db/pg_probackup
db/pg_probackup
1
0
Fork 0
mirror of https://github.com/postgrespro/pg_probackup.git synced 2025-03-17 21:18:00 +02:00
Code Issues Releases Activity

added some improvements based on review

Browse Source
This commit is contained in:
Grigory Smolkin 2019-09-06 11:26:42 +03:00
parent 0e2336fb6f
commit 07f63e473b
1 changed files with 42 additions and 50 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

92
src/backup.c
View File

@ -1399,31 +1399,32 @@ pg_ptrack_get_and_clear(Oid tablespace_oid, Oid db_oid, Oid rel_filenode,
}
/*
* Wait for target 'lsn' or WAL segment, containing 'lsn'.
* Wait for target LSN or WAL segment, containing target LSN.
*
* If current backup started in archive mode wait for 'lsn' to be archived in
* archive 'wal' directory with WAL segment file.
* If current backup started in stream mode wait for 'lsn' to be streamed in
* 'pg_wal' directory.
* Depending on value of flag in_stream_dir wait for target LSN to archived or
* streamed in 'archive_dir' or 'pg_wal' directory.
*
* If 'is_start_lsn' is true then issue warning for first-time users.
* If flag 'is_start_lsn' is set then issue warning for first-time users.
* If flag 'in_prev_segment' is set, look for LSN in previous segment,
* with EndRecPtr >= Target LSN. It should be used only for solving
* invalid XRecOff problem.
* If flag 'segment_only' is set, then, instead of waiting for LSN, wait for segment,
* containing that LSN.
* If flags 'in_prev_segment' and 'segment_only' are both set, then wait for
* previous segment.
*
* If 'in_prev_segment' is set, look for LSN in previous segment.
* If 'segment_only' is set, then instead of looking for LSN, look for segment itself.
* If 'in_prev_segment' and 'segment_only' are both set, then wait for previous segment.
* Flag 'in_stream_dir' determine whether we looking for WAL in 'pg_wal' directory or
* in archive. Do note, that we cannot rely sorely on global variable 'stream_wal' because,
* for example, PAGE backup must(!) look for start_lsn in archive regardless of wal_mode.
*
* Flag 'in_stream_dir' determine whether we looking for wal in 'pg_wal' directory or
* in archive. Do note, that we cannot rely sorely on 'stream_wal' because, for example,
* PAGE backup must(!) look for start_lsn in archive regardless of wal_mode.
* 'timeout_elevel' determine the elevel for timeout elog message. If elevel lighter than
* ERROR is used, then return InvalidXLogRecPtr. TODO: return something more concrete, for example 1.
*
* Returns LSN of last valid record if wait_prev_segment is not true, otherwise
* returns InvalidXLogRecPtr.
* Returns target LSN if such is found, failing that returns LSN of record prior to target LSN.
* Returns InvalidXLogRecPtr if 'segment_only' flag is used.
*/
static XLogRecPtr
wait_wal_lsn(XLogRecPtr lsn, bool is_start_lsn, TimeLineID tli,
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)
{
@ -1442,16 +1443,16 @@ wait_wal_lsn(XLogRecPtr lsn, bool is_start_lsn, TimeLineID tli,
#endif
/* Compute the name of the WAL file containing requested LSN */
GetXLogSegNo(lsn, targetSegNo, instance_config.xlog_seg_size);
GetXLogSegNo(target_lsn, targetSegNo, instance_config.xlog_seg_size);
if (in_prev_segment)
targetSegNo--;
GetXLogFileName(wal_segment, tli, targetSegNo,
instance_config.xlog_seg_size);
/*
* In pg_start_backup we wait for 'lsn' in 'pg_wal' directory if it is
* In pg_start_backup we wait for 'target_lsn' in 'pg_wal' directory if it is
* stream and non-page backup. Page backup needs archived WAL files, so we
* wait for 'lsn' in archive 'wal' directory for page backups.
* wait for 'target_lsn' in archive 'wal' directory for page backups.
*
* In pg_stop_backup it depends only on stream_wal.
*/
@ -1468,6 +1469,7 @@ wait_wal_lsn(XLogRecPtr lsn, bool is_start_lsn, TimeLineID tli,
wal_segment_dir = arclog_path;
}
/* TODO: remove this in 3.0 (it is a cludge against some old bug with archive_timeout) */
if (instance_config.archive_timeout > 0)
timeout = instance_config.archive_timeout;
else
@ -1477,7 +1479,7 @@ wait_wal_lsn(XLogRecPtr lsn, bool is_start_lsn, TimeLineID tli,
elog(LOG, "Looking for segment: %s", wal_segment);
else
elog(LOG, "Looking for LSN %X/%X in segment: %s",
(uint32) (lsn >> 32), (uint32) lsn, wal_segment);
(uint32) (target_lsn >> 32), (uint32) target_lsn, wal_segment);
#ifdef HAVE_LIBZ
snprintf(gz_wal_segment_path, sizeof(gz_wal_segment_path), "%s.gz",
@ -1506,26 +1508,24 @@ wait_wal_lsn(XLogRecPtr lsn, bool is_start_lsn, TimeLineID tli,
if (file_exists)
{
//TODO is this comment also relates to current segment_only case
// and should be updated?
/* Do not check LSN for previous WAL segment */
/* Do not check for target LSN */
if (segment_only)
return InvalidXLogRecPtr;
/*
* A WAL segment found. Check LSN on it.
* A WAL segment found. Look for target LSN in it.
*/
if (!XRecOffIsNull(lsn) &&
wal_contains_lsn(wal_segment_dir, lsn, tli,
if (!XRecOffIsNull(target_lsn) &&
wal_contains_lsn(wal_segment_dir, target_lsn, tli,
instance_config.xlog_seg_size))
/* Target LSN was found */
{
elog(LOG, "Found LSN: %X/%X", (uint32) (lsn >> 32), (uint32) lsn);
return lsn;
elog(LOG, "Found LSN: %X/%X", (uint32) (target_lsn >> 32), (uint32) target_lsn);
return target_lsn;
}
/*
* If we failed to get LSN of valid record in a reasonable time, try
* If we failed to get target LSN in a reasonable time, try
* to get LSN of last valid record prior to the target LSN. But only
* in case of a backup from a replica.
@ -1539,7 +1539,7 @@ wait_wal_lsn(XLogRecPtr lsn, bool is_start_lsn, TimeLineID tli,
{
XLogRecPtr res;
res = get_last_wal_lsn(wal_segment_dir, current.start_lsn, lsn, tli,
res = get_last_wal_lsn(wal_segment_dir, current.start_lsn, target_lsn, tli,
in_prev_segment, instance_config.xlog_seg_size);
if (!XLogRecPtrIsInvalid(res))
@ -1565,7 +1565,7 @@ wait_wal_lsn(XLogRecPtr lsn, bool is_start_lsn, TimeLineID tli,
wal_segment_path, wal_delivery_str);
else
elog(INFO, "Wait for LSN %X/%X in %s WAL segment %s",
(uint32) (lsn >> 32), (uint32) lsn,
(uint32) (target_lsn >> 32), (uint32) target_lsn,
wal_delivery_str, wal_segment_path);
}
@ -1580,7 +1580,7 @@ wait_wal_lsn(XLogRecPtr lsn, bool is_start_lsn, TimeLineID tli,
elog(timeout_elevel, "WAL segment %s was %s, "
"but target LSN %X/%X could not be archived in %d seconds",
wal_segment, wal_delivery_str,
(uint32) (lsn >> 32), (uint32) lsn, timeout);
(uint32) (target_lsn >> 32), (uint32) target_lsn, timeout);
/* If WAL segment doesn't exist or we wait for previous segment */
else
elog(timeout_elevel,
@ -1808,11 +1808,10 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
XLogRecPtr lsn_tmp = InvalidXLogRecPtr;
/*
* TODO Let's rephrase that to be less scary for a user.
* Even though the value is invalid, it's expected postgres behaviour
* and we're trying to fix it below.
*/
elog(WARNING, "Invalid stop_backup_lsn value %X/%X",
elog(LOG, "Null offset in stop_backup_lsn value %X/%X, trying to fix",
(uint32) (stop_backup_lsn_tmp >> 32), (uint32) (stop_backup_lsn_tmp));
/*
@ -1844,15 +1843,12 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
* and to use it in WAL validation.
*
* So we try to do the following:
* 1. Wait for current segment and look in it for the LSN >= STOP_LSN.
* //TODO what is 'current' segment?
* //TODO how long do we wait? is there a timeout?
* It solves the problem of occasional invalid XRecOff
* on write-busy system.
* 1. Wait 'archive_timeout' seconds for segment containing stop_lsn and
* look for the first valid record in it.
* It solves the problem of occasional invalid XRecOff on write-busy system.
* 2. Failing that, look for record in previous segment with endpoint
* equal or greater than 0 offset LSN. It may(!) solve the problem of 0 offset
* on write-idle system.
* //TODO what if if didn't?
* equal or greater than stop_lsn. It may(!) solve the problem of 0 offset
* on write-idle system. If that fails too, error out.
* //TODO what kind of record that refers to?
*/
@ -1864,8 +1860,7 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
lsn_tmp = get_first_wal_lsn(xlog_path, segno, backup->tli,
instance_config.xlog_seg_size);
/* Check if returned LSN is satisfying our requirements */
//TODO what requirements?
/* Check that returned LSN is valid and greater than stop_lsn */
if (XLogRecPtrIsInvalid(lsn_tmp) ||
!XRecOffIsValid(lsn_tmp) ||
lsn_tmp < stop_backup_lsn_tmp)
@ -1893,9 +1888,7 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
stop_backup_lsn = lsn_tmp;
stop_lsn_exists = true;
}
//TODO is it good to throw an error here? Shouldn't we rather
// save it as is and mark backup as DONE instead of OK
// and check it in validation and restore?
/* Replica returned something very illegal, error out */
else
elog(ERROR, "Invalid stop_backup_lsn value %X/%X",
(uint32) (stop_backup_lsn_tmp >> 32), (uint32) (stop_backup_lsn_tmp));
@ -2000,9 +1993,8 @@ pg_stop_backup(pgBackup *backup, PGconn *pg_startbackup_conn,
/*
* Wait for stop_lsn to be archived or streamed.
* //TODO is the sentence below outdated?
* If replica returned non-existent LSN, look for previous record,
* which endpoint >= stop_lsn
* If replica returned valid STOP_LSN of not actually existing record,
* look for previous record with endpoint >= STOP_LSN.
*/
if (!stop_lsn_exists)
stop_backup_lsn = wait_wal_lsn(stop_backup_lsn_tmp, false, backup->tli,