Merge branch 'master' into issue_324

Author: gsmolk

doc/pgprobackup.xml

@@ -155,7 +155,7 @@ doc/src/sgml/pgprobackup.sgml
    recovery of <productname>PostgreSQL</productname> database clusters.
    It is designed to perform periodic backups of the <productname>PostgreSQL</productname> 
    instance that enable you to restore the server in case of a failure.
-   <application>pg_probackup</application> supports <productname>PostgreSQL</productname> 9.5 or higher.
+   <application>pg_probackup</application> supports PostgreSQL 9.5 or higher.
   </para>
 
 <itemizedlist spacing="compact">
@@ -389,7 +389,7 @@ doc/src/sgml/pgprobackup.sgml
     <itemizedlist spacing="compact">
       <listitem>
         <para>
-          <application>pg_probackup</application> only supports <productname>PostgreSQL</productname>  9.5 and higher.
+          <application>pg_probackup</application> only supports PostgreSQL 9.5 and higher.
         </para>
       </listitem>
       <listitem>
@@ -410,7 +410,7 @@ doc/src/sgml/pgprobackup.sgml
       </listitem>
       <listitem>
         <para>
-          For PostgreSQL 9.5, functions
+          For <productname>PostgreSQL</productname> 9.5, functions
           <function>pg_create_restore_point(text)</function> and
           <function>pg_switch_xlog()</function> can be executed only if
           the backup role is a superuser, so backup of a
@@ -599,7 +599,7 @@ pg_probackup add-instance -B <replaceable>backup_dir</replaceable> -D <replaceab
       connection</emphasis> to the <productname>PostgreSQL</productname> server:
     </para>
     <para>
-      For PostgreSQL 9.5:
+      For <productname>PostgreSQL</productname> 9.5:
     </para>
     <programlisting>
 BEGIN;
@@ -1202,8 +1202,9 @@ CREATE EXTENSION ptrack;
        together, which leads to false-positive results when tracking changed
        blocks and increases the incremental backup size as unchanged blocks
        can also be copied into the incremental backup.
-       Setting <varname>ptrack.map_size</varname> to a higher value
-       does not affect PTRACK operation. The maximum allowed value is 1024.
+       Setting <varname>ptrack.map_size</varname> to a higher value does not
+       affect PTRACK operation, but it is not recommended to set this parameter
+       to a value higher than 1024.
       </para>
      </listitem>
     </orderedlist>
@@ -1711,7 +1712,7 @@ pg_probackup restore -B <replaceable>backup_dir</replaceable> --instance <replac
       <para>
         The speed of restore from backup can be significantly improved
         by replacing only invalid and changed pages in already
-        existing PostgreSQL data directory using
+        existing <productname>PostgreSQL</productname> data directory using
         <link linkend="pbk-incremental-restore-options">incremental
         restore options</link> with the <xref linkend="pbk-restore"/>
         command.
@@ -1874,11 +1875,11 @@ pg_probackup restore -B <replaceable>backup_dir</replaceable> --instance <replac
       </note>
       <note>
         <para>
-          Due to recovery specifics of PostgreSQL versions earlier than 12,
+          Due to recovery specifics of <productname>PostgreSQL</productname> versions earlier than 12,
           it is advisable that you set the
           <ulink url="https://postgrespro.com/docs/postgresql/current/runtime-config-replication.html#GUC-HOT-STANDBY">hot_standby</ulink>
           parameter to <literal>off</literal> when running partial
-          restore of a PostgreSQL cluster of version earlier than 12.
+          restore of a <productname>PostgreSQL</productname> cluster of version earlier than 12.
           Otherwise the recovery may fail.
         </para>
       </note>

src/backup.c

@@ -56,7 +56,7 @@ static void pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn, PGNode
 
 static XLogRecPtr wait_wal_lsn(XLogRecPtr lsn, bool is_start_lsn, TimeLineID tli,
 								bool in_prev_segment, bool segment_only,
-								int timeout_elevel, bool in_stream_dir);
+								int timeout_elevel, bool in_stream_dir, pgBackup *backup);
 
 static void check_external_for_tablespaces(parray *external_list,
 										   PGconn *backup_conn);
@@ -165,8 +165,10 @@ do_backup_instance(PGconn *backup_conn, PGNodeInfo *nodeInfo, bool no_sync, bool
 						"trying to look up on previous timelines",
 						current.tli);
 
-			/* TODO: use read_timeline_history */
-			tli_list = catalog_get_timelines(&instance_config);
+			tli_list = get_history_streaming(&instance_config.conn_opt, current.tli, backup_list);
+			if (!tli_list)
+				/* fallback to using archive */
+				tli_list = catalog_get_timelines(&instance_config);
 
 			if (parray_num(tli_list) == 0)
 				elog(WARNING, "Cannot find valid backup on previous timelines, "
@@ -268,7 +270,7 @@ do_backup_instance(PGconn *backup_conn, PGNodeInfo *nodeInfo, bool no_sync, bool
 		 * Because WAL streaming will start after pg_start_backup() in stream
 		 * mode.
 		 */
-		wait_wal_lsn(current.start_lsn, true, current.tli, false, true, ERROR, false);
+		wait_wal_lsn(current.start_lsn, true, current.tli, false, true, ERROR, false, &current);
 	}
 
 	/* start stream replication */
@@ -279,6 +281,12 @@ do_backup_instance(PGconn *backup_conn, PGNodeInfo *nodeInfo, bool no_sync, bool
 
 		start_WAL_streaming(backup_conn, dst_backup_path, &instance_config.conn_opt,
 							current.start_lsn, current.tli);
+
+		/* Make sure that WAL streaming is working
+		 * PAGE backup in stream mode is waited twice, first for
+		 * segment in WAL archive and then for streamed segment
+		 */
+		wait_wal_lsn(current.start_lsn, true, current.tli, false, true, ERROR, true, &current);
 	}
 
 	/* initialize backup's file list */
@@ -1262,7 +1270,7 @@ pg_is_superuser(PGconn *conn)
 static XLogRecPtr
 wait_wal_lsn(XLogRecPtr target_lsn, bool is_start_lsn, TimeLineID tli,
 			 bool in_prev_segment, bool segment_only,
-			 int timeout_elevel, bool in_stream_dir)
+			 int timeout_elevel, bool in_stream_dir, pgBackup *backup)
 {
 	XLogSegNo	targetSegNo;
 	char		pg_wal_dir[MAXPGPATH];
@@ -1294,15 +1302,14 @@ wait_wal_lsn(XLogRecPtr target_lsn, bool is_start_lsn, TimeLineID tli,
 	 */
 	if (in_stream_dir)
 	{
-		pgBackupGetPath2(&current, pg_wal_dir, lengthof(pg_wal_dir),
-						 DATABASE_DIR, PG_XLOG_DIR);
+		join_path_components(pg_wal_dir, backup->database_dir, PG_XLOG_DIR);
 		join_path_components(wal_segment_path, pg_wal_dir, wal_segment);
 		wal_segment_dir = pg_wal_dir;
 	}
 	else
 	{
 		join_path_components(wal_segment_path, arclog_path, wal_segment);
-		wal_segment_dir = arclog_path;
+		wal_segment_dir = arclog_path; /* global var */
 	}
 
 	/* TODO: remove this in 3.0 (it is a cludge against some old bug with archive_timeout) */
@@ -1394,7 +1401,7 @@ wait_wal_lsn(XLogRecPtr target_lsn, bool is_start_lsn, TimeLineID tli,
 
 		sleep(1);
 		if (interrupted)
-			elog(ERROR, "Interrupted during waiting for WAL archiving");
+			elog(ERROR, "Interrupted during waiting for WAL %s", in_stream_dir ? "streaming" : "archiving");
 		try_count++;
 
 		/* Inform user if WAL segment is absent in first attempt */
@@ -1418,9 +1425,10 @@ wait_wal_lsn(XLogRecPtr target_lsn, bool is_start_lsn, TimeLineID tli,
 		{
 			if (file_exists)
 				elog(timeout_elevel, "WAL segment %s was %s, "
-					 "but target LSN %X/%X could not be archived in %d seconds",
+					 "but target LSN %X/%X could not be %s in %d seconds",
 					 wal_segment, wal_delivery_str,
-					 (uint32) (target_lsn >> 32), (uint32) target_lsn, timeout);
+					 (uint32) (target_lsn >> 32), (uint32) target_lsn,
+					 wal_delivery_str, timeout);
 			/* If WAL segment doesn't exist or we wait for previous segment */
 			else
 				elog(timeout_elevel,
@@ -1572,8 +1580,13 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
 	 */
 	if (pg_stop_backup_is_sent && !in_cleanup)
 	{
+		int timeout = ARCHIVE_TIMEOUT_DEFAULT;
 		res = NULL;
 
+		/* kludge against some old bug in archive_timeout. TODO: remove in 3.0.0 */
+		if (instance_config.archive_timeout > 0)
+			timeout = instance_config.archive_timeout;
+
 		while (1)
 		{
 			if (!PQconsumeInput(conn))
@@ -1598,11 +1611,10 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
 				 * If postgres haven't answered in archive_timeout seconds,
 				 * send an interrupt.
 				 */
-				if (pg_stop_backup_timeout > instance_config.archive_timeout)
+				if (pg_stop_backup_timeout > timeout)
 				{
 					pgut_cancel(conn);
-					elog(ERROR, "pg_stop_backup doesn't answer in %d seconds, cancel it",
-						 instance_config.archive_timeout);
+					elog(ERROR, "pg_stop_backup doesn't answer in %d seconds, cancel it", timeout);
 				}
 			}
 			else
@@ -1701,7 +1713,7 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
 			{
 				/* Wait for segment with current stop_lsn, it is ok for it to never arrive */
 				wait_wal_lsn(stop_backup_lsn_tmp, false, backup->tli,
-							 false, true, WARNING, stream_wal);
+							 false, true, WARNING, stream_wal, backup);
 
 				/* Get the first record in segment with current stop_lsn */
 				lsn_tmp = get_first_record_lsn(xlog_path, segno, backup->tli,
@@ -1729,7 +1741,7 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
 					 * because previous record can be the contrecord.
 					 */
 					lsn_tmp = wait_wal_lsn(stop_backup_lsn_tmp, false, backup->tli,
-											true, false, ERROR, stream_wal);
+											true, false, ERROR, stream_wal, backup);
 
 					/* sanity */
 					if (!XRecOffIsValid(lsn_tmp) || XLogRecPtrIsInvalid(lsn_tmp))
@@ -1743,7 +1755,7 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
 			{
 				/* Wait for segment with current stop_lsn */
 				wait_wal_lsn(stop_backup_lsn_tmp, false, backup->tli,
-							 false, true, ERROR, stream_wal);
+							 false, true, ERROR, stream_wal, backup);
 
 				/* Get the next closest record in segment with current stop_lsn */
 				lsn_tmp = get_next_record_lsn(xlog_path, segno, backup->tli,
@@ -1872,7 +1884,7 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
 		 */
 		if (!stop_lsn_exists)
 			stop_backup_lsn = wait_wal_lsn(stop_backup_lsn_tmp, false, backup->tli,
-											false, false, ERROR, stream_wal);
+											false, false, ERROR, stream_wal, backup);
 
 		if (stream_wal)
 		{

src/catalog.c

@@ -48,7 +48,7 @@ typedef struct LockInfo
 	bool exclusive;
 } LockInfo;
 
-static timelineInfo *
+timelineInfo *
 timelineInfoNew(TimeLineID tli)
 {
 	timelineInfo *tlinfo = (timelineInfo *) pgut_malloc(sizeof(timelineInfo));
@@ -74,7 +74,8 @@ timelineInfoFree(void *tliInfo)
 
 	if (tli->backups)
 	{
-		parray_walk(tli->backups, pgBackupFree);
+		/* backups themselves should freed separately  */
+//		parray_walk(tli->backups, pgBackupFree);
 		parray_free(tli->backups);
 	}
 
@@ -972,17 +973,11 @@ catalog_get_backup_list(const char *instance_name, time_t requested_backup_id)
 			continue;
 		}
 		parray_append(backups, backup);
-
-		if (errno && errno != ENOENT)
-		{
-			elog(WARNING, "cannot read data directory \"%s\": %s",
-				 data_ent->d_name, strerror(errno));
-			goto err_proc;
-		}
 	}
+
 	if (errno)
 	{
-		elog(WARNING, "cannot read backup root directory \"%s\": %s",
+		elog(WARNING, "Cannot read backup root directory \"%s\": %s",
 			backup_instance_path, strerror(errno));
 		goto err_proc;
 	}
@@ -2462,7 +2457,7 @@ write_backup_filelist(pgBackup *backup, parray *files, const char *root,
 		{
 			len += sprintf(line+len, ",\"n_headers\":\"%i\"", file->n_headers);
 			len += sprintf(line+len, ",\"hdr_crc\":\"%u\"", file->hdr_crc);
-			len += sprintf(line+len, ",\"hdr_off\":\"%li\"", file->hdr_off);
+			len += sprintf(line+len, ",\"hdr_off\":\"%llu\"", file->hdr_off);
 			len += sprintf(line+len, ",\"hdr_size\":\"%i\"", file->hdr_size);
 		}
 

src/configure.c

@@ -299,6 +299,7 @@ do_set_config(bool missing_ok)
 
 	for (i = 0; instance_options[i].type; i++)
 	{
+		int rc = 0;
 		ConfigOption *opt = &instance_options[i];
 		char	   *value;
 
@@ -319,13 +320,25 @@ do_set_config(bool missing_ok)
 		}
 
 		if (strchr(value, ' '))
-			fprintf(fp, "%s = '%s'\n", opt->lname, value);
+			rc = fprintf(fp, "%s = '%s'\n", opt->lname, value);
 		else
-			fprintf(fp, "%s = %s\n", opt->lname, value);
+			rc = fprintf(fp, "%s = %s\n", opt->lname, value);
+
+		if (rc < 0)
+			elog(ERROR, "Cannot write to configuration file: \"%s\"", path_temp);
+
 		pfree(value);
 	}
 
-	fclose(fp);
+	if (ferror(fp) || fflush(fp))
+		elog(ERROR, "Cannot write to configuration file: \"%s\"", path_temp);
+
+	if (fclose(fp))
+		elog(ERROR, "Cannot close configuration file: \"%s\"", path_temp);
+
+	if (fio_sync(path_temp, FIO_LOCAL_HOST) != 0)
+		elog(ERROR, "Failed to sync temp configuration file \"%s\": %s",
+			 path_temp, strerror(errno));
 
 	if (rename(path_temp, path) < 0)
 	{

src/data.c

@@ -2160,7 +2160,7 @@ get_data_file_headers(HeaderMap *hdr_map, pgFile *file, uint32 backup_version, b
 
 	if (fseek(in, file->hdr_off, SEEK_SET))
 	{
-		elog(strict ? ERROR : WARNING, "Cannot seek to position %lu in page header map \"%s\": %s",
+		elog(strict ? ERROR : WARNING, "Cannot seek to position %llu in page header map \"%s\": %s",
 			file->hdr_off, hdr_map->path, strerror(errno));
 		goto cleanup;
 	}
@@ -2177,7 +2177,7 @@ get_data_file_headers(HeaderMap *hdr_map, pgFile *file, uint32 backup_version, b
 
 	if (fread(zheaders, 1, file->hdr_size, in) != file->hdr_size)
 	{
-		elog(strict ? ERROR : WARNING, "Cannot read header file at offset: %li len: %i \"%s\": %s",
+		elog(strict ? ERROR : WARNING, "Cannot read header file at offset: %llu len: %i \"%s\": %s",
 			file->hdr_off, file->hdr_size, hdr_map->path, strerror(errno));
 		goto cleanup;
 	}
@@ -2208,7 +2208,7 @@ get_data_file_headers(HeaderMap *hdr_map, pgFile *file, uint32 backup_version, b
 	if (hdr_crc != file->hdr_crc)
 	{
 		elog(strict ? ERROR : WARNING, "Header map for file \"%s\" crc mismatch \"%s\" "
-				"offset: %lu, len: %lu, current: %u, expected: %u",
+				"offset: %llu, len: %lu, current: %u, expected: %u",
 			file->rel_path, hdr_map->path, file->hdr_off, read_len, hdr_crc, file->hdr_crc);
 		goto cleanup;
 	}
@@ -2268,7 +2268,7 @@ write_page_headers(BackupPageHeader2 *headers, pgFile *file, HeaderMap *hdr_map,
 	{
 		elog(LOG, "Creating page header map \"%s\"", map_path);
 
-		hdr_map->fp = fopen(map_path, PG_BINARY_W);
+		hdr_map->fp = fopen(map_path, "a");
 		if (hdr_map->fp == NULL)
 			elog(ERROR, "Cannot open header file \"%s\": %s",
 				 map_path, strerror(errno));
@@ -2297,7 +2297,7 @@ write_page_headers(BackupPageHeader2 *headers, pgFile *file, HeaderMap *hdr_map,
 				 file->rel_path, z_len);
 	}
 
-	elog(VERBOSE, "Writing headers for file \"%s\" offset: %li, len: %i, crc: %u",
+	elog(VERBOSE, "Writing headers for file \"%s\" offset: %llu, len: %i, crc: %u",
 			file->rel_path, file->hdr_off, z_len, file->hdr_crc);
 
 	if (fwrite(zheaders, 1, z_len, hdr_map->fp) != z_len)