Re: pg_dumpall failing from possible corrupted shared memory

Richard Broersma Jr <rabroersma(at)yahoo(dot)com> writes:
>> Are you interested in dumping out the page of pg_index that
>> contains this record, using pg_filedump?

> I've attached the results.log to the email.

Well, there's our smoking gun:

Item 11 -- Length: 398 Offset: 6844 (0x1abc) Flags: USED
XMIN: 940 CMIN: 0 XMAX: 0 CMAX|XVAC: 0
Block Id: 1 linp Index: 11 Attributes: 10 Size: 32
infomask: 0x0903 (HASNULL|HASVARWIDTH|XMIN_COMMITTED|XMAX_INVALID)
t_bits: [0]: 0xff [1]: 0x01

1abc: ac030000 00000000 00000000 00000000 ................
1acc: 00000100 0b000a00 030920ff 01000000 .......... .....
1adc: 5b380900 61410000 01000000 00000000 [8..aA..........
1aec: 1a000000 01000000 00000000 15000000 ................
1afc: 01000000 00000000 00000000 1c000000 ................
1b0c: 01000000 00000000 1a000000 01000000 ................
1b1c: 00000000 f7070000 26010000 287b4655 ........&...({FU
1b2c: 4e434558 5052203a 66756e63 69642032 NCEXPR :funcid 2
1b3c: 30323020 3a66756e 63726573 756c7474 020 :funcresultt
1b4c: 79706520 31313134 203a6675 6e637265 ype 1114 :funcre
1b5c: 74736574 2066616c 7365203a 66756e63 tset false :func
1b6c: 666f726d 61742030 203a6172 67732028 format 0 :args (
1b7c: 7b434f4e 5354203a 636f6e73 74747970 {CONST :consttyp
1b8c: 65203235 203a636f 6e73746c 656e202d e 25 :constlen -
1b9c: 31203a63 6f6e7374 62797661 6c206661 1 :constbyval fa
1bac: 6c736520 3a636f6e 73746973 6e756c6c lse :constisnull
1bbc: 2066616c 7365203a 636f6e73 7476616c false :constval
1bcc: 75652037 205b2037 20302030 20302031 ue 7 [ 7 0 0 0 1
1bdc: 30302039 37203132 31205d7d 207b5641 00 97 121 ]} {VA
1bec: 52203a76 61726e6f 2031203a 76617261 R :varno 1 :vara
1bfc: 74746e6f 00000000 00000000 1c000000 ttno............
^^^^^^^^^^^^^^^^^^^^^^^^^^
1c0c: 01000000 00000000 1a000000 6d6f6420 ............mod
^^^^^^^^^^^^^^^^^^^^^^^^^^
1c1c: 2d31203a 7661726c 6576656c 73757020 -1 :varlevelsup
1c2c: 30203a76 61726e6f 6f6c6420 31203a76 0 :varnoold 1 :v
1c3c: 61726f61 74746e6f 20317d29 7d29 aroattno 1})})

The underlined stuff has overwritten what should be expression dump
text.

Seeing that the corruption begins at page offset 1c00 --- that is,
exactly on a 1K boundary --- I'm a bit inclined to suspect a disk
I/O glitch. You might want to run some hardware diagnostics on
your machine.

regards, tom lane