On 6/10/05, Ignizio, Richard <richard.ignizio@xxxxxxxxxx> wrote: > I agree with Tom, we are on a Sun UNIX box and we move the alert.log (25+= databases) every week without any issues and have been doing it for the la= st 5 years. >=20 > Rich I agree it can be done without risks provided the Unix oracle user is allowed to have enough inodes opened simultaneously (and in real environment it's always the case, yes). But it is not clean: Since the instance is not shut down after the "mv", its processes still hold the old file open. When they need to write an alert, it is written to the new alertSID.log but they hold both files open. As for a "rm" instead of "cp"+"mv", it also leaves the inode open by the instance: $ lsof|grep 368024 oracle 10561 oracle 6w REG 254,8 7300 368024 /usr/oracle/admin/TEST/bdump/alert_TEST.log (deleted) oracle 10561 oracle 7w REG 254,8 7300 368024 /usr/oracle/admin/TEST/bdump/alert_TEST.log (deleted) oracle 10563 oracle 6w REG 254,8 7300 368024 /usr/oracle/admin/TEST/bdump/alert_TEST.log (deleted) oracle 10563 oracle 7w REG 254,8 7300 368024 /usr/oracle/admin/TEST/bdump/alert_TEST.log (deleted) ... Definitely the most simple and neat solution is "cp" and then ">alertSID.lo= g". Jerome -- //www.freelists.org/webpage/oracle-l