Rafael Martinez <r(dot)m(dot)guerrero(at)usit(dot)uio(dot)no> writes:
> If the behavior we are talking about is intentional, why do be have
> pg_rotate_logfile() if we can not manually override with this function
> log_rotation_age and log_rotation_size?
It works fine as long as the filename pattern is such that a new file
would be selected.
If we didn't have this filter, then a rotation operation would wipe out
*current* log entries, even those that were made a millisecond ago.
That can't possibly be a good idea. Any rotation scheme ought to
provide for recent entries to survive at least until you've rotated
through whichever other log files you have.
regards, tom lane