Re: Duplicate rows sneaking in despite PRIMARY KEY / UNIQUE

Ian Barwick wrote:
> On 6/6/06, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Travis Cross <travis@crosswirecorp.com> writes:
>> > I'm noticing that a handful (4-16) of rows with duplicate columns
>> > (uid,token) are sneaking into the table every day despite the
>> > primary key constraint.
>>
>> Corrupt index, looks like ... you might try reindexing the index.
>>
>> I don't believe that the PANIC you show has anything directly to do
>> with duplicate entries.  It is a symptom of corrupt index structure.
>> Now a corrupt index might also explain failure to notice duplications,
>> but changing your application isn't going to fix whatever is causing
>> it.  You need to look for server-side causes.
>>
>> Any database or system crashes on this server (before this problem
>> started)?  Do you *know* that the disk drive will not lie about write
>> complete?  What is the platform and storage system, anyway?
> 
> FWIW I've seen similar behaviour to this (PostgreSQL processes exiting
> "abnormally", index corruption with duplicate primary keys) on servers
> with defective RAM chips.

That's a good thought, and also a possibility.  I tend to distrust 
RAM inherently.  If nothing else seems verifiable as the cause, I 
may have to take the system down on a Saturday night for a good 
thorough run through memtest86.

Cheers,

-- Travis