Re: repeatable system index corruption on 7.4.2

Tom Lane wrote:
> Joe Conway <mail(at)joeconway(dot)com> writes:
>
>>I'm seeing the following errors after a few hours of fairly aggressive
>>bulk load of a database running on Postgres 7.4.2:
>
>>cyspec=# select version();
>>ERROR: root page 43 of "pg_proc_proname_args_nsp_index" has level 0,
>>expected 2
>
> [ scratches head... ] Are you just doing bulk *data* load, or are you
> doing extensive schema hacking too? What reason would there be for
> index churn in pg_proc (or pg_class for that matter)?

Here's more detail about what we're doing:

Done once
----------
create the database
create schemas
create "base" tables
create functions

Done in a loop
--------------
for year in X to Y
{
create ~60 tables, inherited from the "base" tables
for each table (up to 6 at a time in parallel)
{
load data via COPY
create index
}
}

I was just informed that the most recent attempt failed again, this time
with no more than 1 table loaded at a time.

> Can you get a pg_filedump dump of that page of the index?

Will do.

BTW, I also tried reindexing one of the corrupt databases and got this:

backend> reindex database cyspec;
2004-08-19 11:13:37 NOTICE: table "pg_class" was reindexed
2004-08-19 11:13:38 ERROR: invalid page header in block 6 of relation
"pg_attribute"

Joe