Re: transaction safety

On Mon, Feb 12, 2001 at 10:22:30AM -0500, Tom Lane wrote:
> DaVinci <bombadil(at)wanadoo(dot)es> writes:
> > On Mon, Feb 12, 2001 at 01:08:01PM -0000, Michael Ansley wrote:
> >> Typically, the insert for a person, and for all the associated addresses
> >> would be done in the same transaction so that if the insert for one of the
> >> addresses failed, then the whole lot would role back (perhaps a bit extreme,
> >> but I think that's what you asked for ;-)
>
> > I thought it is possible to have different transactions opened and insert
> > data in same table from them. It seems my idea was fault, doesn't it?.
> > In sumary: locks with inserts are for table and not for tuple. If this is
> > not true, tell me details, please :)
>
> It's not true. How did you arrive at that conclusion from what Mike
> said?

I'll try to explain. Mike said: "in a transaction make an insert and then a
read in serial current value".

If in gap between those operations occurs another insert from different
transaction, then reading serial is not safe.

In order to understand this well I have made some basic experiments,
freezing a transaction with an insert and making other transaction with an
insert to the same table. Second gets frozen until first commit or cancel.
That is reason of my last message: "locks with inserts are for table and
not for tuple". Perhaps I didn't explain myself very well or there is some
detail about locks that I don't understand at all.

Thanks all for your time.

David