On Mon, Oct 19, 2015 at 9:37 AM, Bruce Momjian <bruce@momjian.us> wrote:
On Mon, Oct 19, 2015 at 09:26:21AM -0700, Jeff Janes wrote: > It seems like gdb eats signals that you send a process while it is being > debugged, so it is hard to figure out what is going on. From strace, it looks > like the children do receive a signal but either ignore it, or set a flag and > then ignore that. > > It doesn't continue to load the entire dump file, it exits once they complete > the current assignment and ask the parent for more work. > > Could just be a matter of adding the local equivalent of CHECK_FOR_INTERRUPTS > in the part of the code that spools COPY data to the backends? I'm not sure > what would happen if it were in the index/constraint building phase, I've never > let it get that far when it reported errors early on. > > (This is linux, sorry for not making that clear)
Well, we are not running COPY in pg_upgrade, just the DDL commands. Index creation is on empty tables, so it should be very quick. What should basically happen is that the pg_restore child processes should exit as forked children, and then the backends for these pg_restore proceses should then exit. My guess is that this problem is not pg_upgrade-specific as there is no signal control in pg_upgrade --- you are just getting the defaults.
(Updated TODO to mention Linux.)
Sorry, I don't know how I managed to screw this up so much. pg_restore, not pg_upgrade. I've never looked into pg_restore much until recently, so my fingers just autocomplete to the one I'm more used to. Let me go fix the TODO page. (I'm pretty sure pg_upgrade terminates itself very early on upon missing extensions)
