Marc Cousin <mcousin(at)sigma(dot)fr> writes:
> [ can't cancel this query ]
> Index Scan using file_jpfid_idx on file (cost=0.00..7327212.07 rows=2128
> width=104)
> Index Cond: (pathid = 120)
> "file_jpfid_idx" btree (jobid, pathid, filenameid)
Hmm ... is it likely that index entries with pathid = 120 are *very* few
and far between in jobid order? It looks like we have no
CHECK_FOR_INTERRUPTS calls inside the loop in _bt_next(), which is
probably a mistake ...
regards, tom lane