Kurt Overberg <kurt@hotdogrecords.com> writes:
> Okay, so I did some digging with pg_filedump, and found the following:
> Block 406 ********************************************************
> Item 1 -- Length: 472 Offset: 7720 (0x1e28) Flags: USED
> XMIN: 1489323584 CMIN: 1 XMAX: 0 CMAX|XVAC: 0
> Block Id: 406 linp Index: 1 Attributes: 6 Size: 32
> infomask: 0x0912 (HASVARWIDTH|HASOID|XMIN_COMMITTED|XMAX_INVALID)
This is pretty much what you'd expect for a never-updated tuple...
> mydb=# select * from _my_cluster.sl_log_1 where ctid = '(406,1)';
> log_origin | log_xid | log_tableid | log_actionseq | log_cmdtype |
> log_cmddata
> ------------+---------+-------------+---------------+-------------
> +-------------
> (0 rows)
so I have to conclude that you've got a wraparound problem. What is the
current XID counter? (pg_controldata will give you that, along with a
lot of other junk.) It might also be interesting to take a look at
"ls -l $PGDATA/pg_clog"; the mod times on the files in there would give
us an idea how fast XIDs are being consumed.
regards, tom lane
