"Tom Lane" <tgl@xxxxxxxxxxxxx> writes: > "Steven Flatt" <steven.flatt@xxxxxxxxx> writes: >> So, can we simply trust what's in pg_class.relpages and ignore looking >> directly at the index? > > No, we can't. In the light of morning I remember more about the reason > for the aforesaid patch: it's actually unsafe to read the pg_class row > at all if you have not got lock on the index. We are reading with > SnapshotNow in order to be sure we see up-to-date info, and that means > that a concurrent update of the row (eg, for REINDEX to report the new > relfilenode) can have the following behavior: Should reindex be doing an in-place update? Don't we have to do in-place updates for other system catalogs which are read in snapshotnow for precisely the same reasons? Alternatively, why does the planner need access to the pg_class entry and not just the pg_index record? -- Gregory Stark EnterpriseDB http://www.enterprisedb.com ---------------------------(end of broadcast)--------------------------- TIP 9: In versions below 8.0, the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match
