Thread: XID comparations

Hi,

I would like to understand better the logic to determine when a xid is older
than another one.

As I could understand, the XID is always incremented, never reset. If it is
true, then we can have rows with cmin ranging
from 1 to 4.294.967.295 (2^32-1).

When xid overflows (32 bits) the next one will be 3 (1 and 2 are reserved).

In this case, we could have have lines with cmin 4.294.967.295 and lines
with cmin 3. How are they compared to determine that
rows with cmin 3 are newer than rows with cmin 4.294.967.295?

Thanks in advance,

Reimer

"Carlos H. Reimer" <carlosreimer@terra.com.br> writes:
> I would like to understand better the logic to determine when a xid is older
> than another one.

It's circular mod 2^32, with a special case for FrozenXID.  It's a
mistake to imagine that XIDs are unsigned ints, really --- the
comparison doesn't work that way.  For an XID of say 1billion, XIDs from
1billion to 3billion are "after", the rest "before".  So once a row
is created, it has to be deleted or frozen within 2 billion
transactions, else its XID wraps around and appears to be "in the
future" rather than "in the past" compared to current XIDs.

            regards, tom lane

On Tue, Jun 13, 2006 at 12:10:26PM -0300, Carlos H. Reimer wrote:
> When xid overflows (32 bits) the next one will be 3 (1 and 2 are reserved).
>
> In this case, we could have have lines with cmin 4.294.967.295 and lines
> with cmin 3. How are they compared to determine that
> rows with cmin 3 are newer than rows with cmin 4.294.967.295?

The same way you handle any circular numbering system, compare the
difference. i.e.

if( (xmin1-xmin2) mod (2^32) < 2^31 )
{
  xmin1 is newer than xmin2
}

In the example you give:

(3 - 4.294.967.295) mod (2^32)
= -4294967292 mod (2^32)
= 4

Hence XID 3 is newer compared to XID 4.294.967.295.

Note in such a circular system, it is possible for A < B, B < C and C <
A to all be simultaneously true. However, we always know where the
current transaction is, so everything is compared to that.

I hope this clears it up for you.

Have a nice day,
--
Martijn van Oosterhout   <kleptog@svana.org>   http://svana.org/kleptog/
> From each according to his ability. To each according to his ability to litigate.

Thanks,

In my first question I would like to use xmin instead of cmin, even so I
could understand the logic.

Then for each XID you have 2 bilions XIDs that are considered lower than and
the other 2 bi higher than.

About row visibility: are all the rows with xmin higher than my XID be
considered in the future and not visible to my XID?

Thanks in advance!

Reimer