Re: Disk file operations I/O

From: kyle Hailey <kylelf_at_gmail.com>
Date: Fri, 20 Jan 2012 10:47:55 -0800
Message-ID: <CADsdiQjeLTqOYWaJMOeh=ObrQEqPseCR77b5WWcpeqepZMuv1g_at_mail.gmail.com>

Thanks for pointing out mnt_timeout. The database AFAIK is not on dNFS, so not sure if mnt_timeout applies. Doc says default value is 10 minutes which is higher than the actual wait, so something else is probably going on, though it might be better to set mnt_timeout low, to actually see if a timeout occurs. A timeout might be handled better than the current situation.
According to Oracle support, on this version, the file functions are:

1 file creation
2 file open
3 file resize
4 file deletion
5 file close
6 wait for all aio requests to finish
7 write verification
8 wait for miscellaneous io (ftp, block dump, passwd file) 9 read from snapshot files

Though the file types should match from DBA_HIST_IOSTAT_FILETYPE_NAME thus the issues is

     file open on data files

The data look suspiciously like the bug, though the bug is suppose to be fixed in 11.2.0.3 patch set (which I'm unclear if that is just 11.2.0.3 or a patch on top fo 11.2.0.3 )

Yes, looks like it might be time for some tracing which will require some script automation.

• Kyle

On Fri, Jan 20, 2012 at 8:34 AM, Yong Huang <yong321_at_yahoo.com> wrote:

> > Jonathan pointed out that there is a bug with "Disk file operations
> > I/O" with dNFS.
> > The database isn't on dNFS but is using NFS
>
> If you can capture the session waiting on the event, can you trace the
> shadow process to see what system call it's making? Is it open()? pstack
> on the process is helpful too.
>
> The bug you referenced also has a description in Bug 13245023. If your
> pstack at the moment shows the same call stack, your process is waiting
> for up to mnt_timeout seconds to open the file on the NFS filesystem. But
> the bug causes it to loop until counter overflow. mnt_timeout is specified
> in $ORACLE_HOME/dbs/oranfstab. But I doubt playing with that number helps
> due to the bug. While waiting for the bug fix, maybe you can have a
> frequent cron job that opens and closes each file on the NFS just to make
> sure their inode info is hot in cache?
>
> Yong Huang
>

--
http://www.freelists.org/webpage/oracle-l

Received on Fri Jan 20 2012 - 12:47:55 CST