I/O error on data file, can't run backup

Leif Biberg Kristensen <leif@solumslekt.org> writes:

Running postgresql 9.0.5 on
balapapa ~ # uname -a
Linux balapapa 2.6.39-gentoo-r3 #1 SMP Sun Jul 17 11:22:15 CEST 2011 x86_64
Intel(R) Core(TM) i7 CPU 930 @ 2.80GHz GenuineIntel GNU/Linux

I'm trying to run pg_dump on my database, and get an error:

pg_dump: SQL command failed
pg_dump: Error message from server: ERROR: could not read block 1 in file
"base/612249/11658": Inn/ut-feil
pg_dump: The command was: SELECT tableoid, oid, opfname, opfnamespace, (SELECT
rolname FROM pg_catalog.pg_roles WHERE oid = opfowner) AS rolname FROM
pg_opfamily

I have tried to stop postgresql and take a filesystem backup of the data
directory with a cp -ax, but it crashes on the same file.

You have a disk failure on some sector of that file, apparently. I'd be
thinking about replacing that disk drive if I were you. Once it starts
showing uncorrectable errors the MTTF is going to be short.

The backup script overwrote my previous backup with a 40 byte file (yes silly
me I know that's bloody stupid - I'm gonna fix that) and now I haven't got a
recent backup anymore.

Is this fixable?

Postgres can't magically resurrect data that your drive lost, if that's
what you were hoping for. However, you might be in luck, because that
file is probably just an index and not original data. Try this:

select relname from pg_class where relfilenode = 11658;

On my 9.0 installation I get "pg_opclass_am_name_nsp_index". If you get
the same (or any other index for that matter) just reindex that index
and you'll be all right ... or at least, you will be if that's the only
file your drive has lost.

regards, tom lane

On Wednesday 5. October 2011 20.42.00 Tom Lane wrote:

Postgres can't magically resurrect data that your drive lost, if that's
what you were hoping for. However, you might be in luck, because that
file is probably just an index and not original data. Try this:

select relname from pg_class where relfilenode = 11658;

On my 9.0 installation I get "pg_opclass_am_name_nsp_index". If you get
the same (or any other index for that matter) just reindex that index
and you'll be all right ... or at least, you will be if that's the only
file your drive has lost.

Tom,
this is what I get:

postgres@balapapa ~ $ psql pgslekt
psql (9.0.5)
Type "help" for help.

pgslekt=# select relname from pg_class where relfilenode = 11658;
relname
-------------
pg_opfamily
(1 row)

regards, Leif

I seemingly fixed the problem by stopping postgres and doing:

balapapa 612249 # mv 11658 11658.old
balapapa 612249 # mv 11658.old 11658

And the backup magically works.

I'm gonna move the data to another disk right now.

regards, Leif

Leif Biberg Kristensen <leif@solumslekt.org> writes:

I seemingly fixed the problem by stopping postgres and doing:
balapapa 612249 # mv 11658 11658.old
balapapa 612249 # mv 11658.old 11658

And the backup magically works.

Wow, that is magic. I was going to suggest copying pg_opfamily from
template0, which would probably work (maybe requiring reindexing) as
long as you didn't have any non-core data types in use. But you
got lucky.

I'm gonna move the data to another disk right now.

Good plan.

regards, tom lane

On Wednesday 5. October 2011 22.41.49 Tom Lane wrote:

Leif Biberg Kristensen <leif@solumslekt.org> writes:

I'm gonna move the data to another disk right now.

Good plan.

Couple of things I forgot to mention, in case it matters:

The disk is a 1 TB Seagate Barracuda S-ATA, and it has been in use for about a
year. I've been using this brand since way back around 1998 without any
problems, but have never used any disk more than 3 years. The file system is
ext3.

I had a hang on the machine a few hours earlier that required a power-off
reboot. That has been a problem with this rig since I built it about a year
ago, it's probably a funky connection somewhere. This may be the direct cause
of the I/O error, which also may mean that the disk is not to blame.

I'm so used to postgres and everything else coming up without a hiccup after a
power-off that I don't usually pay much attention to it. But I'm certainly
going to rework my backup strategy, and keep several generations.

regards, Leif

On 10/05/2011 02:48 PM, Leif Biberg Kristensen wrote:

I had a hang on the machine a few hours earlier that required a power-off
reboot. That has been a problem with this rig since I built it about a year
ago, it's probably a funky connection somewhere. This may be the direct cause
of the I/O error, which also may mean that the disk is not to blame.

I'm so used to postgres and everything else coming up without a hiccup after a
power-off that I don't usually pay much attention to it

PostgreSQL is great, but it can't overcome defective hardware.

I'm thinking perhaps a funky memory problem - you are having odd crashes
after all.

If memory is failing you could have a file that is corrupted not on disk
but in the cache. Perhaps in the process of stopping and starting
PostgreSQL, the data that was causing the trouble got flushed from cache
then reread from disk. You may find this story interesting:
http://blogs.oracle.com/ksplice/entry/attack_of_the_cosmic_rays1

Cheers,
Steve

On Thursday 6. October 2011 00.17.38 Steve Crawford wrote:

I'm thinking perhaps a funky memory problem - you are having odd crashes
after all.

I've been thinking about the memory myself, but it passes memtest86plus with
flying colors. Or at least it did the last time I checked which is a few months
ago.

The problems got a lot better after I replaced a monster Radeon XFX video card
with a very basic fanless NVidia card (with the added bonus that I can now
actually watch Flash videos in full screen), which may point to overheating
issues.

In other news: I discovered that injecting `date +%u` into the backup file name
at an appropriate place will number it by weekday, which is great for keeping
daily backups for a week.

regards, Leif.

On 10/05/2011 03:43 PM, Leif Biberg Kristensen wrote:

On Thursday 6. October 2011 00.17.38 Steve Crawford wrote:

I'm thinking perhaps a funky memory problem - you are having odd crashes
after all.

I've been thinking about the memory myself, but it passes memtest86plus with
flying colors. Or at least it did the last time I checked which is a few months
ago.

I have had two machines pass extensive memtest86plus but fail on heavy
pgbench testing and in both cases the cause was ultimately traced to bad
memory.

Cheers,
Steve

On 10/06/2011 03:06 AM, Leif Biberg Kristensen wrote:

I seemingly fixed the problem by stopping postgres and doing:

balapapa 612249 # mv 11658 11658.old
balapapa 612249 # mv 11658.old 11658

And the backup magically works.

Woooooo! That's ... "interesting".

I'd be inclined to suspect filesystem corruption, a file system bug /
kernel bug (not very likely if you're on ext3), flakey RAM, etc rather
than a failing disk ... though a failing disk _could_ still be the culprit.

Use smartmontools to do a self-test; if 'smartctl -d ata -t long
/dev/sdx' (where 'x' is the drive node) is reported by 'smartctl -d ata
-a /dev/sdx' as having passed, there are no pending or uncorrectable
sectors, and the disk status is reported as 'HEALTHY' your disk is quite
likely OK. Note that a 'PASSED' or 'HEALTHY' report by its self doesn't
mean much, disk firmwares often return HEALTHY even when the disk can't
even read sector 0.

I strongly recommend making a full backup, both a pg_dump *and* a
file-system level copy of the datadir. Personally I'd then do a test
restore of the pg_dump backup on a separate Pg instance and if it looked
OK I'd re-initdb then reload from the dump.

--
Craig Ringer

On Thursday 6. October 2011 07.07.11 Craig Ringer wrote:

On 10/06/2011 03:06 AM, Leif Biberg Kristensen wrote:

I seemingly fixed the problem by stopping postgres and doing:

balapapa 612249 # mv 11658 11658.old
balapapa 612249 # mv 11658.old 11658

And the backup magically works.

Woooooo! That's ... "interesting".

I'd be inclined to suspect filesystem corruption, a file system bug /
kernel bug (not very likely if you're on ext3), flakey RAM, etc rather
than a failing disk ... though a failing disk _could_ still be the culprit.

Use smartmontools to do a self-test; if 'smartctl -d ata -t long
/dev/sdx' (where 'x' is the drive node) is reported by 'smartctl -d ata
-a /dev/sdx' as having passed, there are no pending or uncorrectable
sectors, and the disk status is reported as 'HEALTHY' your disk is quite
likely OK. Note that a 'PASSED' or 'HEALTHY' report by its self doesn't
mean much, disk firmwares often return HEALTHY even when the disk can't
even read sector 0.

I strongly recommend making a full backup, both a pg_dump *and* a
file-system level copy of the datadir. Personally I'd then do a test
restore of the pg_dump backup on a separate Pg instance and if it looked
OK I'd re-initdb then reload from the dump.

Craig,
Thank you very much for the tip on smartmontools, which I didn't know about.
There indeed appears to be some problems with this disk:

8<---

balapapa ~ # smartctl -d ata -a /dev/sdb -s on
smartctl 5.40 2010-10-16 r3189 [x86_64-pc-linux-gnu] (local build)
Copyright (C) 2002-10 by Bruce Allen, http://smartmontools.sourceforge.net

=== START OF INFORMATION SECTION ===
Model Family: Seagate Barracuda 7200.11 family
Device Model: ST31000340AS
Serial Number: 9QJ1ZMHY
Firmware Version: SD15
User Capacity: 1 000 204 886 016 bytes
Device is: In smartctl database [for details use: -P show]
ATA Version is: 8
ATA Standard is: ATA-8-ACS revision 4
Local Time is: Thu Oct 6 07:46:19 2011 CEST

==> WARNING: There are known problems with these drives,
AND THIS FIRMWARE VERSION IS AFFECTED,
see the following Seagate web pages:
http://seagate.custkb.com/seagate/crm/selfservice/search.jsp?DocId=207931
http://seagate.custkb.com/seagate/crm/selfservice/search.jsp?DocId=207951

SMART support is: Available - device has SMART capability.
SMART support is: Disabled

=== START OF ENABLE/DISABLE COMMANDS SECTION ===
SMART Enabled.

=== START OF READ SMART DATA SECTION ===
SMART overall-health self-assessment test result: PASSED

General SMART Values:
Offline data collection status: (0x82) Offline data collection activity
was completed without error.
Auto Offline Data Collection: Enabled.
Self-test execution status: ( 25) The self-test routine was aborted by
the host.
Total time to complete Offline
data collection: ( 650) seconds.
Offline data collection
capabilities: (0x7b) SMART execute Offline immediate.
Auto Offline data collection on/off
support.
Suspend Offline collection upon new
command.
Offline surface scan supported.
Self-test supported.
Conveyance Self-test supported.
Selective Self-test supported.
SMART capabilities: (0x0003) Saves SMART data before entering
power-saving mode.
Supports SMART auto save timer.
Error logging capability: (0x01) Error logging supported.
General Purpose Logging supported.
Short self-test routine
recommended polling time: ( 1) minutes.
Extended self-test routine
recommended polling time: ( 236) minutes.
Conveyance self-test routine
recommended polling time: ( 2) minutes.
SCT capabilities: (0x103b) SCT Status supported.
SCT Error Recovery Control supported.
SCT Feature Control supported.
SCT Data Table supported.

SMART Attributes Data Structure revision number: 10
Vendor Specific SMART Attributes with Thresholds:
ID# ATTRIBUTE_NAME FLAG VALUE WORST THRESH TYPE UPDATED
WHEN_FAILED RAW_VALUE
1 Raw_Read_Error_Rate 0x000f 114 099 006 Pre-fail Always
- 61796058
3 Spin_Up_Time 0x0003 094 092 000 Pre-fail Always
- 0
4 Start_Stop_Count 0x0032 100 100 020 Old_age Always
- 46
5 Reallocated_Sector_Ct 0x0033 100 100 036 Pre-fail Always
- 1
7 Seek_Error_Rate 0x000f 045 045 030 Pre-fail Always
- 3848329867033
9 Power_On_Hours 0x0032 076 076 000 Old_age Always
- 21358
10 Spin_Retry_Count 0x0013 100 100 097 Pre-fail Always
- 0
12 Power_Cycle_Count 0x0032 100 100 020 Old_age Always
- 141
184 End-to-End_Error 0x0032 100 100 099 Old_age Always
- 0
187 Reported_Uncorrect 0x0032 016 016 000 Old_age Always
- 84
188 Command_Timeout 0x0032 100 100 000 Old_age Always
- 0
189 High_Fly_Writes 0x003a 100 100 000 Old_age Always
- 0
190 Airflow_Temperature_Cel 0x0022 063 049 045 Old_age Always
- 37 (Min/Max 37/37)
194 Temperature_Celsius 0x0022 037 051 000 Old_age Always
- 37 (0 23 0 0)
195 Hardware_ECC_Recovered 0x001a 018 014 000 Old_age Always
- 61796058
197 Current_Pending_Sector 0x0012 100 100 000 Old_age Always
- 2
198 Offline_Uncorrectable 0x0010 100 100 000 Old_age Offline -
2
199 UDMA_CRC_Error_Count 0x003e 200 200 000 Old_age Always
- 0

SMART Error Log Version: 1
ATA Error Count: 84 (device log contains only the most recent five errors)
CR = Command Register [HEX]
FR = Features Register [HEX]
SC = Sector Count Register [HEX]
SN = Sector Number Register [HEX]
CL = Cylinder Low Register [HEX]
CH = Cylinder High Register [HEX]
DH = Device/Head Register [HEX]
DC = Device Command Register [HEX]
ER = Error register [HEX]
ST = Status register [HEX]
Powered_Up_Time is measured from power on, and printed as
DDd+hh:mm:SS.sss where DD=days, hh=hours, mm=minutes,
SS=sec, and sss=millisec. It "wraps" after 49.710 days.

Error 84 occurred at disk power-on lifetime: 19483 hours (811 days + 19 hours)
When the command that caused the error occurred, the device was active or
idle.

After command completion occurred, registers were:
ER ST SC SN CL CH DH
-- -- -- -- -- -- --
40 51 00 ff ff ff 0f Error: UNC at LBA = 0x0fffffff = 268435455

Commands leading to the command that caused the error were:
CR FR SC SN CL CH DH DC Powered_Up_Time Command/Feature_Name
-- -- -- -- -- -- -- -- ---------------- --------------------
25 00 08 ff ff ff ef 00 11:02:43.659 READ DMA EXT
27 00 00 00 00 00 e0 00 11:02:43.658 READ NATIVE MAX ADDRESS EXT
ec 00 00 00 00 00 a0 00 11:02:43.638 IDENTIFY DEVICE
ef 03 46 00 00 00 a0 00 11:02:43.619 SET FEATURES [Set transfer mode]
27 00 00 00 00 00 e0 00 11:02:43.558 READ NATIVE MAX ADDRESS EXT

Error 83 occurred at disk power-on lifetime: 19483 hours (811 days + 19 hours)
When the command that caused the error occurred, the device was active or
idle.

After command completion occurred, registers were:
ER ST SC SN CL CH DH
-- -- -- -- -- -- --
40 51 00 ff ff ff 0f Error: UNC at LBA = 0x0fffffff = 268435455

Commands leading to the command that caused the error were:
CR FR SC SN CL CH DH DC Powered_Up_Time Command/Feature_Name
-- -- -- -- -- -- -- -- ---------------- --------------------
25 00 08 ff ff ff ef 00 11:02:40.589 READ DMA EXT
27 00 00 00 00 00 e0 00 11:02:40.588 READ NATIVE MAX ADDRESS EXT
ec 00 00 00 00 00 a0 00 11:02:40.568 IDENTIFY DEVICE
ef 03 46 00 00 00 a0 00 11:02:40.549 SET FEATURES [Set transfer mode]
27 00 00 00 00 00 e0 00 11:02:40.498 READ NATIVE MAX ADDRESS EXT

Error 82 occurred at disk power-on lifetime: 19483 hours (811 days + 19 hours)
When the command that caused the error occurred, the device was active or
idle.

After command completion occurred, registers were:
ER ST SC SN CL CH DH
-- -- -- -- -- -- --
40 51 00 ff ff ff 0f Error: UNC at LBA = 0x0fffffff = 268435455

Commands leading to the command that caused the error were:
CR FR SC SN CL CH DH DC Powered_Up_Time Command/Feature_Name
-- -- -- -- -- -- -- -- ---------------- --------------------
25 00 08 ff ff ff ef 00 11:02:37.539 READ DMA EXT
27 00 00 00 00 00 e0 00 11:02:37.538 READ NATIVE MAX ADDRESS EXT
ec 00 00 00 00 00 a0 00 11:02:37.518 IDENTIFY DEVICE
ef 03 46 00 00 00 a0 00 11:02:37.499 SET FEATURES [Set transfer mode]
27 00 00 00 00 00 e0 00 11:02:37.448 READ NATIVE MAX ADDRESS EXT

Error 81 occurred at disk power-on lifetime: 19483 hours (811 days + 19 hours)
When the command that caused the error occurred, the device was active or
idle.

After command completion occurred, registers were:
ER ST SC SN CL CH DH
-- -- -- -- -- -- --
40 51 00 ff ff ff 0f Error: UNC at LBA = 0x0fffffff = 268435455

Commands leading to the command that caused the error were:
CR FR SC SN CL CH DH DC Powered_Up_Time Command/Feature_Name
-- -- -- -- -- -- -- -- ---------------- --------------------
25 00 08 ff ff ff ef 00 11:02:34.459 READ DMA EXT
27 00 00 00 00 00 e0 00 11:02:34.458 READ NATIVE MAX ADDRESS EXT
ec 00 00 00 00 00 a0 00 11:02:34.438 IDENTIFY DEVICE
ef 03 46 00 00 00 a0 00 11:02:34.419 SET FEATURES [Set transfer mode]
27 00 00 00 00 00 e0 00 11:02:34.348 READ NATIVE MAX ADDRESS EXT

Error 80 occurred at disk power-on lifetime: 19483 hours (811 days + 19 hours)
When the command that caused the error occurred, the device was active or
idle.

After command completion occurred, registers were:
ER ST SC SN CL CH DH
-- -- -- -- -- -- --
40 51 00 ff ff ff 0f Error: UNC at LBA = 0x0fffffff = 268435455

Commands leading to the command that caused the error were:
CR FR SC SN CL CH DH DC Powered_Up_Time Command/Feature_Name
-- -- -- -- -- -- -- -- ---------------- --------------------
25 00 08 ff ff ff ef 00 11:02:31.369 READ DMA EXT
27 00 00 00 00 00 e0 00 11:02:31.368 READ NATIVE MAX ADDRESS EXT
ec 00 00 00 00 00 a0 00 11:02:31.348 IDENTIFY DEVICE
ef 03 46 00 00 00 a0 00 11:02:31.329 SET FEATURES [Set transfer mode]
27 00 00 00 00 00 e0 00 11:02:31.278 READ NATIVE MAX ADDRESS EXT

SMART Self-test log structure revision number 1
No self-tests have been logged. [To run self-tests, use: smartctl -t]

SMART Selective self-test log data structure revision number 1
SPAN MIN_LBA MAX_LBA CURRENT_TEST_STATUS
1 0 0 Not_testing
2 0 0 Not_testing
3 0 0 Not_testing
4 0 0 Not_testing
5 0 0 Not_testing
Selective self-test flags (0x0):
After scanning selected spans, do NOT read-scan remainder of disk.
If Selective self-test is pending on power-up, resume after 0 minute delay.

8<---

I'm not a hard disk guru, and this info doesn't really say anything specific to
me. Wrt the firmware issue, I went to the Seagate site and downloaded and
burned an ISO, rebooted and followed the instructions, but the firmware update
failed with "unexpected disk type" or something to that effect, although I'm
positively certain that I downloaded the correct ISO and the program initially
identified the disk correctly. The download page is
<http://seagate.custkb.com/seagate/crm/selfservice/search.jsp?DocId=207951&gt;
and the ISO is the bottommost, for ST31000340AS. I'm going to send a mail to
the Seagate support about it, but first I have to rerun the procedure and get
the exact messages.

regards, Leif.

On 10/06/2011 02:15 PM, Leif Biberg Kristensen wrote:

Model Family: Seagate Barracuda 7200.11 family
Device Model: ST31000340AS
Serial Number: 9QJ1ZMHY
Firmware Version: SD15

Oh, joy. I have some of those, and can confirm their data-eating powers.
Thankfully mine were in a backup server as part of a regularly verified
RAID array with ECC on the volumes, so I didn't lose any data, but it
was certainly frustrating.

The firmware updater is a right pain, because it only supports certain
SATA controllers and you have to boot into it. Grr.

smartctl is a *vital* tool; the more people know about it and how
awesome it is, the better.

--
Craig Ringer