Sam Barnett-Cormack <s.barnett-cormack@lancaster.ac.uk> writes:
> [ accidentally clobbered pg_control ]
It sounds like the main problem is too small a value in the XID
(transaction ID) counter, so that all your data tuples appear to be in
the future. pg_resetxlog has an option to force an initial XID but
you need to use it. You can guess a suitable XID by looking at the
names of the old clog files --- you want something that will be past
the end of the newest clog file, keeping in mind that there are 1M
(1024^2) XIDs per clog file and that the clog file names are in hex.
(I am not sure, but you might also need to provide a zero-filled clog
file spanning the point you start XID at.)
You will also need to be sure that the WAL file position
is greater than the end of the old WAL --- again, look at the existing
WAL files to see the largest number, and pick something bigger.
BTW, make real sure that pg_control is set up with the same locale
values as you were using before, or you'll be in a world of hurt
with corrupt indexes. Use pg_controldata to look at what you have put
into pg_control. (If you're not sure what was there before, get an
old copy of pg_control off your last backup tape ... you do have a
filesystem backup someplace, I trust ...)
> I still have the old
> /data/base dir, all that has been altered in there (recent mtime) is
> base/1175459/pg_internal.init
You can and probably should blow away pg_internal.init --- it's just a
cache file and can be rebuilt. I'd not trust it at this point.
regards, tom lane