Hi, On 2018-05-22 21:43:01 +0200, Paolo Crosato wrote: > > Could you report the result of > > select ctid, xmin, xmax from pg_authid ; > > > > This is the result: > > postgres=# select ctid, xmin, xmax from pg_authid ; > (0,16) | 3031994631 | 0 > 16 | 6496 | 1 | 144 | 3031994631 | 0 | 0 | (0,16) > | 32779 | 10507 | 32 | 1111111111000000 | 675851 | > \x6e6167696f73000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000001000001000000ffffffff496d64356333633236616163636439616665346437383061396239613464663634653639 > postgres=# select relfrozenxid from pg_class where relname='pg_authid'; > relfrozenxid > -------------- > 400011439 That tuple looks, to me, like it indeed shouldn't exist, given the relfrozenxid. Decoding infomask (10507 / 0x290B), if I did it right, shows: HEAP_HASNULL HEAP_HASVARWIDTH HEAP_HASOID HEAP_XMIN_COMMITTED HEAP_XMAX_INVALID HEAP_UPDATED so it's not frozen. That suggests there's either a bug, or you had corruption in your cluster. Could you give a bit of "history" about that postgres instance? What version of postgres was it run on earliest? Was it ever pg_upgraded? Were there any OS crashes? Other hardware issues? Was the cluster ever used with fsync=off or full_page_writes=off? Greetings, Andres Freund
