"Luca Ferrari" <fluca1978@xxxxxxxxxxx> writes: >> If you really want to understand how snapshots work at this level you could >> read (slowly -- it's pretty dense stuff) through >> src/backend/utils/time/tqual.c:HeapTupleSatisfiesMVCC() > > I don't find it, I think you mean HeapTupleSatisfiesNow and > HeapTupleSatisfiesSnapshot. Sorry, HeapTupleSatisfiesSnapshot was renamed to HeapTupleSatisfiesMVCC, they're two different versions of the same function. > Just for confirmation: the relfrozenxid of a fresh table is the xid of the > transaction that created it, isn't it? Yes, easily enough checked: postgres=# create table xyz (i integer); CREATE TABLE postgres=# select xmin,relfrozenxid from pg_class where relname = 'xyz'; xmin | relfrozenxid ---------+-------------- 4971524 | 4971524 (1 row) But once you run vacuum that value will be changed. It represents the oldest transaction id which can occur in the table. Every time vacuum runs if that transaction is too old it will try to move it forward. -- Gregory Stark EnterpriseDB http://www.enterprisedb.com
