Merge branch 'master' into release_2_5

Author: gsmolk

.travis.yml

@@ -45,3 +45,8 @@ env:
 jobs:
     allow_failures:
         - if: env(MODE) IN (archive, backup, delta, locking, merge, replica, retention, restore)
+
+# Only run CI for master branch commits to limit our travis usage
+branches:
+    only:
+    - master

Documentation.md

@@ -56,9 +56,17 @@ Regardless of the chosen backup type, all backups taken with `pg_probackup` supp
 
 Documentation can be found at [github](https://postgrespro.github.io/pg_probackup) and [Postgres Professional documentation](https://postgrespro.com/docs/postgrespro/current/app-pgprobackup)
 
+## Development
+
+* Stable version state can be found under the respective [release tag](https://github.com/postgrespro/pg_probackup/releases).
+* `master` branch contains minor fixes that are planned to the nearest minor release.
+* Upcoming major release is developed in a release branch i.e. `release_2_5`.
+
+For detailed release plans check [Milestones](https://github.com/postgrespro/pg_probackup/milestones)
+
 ## Installation and Setup
 ### Windows Installation
-Installers are available in release **assets**. [Latests](https://github.com/postgrespro/pg_probackup/releases/2.4.4).
+Installers are available in release **assets**. [Latests](https://github.com/postgrespro/pg_probackup/releases/2.4.9).
 
 ### Linux Installation
 #### pg_probackup for vanilla PostgreSQL
@@ -160,7 +168,7 @@ sudo apt-get install pg_probackup-{std,ent}-{12,11,10,9.6}
 sudo apt-get install pg_probackup-{std,ent}-{12,11,10,9.6}-debuginfo
 ```
 
-Once you have `pg_probackup` installed, complete [the setup](https://github.com/postgrespro/pg_probackup/blob/master/Documentation.md#installation-and-setup).
+Once you have `pg_probackup` installed, complete [the setup](https://postgrespro.github.io/pg_probackup/#pbk-install-and-setup).
 
 ## Building from source
 ### Linux

README.md

@@ -2,4 +2,4 @@
 ```
 xmllint --noout --valid probackup.xml
 xsltproc stylesheet.xsl probackup.xml >pg-probackup.html
-```
+```

doc/Readme.md

@@ -1206,18 +1206,6 @@ CREATE EXTENSION ptrack;
        does not affect PTRACK operation. The maximum allowed value is 1024.
       </para>
      </listitem>
-     <listitem>
-      <para>
-       Grant the right to execute <application>PTRACK</application>
-       functions to the <literal>backup</literal> role
-       in the database used to connect to the cluster:
-      </para>
-      <programlisting>
-GRANT EXECUTE ON FUNCTION pg_ptrack_get_pagemapset(pg_lsn) TO backup;
-GRANT EXECUTE ON FUNCTION pg_ptrack_control_lsn() TO backup;
-GRANT EXECUTE ON FUNCTION pg_ptrack_get_block(oid, oid, oid, bigint) TO backup;
-</programlisting>
-      </listitem>
     </orderedlist>
 
    <note>
@@ -1254,7 +1242,6 @@ GRANT EXECUTE ON FUNCTION pg_ptrack_get_block(oid, oid, oid, bigint) TO backup;
         <programlisting>
 GRANT EXECUTE ON FUNCTION pg_catalog.pg_ptrack_clear() TO backup;
 GRANT EXECUTE ON FUNCTION pg_catalog.pg_ptrack_get_and_clear(oid, oid) TO backup;
-GRANT EXECUTE ON FUNCTION pg_catalog.pg_stop_backup() TO backup;
 </programlisting>
       </listitem>
     </orderedlist>
@@ -1875,8 +1862,9 @@ pg_probackup restore -B <replaceable>backup_dir</replaceable> --instance <replac
         <command>DROP DATABASE</command> command.
       </para>
       <para>
-        If want to decouple a single cluster with multiple databases into separate clusters with minimal downtime, then
-        you can partially restore cluster as standby via <option>--restore-as-replica</option> options.
+        To decouple a single cluster containing multiple databases into separate clusters with minimal downtime,
+        you can do partial restore of the cluster as a standby using the <option>--restore-as-replica</option> option
+        for specific databases.
       </para>
       <note>
         <para>
@@ -1886,10 +1874,12 @@ pg_probackup restore -B <replaceable>backup_dir</replaceable> --instance <replac
       </note>
       <note>
         <para>
-          Due to how recovery works in PostgreSQL < 12 it is advisable to
-          disable <option>hot_standby</option> option, when running partial
-          restore of PostgreSQL cluster of version less than <option>12</option>.
-          Otherwise recovery may fail.
+          Due to recovery specifics of PostgreSQL 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.
+          Otherwise the recovery may fail.
         </para>
       </note>
     </refsect3>
@@ -3978,10 +3968,11 @@ pg_probackup restore -B <replaceable>backup_dir</replaceable> --instance <replac
         this flag if you need to restore the
         <productname>PostgreSQL</productname> cluster from a corrupt or an invalid backup.
         Use with caution.
-        When used with <link linkend="pbk-incremental-restore">incremental restore</link> this flag
-        allows to replace already existing PGDATA with different system ID. In case of tablespaces,
-        remapped via <literal>--tablespace-mapping</literal> option into not empty directories,
-        the old content of such directories will be deleted.
+        If <envar>PGDATA</envar> contains a non-empty directory with system ID different from that
+        of the backup being restored, <link linkend="pbk-incremental-restore">incremental restore</link>
+        with this flag overwrites the directory contents (while an error occurs without the flag). If tablespaces
+        are remapped through the <literal>--tablespace-mapping</literal> option into non-empty directories,
+        the contents of such directories will be deleted.
       </para>
       </listitem>
       </varlistentry>
@@ -4427,7 +4418,7 @@ pg_probackup archive-get -B <replaceable>backup_dir</replaceable> --instance <re
       <listitem>
       <para>
         Specifies the timestamp up to which recovery will proceed.
-        If timezone offset is not specified, local timezone is used.
+        If the time zone offset is not specified, the local time zone is used.
       </para>
       <para>
        Example: <literal>--recovery-target-time='2020-01-01 00:00:00+03'</literal>
@@ -4617,7 +4608,7 @@ pg_probackup archive-get -B <replaceable>backup_dir</replaceable> --instance <re
       <para>
           Specifies the timestamp up to which the backup will stay
           pinned. Must be an ISO-8601 complaint timestamp.
-          If timezone offset is not specified, local timezone is used.
+          If the time zone offset is not specified, the local time zone is used.
       </para>
       <para>
        Example: <literal>--expire-time='2020-01-01 00:00:00+03'</literal>

doc/pgprobackup.xml

@@ -421,6 +421,8 @@ push_file_internal_uncompressed(const char *wal_file_name, const char *pg_xlog_d
 	int			partial_try_count = 0;
 	int			partial_file_size = 0;
 	bool		partial_is_stale = true;
+	/* remote agent error message */
+	char       *errmsg = NULL;
 
 	/* from path */
 	join_path_components(from_fullpath, pg_xlog_dir, wal_file_name);
@@ -579,7 +581,7 @@ push_file_internal_uncompressed(const char *wal_file_name, const char *pg_xlog_d
 						from_fullpath, strerror(errno));
 		}
 
-		if (read_len > 0 && fio_write(out, buf, read_len) != read_len)
+		if (read_len > 0 && fio_write_async(out, buf, read_len) != read_len)
 		{
 			fio_unlink(to_fullpath_part, FIO_BACKUP_HOST);
 			elog(ERROR, "Cannot write to destination temp file \"%s\": %s",
@@ -593,6 +595,14 @@ push_file_internal_uncompressed(const char *wal_file_name, const char *pg_xlog_d
 	/* close source file */
 	fclose(in);
 
+	/* Writing is asynchronous in case of push in remote mode, so check agent status */
+	if (fio_check_error_fd(out, &errmsg))
+	{
+		fio_unlink(to_fullpath_part, FIO_BACKUP_HOST);
+		elog(ERROR, "Cannot write to the remote file \"%s\": %s",
+					to_fullpath_part, errmsg);
+	}
+
 	/* close temp file */
 	if (fio_close(out) != 0)
 	{
@@ -652,6 +662,8 @@ push_file_internal_gz(const char *wal_file_name, const char *pg_xlog_dir,
 	int			partial_try_count = 0;
 	int			partial_file_size = 0;
 	bool		partial_is_stale = true;
+	/* remote agent errormsg */
+	char       *errmsg = NULL;
 
 	/* from path */
 	join_path_components(from_fullpath, pg_xlog_dir, wal_file_name);
@@ -804,6 +816,7 @@ push_file_internal_gz(const char *wal_file_name, const char *pg_xlog_dir,
 	}
 
 	/* copy content */
+	/* TODO: move to separate function */
 	for (;;)
 	{
 		size_t  read_len = 0;
@@ -831,7 +844,15 @@ push_file_internal_gz(const char *wal_file_name, const char *pg_xlog_dir,
 	/* close source file */
 	fclose(in);
 
-	/* close temp file */
+	/* Writing is asynchronous in case of push in remote mode, so check agent status */
+	if (fio_check_error_fd_gz(out, &errmsg))
+	{
+		fio_unlink(to_fullpath_gz_part, FIO_BACKUP_HOST);
+		elog(ERROR, "Cannot write to the remote compressed file \"%s\": %s",
+					to_fullpath_gz_part, errmsg);
+	}
+
+	/* close temp file, TODO: make it synchronous */
 	if (fio_gzclose(out) != 0)
 	{
 		fio_unlink(to_fullpath_gz_part, FIO_BACKUP_HOST);

src/archive.c

@@ -1084,21 +1084,22 @@ restore_data_file_internal(FILE *in, FILE *out, pgFile *file, uint32 backup_vers
 			cur_pos_out = write_pos;
 		}
 
-		/* If page is compressed and restore is in remote mode, send compressed
-		 * page to the remote side.
+		/*
+		 * If page is compressed and restore is in remote mode,
+		 * send compressed page to the remote side.
 		 */
 		if (is_compressed)
 		{
 			ssize_t rc;
-			rc = fio_fwrite_compressed(out, page.data, compressed_size, file->compress_alg);
+			rc = fio_fwrite_async_compressed(out, page.data, compressed_size, file->compress_alg);
 
 			if (!fio_is_remote_file(out) && rc != BLCKSZ)
 				elog(ERROR, "Cannot write block %u of \"%s\": %s, size: %u",
 					 blknum, to_fullpath, strerror(errno), compressed_size);
 		}
 		else
 		{
-			if (fio_fwrite(out, page.data, BLCKSZ) != BLCKSZ)
+			if (fio_fwrite_async(out, page.data, BLCKSZ) != BLCKSZ)
 				elog(ERROR, "Cannot write block %u of \"%s\": %s",
 					 blknum, to_fullpath, strerror(errno));
 		}
@@ -1144,7 +1145,7 @@ restore_non_data_file_internal(FILE *in, FILE *out, pgFile *file,
 
 		if (read_len > 0)
 		{
-			if (fio_fwrite(out, buf, read_len) != read_len)
+			if (fio_fwrite_async(out, buf, read_len) != read_len)
 				elog(ERROR, "Cannot write to \"%s\": %s", to_fullpath,
 					 strerror(errno));
 		}

src/data.c

@@ -1206,7 +1206,7 @@ extern datapagemap_t *fio_get_lsn_map(const char *fullpath, uint32 checksum_vers
 							fio_location location);
 extern pid_t fio_check_postmaster(const char *pgdata, fio_location location);
 
-extern int32 fio_decompress(void* dst, void const* src, size_t size, int compress_alg);
+extern int32 fio_decompress(void* dst, void const* src, size_t size, int compress_alg, char **errormsg);
 
 /* return codes for fio_send_pages() and fio_send_file() */
 #define SEND_OK       (0)

src/pg_probackup.h

@@ -1109,6 +1109,7 @@ restore_files(void *arg)
 		bool     already_exists = false;
 		PageState      *checksum_map = NULL; /* it should take ~1.5MB at most */
 		datapagemap_t  *lsn_map = NULL;      /* it should take 16kB at most */
+		char           *errmsg = NULL;       /* remote agent error message */
 		pgFile	*dest_file = (pgFile *) parray_get(arguments->dest_files, i);
 
 		/* Directories were created before */
@@ -1262,6 +1263,10 @@ restore_files(void *arg)
 		}
 
 done:
+		/* Writing is asynchronous in case of restore in remote mode, so check the agent status */
+		if (fio_check_error_file(out, &errmsg))
+			elog(ERROR, "Cannot write to the remote file \"%s\": %s", to_fullpath, errmsg);
+
 		/* close file */
 		if (fio_fclose(out) != 0)
 			elog(ERROR, "Cannot close file \"%s\": %s", to_fullpath,
@@ -1510,8 +1515,8 @@ update_recovery_options(pgBackup *backup,
 	char		postgres_auto_path[MAXPGPATH];
 	char		postgres_auto_path_tmp[MAXPGPATH];
 	char		path[MAXPGPATH];
-	FILE	   *fp;
-	FILE	   *fp_tmp;
+	FILE	   *fp = NULL;
+	FILE	   *fp_tmp = NULL;
 	struct stat st;
 	char		current_time_str[100];
 	/* postgresql.auto.conf parsing */
@@ -1535,9 +1540,13 @@ update_recovery_options(pgBackup *backup,
 				 strerror(errno));
 	}
 
-	fp = fio_open_stream(postgres_auto_path, FIO_DB_HOST);
-	if (fp == NULL && errno != ENOENT)
-		elog(ERROR, "cannot open \"%s\": %s", postgres_auto_path, strerror(errno));
+	/* Kludge for 0-sized postgresql.auto.conf file. TODO: make something more intelligent */
+	if (st.st_size > 0)
+	{
+		fp = fio_open_stream(postgres_auto_path, FIO_DB_HOST);
+		if (fp == NULL)
+			elog(ERROR, "cannot open \"%s\": %s", postgres_auto_path, strerror(errno));
+	}
 
 	sprintf(postgres_auto_path_tmp, "%s.tmp", postgres_auto_path);
 	fp_tmp = fio_fopen(postgres_auto_path_tmp, "w", FIO_DB_HOST);
@@ -1577,9 +1586,11 @@ update_recovery_options(pgBackup *backup,
 	if (fp)
 		fio_close_stream(fp);
 
-	/* TODO: detect remote error */
-	if (buf_len > 0)
-		fio_fwrite(fp_tmp, buf, buf_len);
+	/* Write data to postgresql.auto.conf.tmp */
+	if (buf_len > 0 &&
+		(fio_fwrite(fp_tmp, buf, buf_len) != buf_len))
+			elog(ERROR, "Cannot write to \"%s\": %s",
+					postgres_auto_path_tmp, strerror(errno));
 
 	if (fio_fflush(fp_tmp) != 0 ||
 		fio_fclose(fp_tmp))