Re: Detecting corrupt table

[Willy-Bas Loos <willybas@xxxxxxxxx>]

you might want to log any errors resulting from  pg_dump and then grep through them to verify. or you could record the exit status ( $? ) for each pg_dump command.

I was also thinking about how to check if something malformed your data on disk. I could think of some ways to do that, but it doesn't look like you are looking for that.

hth

WBL

On Wed, Apr 18, 2012 at 12:05 AM, David Larochelle <dlarochelle@xxxxxxxxxxxxxxxxxxxxx> wrote:

Our database has some corrupt tables and I'm trying to figure out what data can be salvaged and what needs to be restored from backup or regenerated.

Initially I tried running select count(*) on all user tables. While this did detect some corrupt tables, it missed others. For example, I was able to run count(*) on a table but then got an error while trying to back it up.

pg_dump: Error message from server: ERROR:  missing chunk number 0 for toast value 368243665 in pg_toast_284730161

pg_dump: The command was: COPY public.stories (stories_id, media_id, url, guid, title, description, publish_date, collect_date, story_texts_id, full_text_rss) TO stdout;

Is there a simple way to determine which parts of the database are corrupt? I'm currently running a script to back up each table individually using something like the following:

psql -c "select tablename from pg_tables where tableowner = 'db_user' ORDER by tablename " | tail -n +3 | head -n -2 | xargs -n 1  -i pg_dump --verbose --table={} --file={}_.dump

But I'm worried that this approach will also miss database corruption and was wondering if anyone has other suggestions.

Thanks,

David

-- 
"Patriotism is the conviction that your country is superior to all others because you were born in it." -- George Bernard Shaw