Re: Statistics visibility in SERIALIZABLE transactions

Josh Berkus wrote:
> Tom,
>
>   
>>  Try looking at a table that was ALTERed since
>> your transaction began.)
>>     
>
> Shouldn't the transaction fail, then?
>
>   
Neil Conway and I opened this VERY BIG can of worms a couple of years 
ago on hackers.

I can't find the references to it in the archives.  I think I need to go 
back more than the start of 2005.  And I can't remember how to do that.
It was in relation to CLUSTER on a table and how previous transactions 
would not use the correct view of the table.

I recall something to do with issues relating to reading the catalog 
before transaction startup and problems with needing a transaction to 
use the catalog.
My memory of it is fuzzy.  If Neil or Tom remember the discussion they 
may be able to post a link.

Sorry if this wasn't much help.

Russell Smith