Hello,
Most excellent!!!
Thank you both Jorge and Tom. The script worked correctly after Toms suggestion and gave me the correct id of the
brokenrecord.
After removing this record from the table, reindexing both pg_toast and the broken table, then vacuuming it – table is
workingnow as expected.
Now backups and selects work again.
Thank you and have a great day,
Maris.
----- Original Message -----
From: "Tom Lane" <tgl@sss.pgh.pa.us>
To: "Maris Jansons" <maris@lailio.net>
Cc: "Jorge Torralba" <jorge.torralba@gmail.com>, "pgsql-admin" <pgsql-admin@postgresql.org>
Sent: Thursday, January 26, 2017 6:03:03 PM
Subject: Re: [ADMIN] Help: unfixable/undead error - missing chunk number 0 for toast value 413511 in pg_toast_25477
Maris Jansons <maris@lailio.net> writes:
> I have identified the problematic column, in my case it is api_response, my version of script:
> DO $f$
> DECLARE
> baddata TEXT;
> badid INT;
> BEGIN
> FOR badid IN SELECT id FROM api_logs LOOP
> BEGIN
> SELECT api_response
> INTO baddata
> FROM api_logs where id = badid;
> EXCEPTION
> WHEN OTHERS THEN
> RAISE NOTICE 'Data for ID % is corrupt', badid;
> CONTINUE;
> END;
> END LOOP;
> END;
> $f$
> However this script finishes without errors, but if I run simple
>> select api_response from api_logs;
> I get:
> ERROR: missing chunk number 0 for toast value 413511 in pg_toast_25477
> Maybe I have messed something up in the script?
Oh ... what is happening, I think, is that the toast pointers are getting
stored into the "baddata" local variable as-is. The problem would only
get detected if the system has to fetch the toasted value, and this code
doesn't make it do so.
You could do something that requires fetching the value, perhaps
SELECT api_response || '' INTO baddata ...
regards, tom lane
