Actually I found these options given to us by the Oracle ZFSSA or DNFS team: rw,bg,hard,tcp,nfsvers=3,timeo=600,rsize=1048576,wsize=1048576,nointr,noatime Trying those first. Not sure why we hadn't propagated those to the other systems by now. Don. On Wed, Oct 8, 2014 at 1:43 PM, Don Seiler <don@xxxxxxxxx> wrote: > I did forget to mention that this is on 11.2.0.3. > > Also this problem only seems to occur when starting our STAGING database > which is on NFS (direct NFS) on the ZFSSA. With our DEV database on ASM we > don't see this problem. > > I just did a long overdue review of our NFS mount options. They were: > > defaults,rw,noatime,bg,hard,tcp,nfsvers=3,timeo=600,rsize=1048576,wsize=1048576,noacl > > I changed them to: > > rw,bg,hard,nointr,tcp,vers=3,timeo=600,rsize=1048576,wsize=1048576,actimeo=0 > > We'll see how it goes this weekend. > > On Tue, Oct 7, 2014 at 9:21 PM, Don Seiler <don@xxxxxxxxx> wrote: > >> It is creating the flashback log as part of the mounting process. Here is >> the alert log snippet that I'm basing this on: >> >> Sun Sep 21 09:40:38 2014 >> Successful mount of redo thread 3, with mount id 4175062721 >> Sun Sep 21 10:23:02 2014 >> Allocated 331130296 bytes in shared pool for flashback generation buffer >> Starting background process RVWR >> Sun Sep 21 10:23:02 2014 >> RVWR started with pid=28, OS id=69631 >> Database mounted in Exclusive Mode >> Lost write protection disabled >> Completed: alter database mount >> Sun Sep 21 10:23:15 2014 >> alter database flashback off >> Stopping background process RVWR >> Deleted Oracle managed file >> /mnt/foo/fra/FOO/flashback/o1_mf__1411306838866824_.flb >> Flashback Database Disabled >> Completed: alter database flashback off >> >> You can see the 43 minute delay between the mount of the last redo thread >> and the flashback generation buffer. During that time I see the flashback >> .flb log file being written to on the NFS mount. When that finishes, the >> database is mounted and flashback is immediately disabled (per the script). >> >> Don. >> >> >> On Tue, Oct 7, 2014 at 9:09 PM, Seth Miller <sethmiller.sm@xxxxxxxxx> >> wrote: >> >>> Don, >>> >>> Are you saying that the instance is creating a flashback log before it >>> is mounted? >>> >>> How did you come to the conclusion that it is the log creation causing >>> the wait? >>> >>> Seth Miller >>> >>> >>> On Tuesday, October 7, 2014, Don Seiler <don@xxxxxxxxx> wrote: >>> >>>> Good evening, >>>> >>>> Part of my staging/dev refresh process involves taking an image copy >>>> backup of prod datafiles and controlfiles and firing up a new instance from >>>> them. The first thing I do when mounting the new database is to turn off >>>> flashback. However recently we're seeing that the time to mount takes >>>> almost 40 minutes while the instance creates a new flashback log. Seems >>>> silly to have to wait for this when we plan to disable it right away >>>> anyway. >>>> >>>> Is there a way to disable flashback database without having to mount? >>>> I'm guessing the worst case scenario is having to create a trace of the >>>> controlfile and create a new one with flashback disabled. >>>> >>>> Don. >>>> >>>> -- >>>> Don Seiler >>>> http://www.seiler.us >>>> >>> >> >> >> -- >> Don Seiler >> http://www.seiler.us >> > > > > -- > Don Seiler > http://www.seiler.us > -- Don Seiler http://www.seiler.us