Thanks to Tom & Adrian, here’s what happened (my version was 9.1, sorry I forgot to mention it)
1) 10 hours after my email, the VACUUM had used up about 3.5TB but had stopped using up more disk space, it was now
“simply”reading data from the file system
2) I attempted to interrupt using CTRL-D to no avail so I interrupted with CTRL-C. That stopped it with a clean message
(butdid not relinquish filesystem space)
3) I exited the backend successfully using CTRL-D and relaunched it with the additional “–r” command line argument
4) I ran the query to see which tables were the “oldest” and did not recognize the ones before I started the vacuuming
(encouraging!)
5) I ran CHECKPOINT on the backend and got all the disk space back
6) I realized that the message regarding wraparound was no longer an ERROR but a WARNING so I was able to restart
postgres“normally”
7) I ran a query based on my previous query to build VACUUM VERBOSE commands on the tables with the oldest transaction
idsand wrote it to a text file and then execute that file, I now have tens of millions of transactions back and can
restartmy server. I’ll do the rest of the VACUUM maintenance during low-load periods.
Cheers
Philippe