Avoid deleting critical WAL segments during pg_rewind

Previously, in unlucky cases, it was possible for pg_rewind to remove
certain WAL segments from the rewound demoted primary.  In particular
this happens if those files have been marked for archival (i.e., their
.ready files were created) but not yet archived; the newly promoted node
no longer has such files because of them having been recycled, but they
are likely critical for recovery in the demoted node.  If pg_rewind
removes them, recovery is not possible anymore.

Fix this by maintaining a hash table of files in this situation in the
scan that looks for a checkpoint, which the decide_file_actions phase
can consult so that it knows to preserve them.

Backpatch to 14.  The problem also exists in 13, but that branch was not
blessed with commit eb00f1d4bf, so this patch is difficult to apply
there.  Users of older releases will just have to continue to be extra
careful when rewinding.

Co-authored-by: Полина Бунгина (Polina Bungina) <bungina@gmail.com>
Co-authored-by: Alexander Kukushkin <cyberdemn@gmail.com>
Reviewed-by: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Reviewed-by: Atsushi Torikoshi <torikoshia@oss.nttdata.com>
Discussion: https://postgr.es/m/CAAtGL4AhzmBRsEsaDdz7065T+k+BscNadfTqP1NcPmsqwA5HBw@mail.gmail.com

src/bin/pg_rewind/filemap.c

@@ -39,14 +39,14 @@
  * appearing in source and target systems.
  */
 static uint32 hash_string_pointer(const char *s);
-#define SH_PREFIX		filehash
-#define SH_ELEMENT_TYPE	file_entry_t
-#define SH_KEY_TYPE		const char *
-#define	SH_KEY			path
+#define SH_PREFIX				filehash
+#define SH_ELEMENT_TYPE			file_entry_t
+#define SH_KEY_TYPE				const char *
+#define SH_KEY					path
 #define SH_HASH_KEY(tb, key)	hash_string_pointer(key)
 #define SH_EQUAL(tb, a, b)		(strcmp(a, b) == 0)
-#define	SH_SCOPE		static inline
-#define SH_RAW_ALLOCATOR	pg_malloc0
+#define SH_SCOPE				static inline
+#define SH_RAW_ALLOCATOR		pg_malloc0
 #define SH_DECLARE
 #define SH_DEFINE
 #include "lib/simplehash.h"
@@ -61,7 +61,36 @@ static char *datasegpath(RelFileNode rnode, ForkNumber forknum,
 
 static file_entry_t *insert_filehash_entry(const char *path);
 static file_entry_t *lookup_filehash_entry(const char *path);
+
+/*
+ * A separate hash table which tracks WAL files that must not be deleted.
+ */
+typedef struct keepwal_entry
+{
+	const char *path;
+	uint32		status;
+} keepwal_entry;
+
+#define SH_PREFIX				keepwal
+#define SH_ELEMENT_TYPE			keepwal_entry
+#define SH_KEY_TYPE				const char *
+#define SH_KEY					path
+#define SH_HASH_KEY(tb, key)	hash_string_pointer(key)
+#define SH_EQUAL(tb, a, b)		(strcmp(a, b) == 0)
+#define SH_SCOPE				static inline
+#define SH_RAW_ALLOCATOR		pg_malloc0
+#define SH_DECLARE
+#define SH_DEFINE
+#include "lib/simplehash.h"
+
+#define KEEPWAL_INITIAL_SIZE	1000
+
+
+static keepwal_hash *keepwal = NULL;
+static bool keepwal_entry_exists(const char *path);
+
 static int	final_filemap_cmp(const void *a, const void *b);
+
 static bool check_file_excluded(const char *path, bool is_source);
 
 /*
@@ -207,6 +236,39 @@ lookup_filehash_entry(const char *path)
 	return filehash_lookup(filehash, path);
 }
 
+/*
+ * Initialize a hash table to store WAL file names that must be kept.
+ */
+void
+keepwal_init(void)
+{
+	/* An initial hash size out of thin air */
+	keepwal = keepwal_create(KEEPWAL_INITIAL_SIZE, NULL);
+}
+
+/* Mark the given file to prevent its removal */
+void
+keepwal_add_entry(const char *path)
+{
+	keepwal_entry *entry;
+	bool		found;
+
+	/* Should only be called with keepwal initialized */
+	Assert(keepwal != NULL);
+
+	entry = keepwal_insert(keepwal, path, &found);
+
+	if (!found)
+		entry->path = pg_strdup(path);
+}
+
+/* Return true if file is marked as not to be removed, false otherwise */
+static bool
+keepwal_entry_exists(const char *path)
+{
+	return keepwal_lookup(keepwal, path) != NULL;
+}
+
 /*
  * Callback for processing source file list.
  *
@@ -686,7 +748,15 @@ decide_file_action(file_entry_t *entry)
 	}
 	else if (entry->target_exists && !entry->source_exists)
 	{
-		/* File exists in target, but not source. Remove it. */
+		/*
+		 * For files that exist in target but not in source, we check the
+		 * keepwal hash table; any files listed therein must not be removed.
+		 */
+		if (keepwal_entry_exists(path))
+		{
+			pg_log_debug("Not removing file \"%s\" because it is required for recovery", path);
+			return FILE_ACTION_NONE;
+		}
 		return FILE_ACTION_REMOVE;
 	}
 	else if (!entry->target_exists && !entry->source_exists)

src/bin/pg_rewind/filemap.h

@@ -110,4 +110,7 @@ extern filemap_t *decide_file_actions(void);
 extern void calculate_totals(filemap_t *filemap);
 extern void print_filemap(filemap_t *filemap);
 
+extern void keepwal_init(void);
+extern void keepwal_add_entry(const char *path);
+
 #endif							/* FILEMAP_H */

src/bin/pg_rewind/parsexlog.c

@@ -175,6 +175,8 @@ findLastCheckpoint(const char *datadir, XLogRecPtr forkptr, int tliIndex,
 	XLogReaderState *xlogreader;
 	char	   *errormsg;
 	XLogPageReadPrivate private;
+	XLogSegNo	current_segno = 0;
+	TimeLineID	current_tli = 0;
 
 	/*
 	 * The given fork pointer points to the end of the last common record,
@@ -217,6 +219,25 @@ findLastCheckpoint(const char *datadir, XLogRecPtr forkptr, int tliIndex,
 						 LSN_FORMAT_ARGS(searchptr));
 		}
 
+		/* Detect if a new WAL file has been opened */
+		if (xlogreader->seg.ws_tli != current_tli ||
+			xlogreader->seg.ws_segno != current_segno)
+		{
+			char		xlogfname[MAXFNAMELEN];
+
+			snprintf(xlogfname, MAXFNAMELEN, XLOGDIR "/");
+
+			/* update curent values */
+			current_tli = xlogreader->seg.ws_tli;
+			current_segno = xlogreader->seg.ws_segno;
+
+			XLogFileName(xlogfname + sizeof(XLOGDIR),
+						 current_tli, current_segno, WalSegSz);
+
+			/* Track this filename as one to not remove */
+			keepwal_add_entry(xlogfname);
+		}
+
 		/*
 		 * Check if it is a checkpoint record. This checkpoint record needs to
 		 * be the latest checkpoint before WAL forked and not the checkpoint

src/bin/pg_rewind/pg_rewind.c

@@ -405,6 +405,9 @@ main(int argc, char **argv)
 		exit(0);
 	}
 
+	/* Initialize hashtable that tracks WAL files protected from removal */
+	keepwal_init();
+
 	findLastCheckpoint(datadir_target, divergerec, lastcommontliIndex,
 					   &chkptrec, &chkpttli, &chkptredo, restore_command);
 	pg_log_info("rewinding from last common checkpoint at %X/%X on timeline %u",

src/bin/pg_rewind/t/010_keep_recycled_wals.pl

@@ -0,0 +1,62 @@
+# Copyright (c) 2021-2024, PostgreSQL Global Development Group
+#
+# Test situation where a target data directory contains
+# WAL files that were already recycled by the new primary.
+#
+
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+use FindBin;
+use lib $FindBin::RealBin;
+use RewindTest;
+
+RewindTest::setup_cluster();
+$node_primary->enable_archiving();
+RewindTest::start_primary();
+
+RewindTest::create_standby();
+$node_standby->enable_restoring($node_primary, 0);
+$node_standby->reload();
+
+RewindTest::primary_psql("CHECKPOINT");    # last common checkpoint
+
+# We use "perl -e 'exit(1)'" as an alternative to "false", because the latter
+# might not be available on Windows.
+my $false = "$^X -e 'exit(1)'";
+$node_primary->append_conf(
+	'postgresql.conf', qq(
+archive_command = '$false'
+));
+$node_primary->reload();
+
+# advance WAL on primary; this WAL segment will never make it to the archive
+RewindTest::primary_psql("CREATE TABLE t(a int)");
+RewindTest::primary_psql("INSERT INTO t VALUES(0)");
+RewindTest::primary_psql("SELECT pg_switch_wal()");
+
+RewindTest::promote_standby;
+
+# new primary loses diverging WAL segment
+RewindTest::standby_psql("INSERT INTO t values(0)");
+RewindTest::standby_psql("SELECT pg_switch_wal()");
+
+$node_standby->stop();
+$node_primary->stop();
+
+my ($stdout, $stderr) = run_command(
+	[
+		'pg_rewind', '--debug',
+		'--source-pgdata', $node_standby->data_dir,
+		'--target-pgdata', $node_primary->data_dir,
+		'--no-sync',
+	]);
+
+like(
+	$stderr,
+	qr/Not removing file .* because it is required for recovery/,
+	"some WAL files were skipped");
+
+done_testing();

src/tools/pgindent/typedefs.list

@@ -3368,6 +3368,8 @@ json_manifest_perwalrange_callback
 json_ofield_action
 json_scalar_action
 json_struct_action
+keepwal_entry
+keepwal_hash
 keyEntryData
 key_t
 lclContext