William Temperley <willtemperley(at)gmail(dot)com> writes:
> I've got two transactions I tried to kill 3 days ago using "select
> pg_cancel_backend(<pid>)", then SIGTERM, and have since then been
> using 100% of a cpu core each. They were supposed to insert the
> results of large unions with PostGIS and appear to have failed.
> Could someone tell me what's the least worst option here please? If I
> kill -9 will I corrupt my data directory?
No, you'll just lose all your open sessions.
It might be worth trying to identify where they're looping before
you zap them, though. A stack trace from gdb would help.
regards, tom lane