Обсуждение: (pgsql8.4) DATA Corruption

On Fri, Aug 19, 2011 at 1:08 PM, Mikola Rose <mrose@power-soft.com> wrote:
> Happy Friday people!
>
>
>
> I was wondering if anyone had any suggestions on how to resolve this
> issue...  I am moving otrs to another server and during the backup process I
> am running into this error
>
> pg_dump: dumping contents of table article_attachment
>
>
>
>
>
> pg_dump: SQL command failed
>
> pg_dump: Error message from server: ERROR:  unexpected chunk number 7
> (expected 6) for toast value 77281 in pg_toast_57366
>
>
>
>
>
> pg_dump: The command was: COPY public.article_attachment (id, article_id,
> filename, content_size, content_type, content, create_time, create_by,
> change_time, change_by, content_id, content_alternative) TO stdout;
>
>
>
>
>
> pg_dump: *** aborted because of error
>
>
>
>
>
> I have tried a vacuum and reindex with no successes.

You'll likely have to figure which blocks are corrupted, and copy out
the good data using a where clause the excludes it, then get what you
can out of it, truncate the table, and reinsert the data.

Then figure out what part of your hardware is / might be dodgy.  mem
test, disk check, etc.

Thanks for the suggestions Scott.

On 20/08/2011 4:27 AM, Mik Rose wrote:

Thanks for the suggestions Scott.

I was looking at the zero_damage_page option and enabled it in my postgres.conf.  I saw that it detected bad pages and zero'd them out during a number of reindex and a vacuum's.

Now when i reindex though,  I am getting this error.

NOTICE:  table "pg_foreign_server" was reindexedERROR:  could not create unique index "pg_toast_57366_index"

DETAIL:  Table contains duplicated values.

ERROR:  could not create unique index "pg_toast_57366_index"

DETAIL:  Table contains duplicated values.

You'll have to figure out what what values are duplicates within each table and DELETE one of them after determining which is the correct one - if you can. Then you should be able to re-create the index. If the DELETE fails you might find you have to DROP the index, do the DELETE(s), then CREATE the index again.

If the index is a partial index (ie it has a WHERE clause) you need to find duplicates that match that WHERE clause, you can ignore ones that don't match.

I hope you took a file-system-level backup of your database before starting to mess with it. People here will normally advise you to do that first and I'm kind of surprised nobody did.

--
Craig Ringer