Tom Lane wrote:
>Bruce Momjian <pgman@candle.pha.pa.us> writes:
>
>
>>Andrew Dunstan wrote:
>>
>>
>>>Bruce and others have suggested that PID is not sufficiently unique.
>>>
>>>
>
>
>
>>The nice things about using xid for session id is that is is unique for
>>a long time, rather than pid.
>>
>>
>
>Hmm. Now that I think about it, InitPostgres() always runs a
>transaction during backend startup. If we simply saved aside the XID
>assigned to that transaction, it would cost next to nothing to do and
>would provide a reasonably unique ID. However, this happens much later
>than the postmaster currently prints the log_connection message ...
>
>
>
Right. And if we have sessionids we would want them logged there, I
think. And that would rule out anything based on xid or backend pid.
OTOH I am quite sure I can implement the time based counter idea I
suggested earlier with a very small amount of disturbance and very low
cost. Unless someone can see a flaw in the scheme.
cheers
andrew