Re: Invalid page header in pg_class

Hi Tom,

Enabling the zero_damaged_pages solved the problem. I
am in the process of dumping & restoring.

Thanks for the help.
Gokul.
--- Tom Lane <tgl@sss.pgh.pa.us> wrote:

gokulnathbabu manoharan <gokulnathbabu@yahoo.com>
writes:

In my sample databases the relfilenode for

pg_class

was 1259. So I checked the block number 190805 of

the

1259 file. Since the block size is 8K, 1259 was

in

two files 1259 & 1259.1. The block number 190805
falls in the second file whose block number is
58733((190805 - (1G/8K)) = 58733).

You've got a pg_class catalog exceeding a gigabyte??
Apparently you've been exceedingly lax about
vacuuming.
You need to do something about that, because it's
surely
hurting performance.

You did the math wrong --- the damaged block would
be 59733, not
58733, which is why pg_filedump isn't noticing
anything wrong here.

It seems almost certain that there are only dead
rows in the
damaged block, so it'd be sufficient to zero out the
block,
either manually with dd or by turning on
zero_damaged_pages.
After that I'd recommend a dump, initdb, reload,
since there may
be other damage you don't know about.

regards, tom lane

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com