Thread: Database corruption caused by long insert strings

Database corruption caused by long insert strings

From
andrew@mycomport.com
Date:
The system: Linux RedHat 6.2, running postgresql 6.5.3 (installed from the
RH 6.2 RPM)

The bug: attempts to insert large strings result in database
corruption.  Not only does the attempted insert fail with a "too
large" error, but the table is corrupted, and further accesses can cause
the system to crash.

I originally discovered this bug while accessing the database using Perl,
but managed to replicate the problem using the command-line tools only.

Test case:
Create a database, connect to it using psql.

testdb=>CREATE TABLE test (testcol text);
CREATE
testdb=>INSERT INTO test (testcol) values (<a really long string literal>);
ERROR:  Tuple is too big: size 10584

At this point, the database is corrupt.  "Vacuum"ing the database will
detect and fix the corruption, but further queries without vacuuming may
cause the system to crash (the postmaster daemon consumes all available
CPU cycles).  I have not determined exactly how to trigger this.


Thanks for an (otherwise) excellent product :-)

Andrew Bishop

Re: Database corruption caused by long insert strings

From
Peter Eisentraut
Date:
On Wed, 3 May 2000 andrew@mycomport.com wrote:

> The bug: attempts to insert large strings result in database
> corruption.  Not only does the attempted insert fail with a "too
> large" error, but the table is corrupted, and further accesses can cause
> the system to crash.

That's a known and fixed problem. Feel free to try our new product 7.0 --
should be out on Monday last I heard.


--
Peter Eisentraut                  Sernanders väg 10:115
peter_e@gmx.net                   75262 Uppsala
http://yi.org/peter-e/            Sweden