Re: page is uninitialized --- fixing

Brad Nicholson <bnichols@ca.afilias.info> writes:
> We just took a test database down (PG 8.1.11) fairly hard (pulled a SAN
> switch out while it was under load), which caused the DB to crash.  It
> started up fine, when I vacuumed the DB, I saw the following messages.

> WARNING:  relation "my_table" page 652139 is uninitialized --- fixing
> WARNING:  relation "my_table" page 652140 is uninitialized --- fixing
> WARNING:  relation "my_table" page 652940 is uninitialized --- fixing
> WARNING:  relation "my_table" page 652941 is uninitialized --- fixing

> That sort of looks like it could be data loss, can someone explain what
> happened?

It could be that but not necessarily.  These could be pages that were
allocated to put new tuples into, but the crash happened before the
inserting transaction committed (and, in fact, before any of the
associated WAL entries had made their way to disk -- else the empty
pages would've been initialized during WAL replay).

It would be easier to believe that if the uninitialized pages were all
contiguous though.  Do you know that this table was under heavy insert
load at the time?

            regards, tom lane