}".format(recovery_time)])
 
         # Clean after yourself
         self.del_test_dir(module_name, fname)