edward(at)heroku(dot)com writes:
> We have this thing where we kill the restore command sometimes, to ensure
> it's not stuck.
Um, what makes you think that's a good idea?
> This has rarely led to postmaster dying afterwards instead of retrying.
"Rarely"? As I read the code, it will happen every single time.
Failure of the restore process is a failure, we do not retry.
(The general theory is that if you think you need retries, you
should code the restore_command to handle that.)
regards, tom lane