Skip to content

Commit 3dba9cb

Browse files
tglsfdctglsfdc
committed
Add a check on file size as an additional safety check that a WAL file
recovered from archive is not corrupt. It's not much but it will catch one common problem, viz out-of-disk-space. Also, force a WAL recovery scan when recovery.conf is present, even if pg_control shows a clean shutdown. This allows recovery with a tar backup that was taken with the postmaster shut down, as per complaint from Mark Kirkwood.
1 parent 406e8aa commit 3dba9cb

File tree

1 file changed

+69
-23
lines changed
  • src/backend/access/transam

1 file changed

+69
-23
lines changed

‎src/backend/access/transam/xlog.c

Lines changed: 69 additions & 23 deletions
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@
77
* Portions Copyright (c) 1996-2003, PostgreSQL Global Development Group
88
* Portions Copyright (c) 1994, Regents of the University of California
99
*
10-
* $PostgreSQL: pgsql/src/backend/access/transam/xlog.c,v 1.150 2004/07/21 22:31:20 tgl Exp $
10+
* $PostgreSQL: pgsql/src/backend/access/transam/xlog.c,v 1.151 2004/07/22 20:18:40 tgl Exp $
1111
*
1212
*-------------------------------------------------------------------------
1313
*/
@@ -432,7 +432,7 @@ static bool InstallXLogFileSegment(uint32 log, uint32 seg, char *tmppath,
432432
static int XLogFileOpen(uint32 log, uint32 seg);
433433
static int XLogFileRead(uint32 log, uint32 seg, int emode);
434434
static bool RestoreArchivedFile(char *path, const char *xlogfname,
435-
const char *recovername);
435+
const char *recovername, off_t expectedSize);
436436
static void PreallocXlogFiles(XLogRecPtr endptr);
437437
static void MoveOfflineLogs(uint32 log, uint32 seg, XLogRecPtr endptr);
438438
static XLogRecord *ReadRecord(XLogRecPtr *RecPtr, int emode, char *buffer);
@@ -1838,7 +1838,8 @@ XLogFileRead(uint32 log, uint32 seg, int emode)
18381838
{
18391839
XLogFileName(xlogfname, tli, log, seg);
18401840
restoredFromArchive = RestoreArchivedFile(path, xlogfname,
1841-
"RECOVERYXLOG");
1841+
"RECOVERYXLOG",
1842+
XLogSegSize);
18421843
}
18431844
else
18441845
XLogFilePath(path, tli, log, seg);
@@ -1876,10 +1877,14 @@ XLogFileRead(uint32 log, uint32 seg, int emode)
18761877
* If not successful, fill "path" with the name of the normal on-line file
18771878
* (which may or may not actually exist, but we'll try to use it), and return
18781879
* FALSE.
1880+
*
1881+
* For fixed-size files, the caller may pass the expected size as an
1882+
* additional crosscheck on successful recovery. If the file size is not
1883+
* known, set expectedSize = 0.
18791884
*/
18801885
static bool
18811886
RestoreArchivedFile(char *path, const char *xlogfname,
1882-
const char *recovername)
1887+
const char *recovername, off_t expectedSize)
18831888
{
18841889
char xlogpath[MAXPGPATH];
18851890
char xlogRestoreCmd[MAXPGPATH];
@@ -1991,19 +1996,42 @@ RestoreArchivedFile(char *path, const char *xlogfname,
19911996
rc = system(xlogRestoreCmd);
19921997
if (rc == 0)
19931998
{
1994-
/* restore success ... assuming file is really there now ... */
1995-
if (stat(xlogpath, &stat_buf) == 0) {
1996-
ereport(LOG,
1997-
(errmsg("restored log file \"%s\" from archive",
1998-
xlogfname)));
1999-
strcpy(path, xlogpath);
2000-
return true;
1999+
/*
2000+
* command apparently succeeded, but let's make sure the file is
2001+
* really there now and has the correct size.
2002+
*
2003+
* XXX I made wrong-size a fatal error to ensure the DBA would
2004+
* notice it, but is that too strong? We could try to plow ahead
2005+
* with a local copy of the file ... but the problem is that there
2006+
* probably isn't one, and we'd incorrectly conclude we've reached
2007+
* the end of WAL and we're done recovering ...
2008+
*/
2009+
if (stat(xlogpath, &stat_buf) == 0)
2010+
{
2011+
if (expectedSize > 0 && stat_buf.st_size != expectedSize)
2012+
ereport(FATAL,
2013+
(errmsg("archive file \"%s\" has wrong size: %lu instead of %lu",
2014+
xlogfname,
2015+
(unsigned long) stat_buf.st_size,
2016+
(unsigned long) expectedSize)));
2017+
else
2018+
{
2019+
ereport(LOG,
2020+
(errmsg("restored log file \"%s\" from archive",
2021+
xlogfname)));
2022+
strcpy(path, xlogpath);
2023+
return true;
2024+
}
2025+
}
2026+
else
2027+
{
2028+
/* stat failed */
2029+
if (errno != ENOENT)
2030+
ereport(FATAL,
2031+
(errcode_for_file_access(),
2032+
errmsg("could not stat \"%s\": %m",
2033+
xlogpath)));
20012034
}
2002-
if (errno != ENOENT)
2003-
ereport(FATAL,
2004-
(errcode_for_file_access(),
2005-
errmsg("could not stat \"%s\": %m",
2006-
xlogpath)));
20072035
}
20082036

20092037
/*
@@ -2664,7 +2692,7 @@ readTimeLineHistory(TimeLineID targetTLI)
26642692
if (InArchiveRecovery)
26652693
{
26662694
TLHistoryFileName(histfname, targetTLI);
2667-
RestoreArchivedFile(path, histfname, "RECOVERYHISTORY");
2695+
RestoreArchivedFile(path, histfname, "RECOVERYHISTORY", 0);
26682696
}
26692697
else
26702698
TLHistoryFilePath(path, targetTLI);
@@ -2749,7 +2777,7 @@ existsTimeLineHistory(TimeLineID probeTLI)
27492777
if (InArchiveRecovery)
27502778
{
27512779
TLHistoryFileName(histfname, probeTLI);
2752-
RestoreArchivedFile(path, histfname, "RECOVERYHISTORY");
2780+
RestoreArchivedFile(path, histfname, "RECOVERYHISTORY", 0);
27532781
}
27542782
else
27552783
TLHistoryFilePath(path, probeTLI);
@@ -2853,7 +2881,7 @@ writeTimeLineHistory(TimeLineID newTLI, TimeLineID parentTLI,
28532881
if (InArchiveRecovery)
28542882
{
28552883
TLHistoryFileName(histfname, parentTLI);
2856-
RestoreArchivedFile(path, histfname, "RECOVERYHISTORY");
2884+
RestoreArchivedFile(path, histfname, "RECOVERYHISTORY", 0);
28572885
}
28582886
else
28592887
TLHistoryFilePath(path, parentTLI);
@@ -4042,6 +4070,11 @@ StartupXLOG(void)
40424070
if (checkPoint.undo.xrecoff == 0)
40434071
checkPoint.undo = RecPtr;
40444072

4073+
/*
4074+
* Check whether we need to force recovery from WAL. If it appears
4075+
* to have been a clean shutdown and we did not have a recovery.conf
4076+
* file, then assume no recovery needed.
4077+
*/
40454078
if (XLByteLT(checkPoint.undo, RecPtr) ||
40464079
XLByteLT(checkPoint.redo, RecPtr))
40474080
{
@@ -4054,13 +4087,23 @@ StartupXLOG(void)
40544087
InRecovery = true;
40554088

40564089
/* REDO */
4057-
if (InRecovery)
4090+
if (InRecovery || InArchiveRecovery)
40584091
{
40594092
int rmid;
40604093

4061-
ereport(LOG,
4062-
(errmsg("database system was not properly shut down; "
4063-
"automatic recovery in progress")));
4094+
if (InRecovery)
4095+
{
4096+
ereport(LOG,
4097+
(errmsg("database system was not properly shut down; "
4098+
"automatic recovery in progress")));
4099+
}
4100+
else
4101+
{
4102+
/* force recovery due to presence of recovery.conf */
4103+
InRecovery = true;
4104+
ereport(LOG,
4105+
(errmsg("automatic recovery in progress")));
4106+
}
40644107
ControlFile->state = DB_IN_RECOVERY;
40654108
ControlFile->time = time(NULL);
40664109
UpdateControlFile();
@@ -4158,8 +4201,11 @@ StartupXLOG(void)
41584201
InRedo = false;
41594202
}
41604203
else
4204+
{
4205+
/* there are no WAL records following the checkpoint */
41614206
ereport(LOG,
41624207
(errmsg("redo is not required")));
4208+
}
41634209
}
41644210

41654211
/*

0 commit comments

Comments
 (0)