[Sunhelp] Should I ever reboot a Solaris box if I don't need to?

[Tim Conrad]

This was posted to the Sun-Managers  mailing list a few days ago...

From: "Thomas Carter" <TCarter at memc.com>
Date:  Fri, 1 Sep 2000 11:07:44 -0500
Subject: Update FYI: If you're running Oracle 8 Databases

Upon further investigation, I believe this also applies to Oracle 7.x.

The time below is actually  2147483647 1/100s of a second (sorry for the
confusion).

Thomas Carter
MEMC Southwest

Original Post:

I thought I'd comunicate a bug we recently ran into, since it wasn't
communicated to us until after it bit us. We're running Solaris 2.6 with
Oracle 8.0.5, but I believe it's a problem with all versions of Oracle <
8.1.7. When a Sun server has been up for 248 days (doesn't matter how long

the Oracle processes have been running), problems start cropping up with
the database. The problem is Oracle uses the times(2) system call in
various places, and from the times(2) man page:

>The return value may overflow the  possible  range  of  type clock_t.  If

times() fails, (clock_t)-1 is returned
> and errno is set to indicate the error.

clock_t is defined as a signed long, which is 32-bit on Solaris 2.6 (can't

speak for 7 or 8), thus allowing a max value of 2147483647 seconds (~248
days) before wrapping around to negative. The Oracle software does not
handle this, causing the problems.

You may reference SRDB #23550 on SunSolve, as well as Oracle bug report
1084273.

Just a heads-up,
Thomas Carter
MEMC Southwest

This could be part of what your DBA is talking about. I only mention this
because of the reference to Oracle. Other than that, very few things are
ever fixed by rebooting a Unix box. If something is broke, 99.9% of the
time (At least in my experience) rebooting the box will only result in the
same problem occuring on reboot. Typically, the things that need reboot
have something to do with failed (or partially failed) hardware, or some
deeply-rooted kernel issue that, again, has to do with funky hardware.

Moo!

Tim

Hal Flynn wrote:

> Short answer.  No
>
> Long answer.  Hell no.
>
> He's a dba because he couldn't cut being a real SA (offense intended).
> rm -rf $USERID is the best recourse of action.
>
> More seriously, no.  It should run fine.  The only time you should
> really need to reboot a Sun server is when adding new hardware.
>
> On a side note, I haven't known many NT servers with an uptime of 3-4
> months.  Just out of curiousity, anybody else know of any?
>
> Hal
>
> "Adams, Christopher" wrote:
>
> >
> >
> > I have a E 450 UltraSparc-II with 2 gig of ram and 6 18 gig Segate
> > Cheetah drives (non-raid).  This 450 is "only" running an Oracle
> > database (8.1.5) and has been running with no problems for 6 months
> > straight without being rebooted.  This is a production server and has
> > mission critical data on it.  Last weekend our Oracle DBA made a call
> > to "reboot" the server because he heard through the grapevine that
> > "rebooting" a Solaris server (even if there are no problems or
> > configuration changes necessary for this) every 3 to 4 months is a
> > good thing and should be done.  My question is, "Is this true? Should
> > you reboot a running server that doesn't need to be rebooted for any
> > administrative purposes every 3 to 4 months just for good measure?".
> > I think he is getting confused with Microsoft (no offense of course)
> > NT Server.
> >
> > Please Fire away with any advice you can give...
> >
> > Thanks in advance..
> >
> >
> > Christopher A.