On 10/27/06, Alvaro Herrera <alvherre@xxxxxxxxxxxxxxxxx> wrote:
> Do they vacuum enough? I have seen problems with PostgreSQL (albeit not > since 7.3) where a unique constraint would not enforce because of index > bloat. Huh?? This would qualify as a serious bug. Failure to vacuum should bring performance loss, but not functionality loss (modulo the Xid wraparound issue).
right, i think he was talking about the wraparound issue. definately does not apply here.
I do remember vaguely the failure Merlin alludes to, and IIRC it has been reported a couple of times by other people but has never been resolved because it was awfully difficult to reproduce. Maybe it has something to do with the btree bug that Tom diagnosed on Wednesday? The uniqueness-checking code is ... weird.
I'm hoping this is the case. When 8.2 comes out I'm going to upgrade their servers to that version and hope for the best.
I guess if it was the same bug, you could not vacuum the table, which I assume you do regularly.
right. Since these are gererally not 24 hour operations, vacuum is performed regularly on a schedule. Also, I am going to implement a sweep which checks each table for duplicates on each constraint. Since this is a converted ISAM system, the query volume is enormous but the data turnover is not. Pessimistic locks are enforced with the userlock module. Statements executed over ExecPrepared 100% of the time. merlin
