[Discuss] Issuing the 'sync' command more than once (and a tangent on how not to run a high-tech company)

[gaf at blu.org (Jerry Feldman)]

On 06/19/2012 03:11 PM, Derek Martin wrote:
> On Sat, Jun 16, 2012 at 12:59:44PM -0400, MBR wrote:
>> On 6/12/2012 11:22 PM, Jack Coats wrote:
>>> In old SunOS days, we could issue the 'sync' command, twice, to ensure
>>> all system
>>> buffers had been written to disk.  You could experiment to see if
>>> issuing it occasionally
>>> in your script helps.  Or issue it outside the script, even in a chron
>>> might help.
>> Actually, calling 'sync' multiple times from a script really won't
>> help.  To the best of my knowledge, no Unix kernel has ever
>> contained code that counts the number of times sync() (the system
>> call that the 'sync' command issues) has been called.  
> The reason I was taught to do this differs from what you put forth,
> and regardless it's certainly true that no modern Unix should ever
> require a user to run sync manually, except possibly in very rare
> circumstances.
>
> I don't claim to know the veracity of this, but I was taught (by a
> college professor who taught Unix system adminsistration as a course,
> for whatever that's worth) that the reason to sync twice (not three
> times) is that, as you say, the first call to sync schedules the
> kernel to sync the buffers, but does not necessarily complete before
> the system call returns; however (as I was told) a SUBSEQUENT call to
> the sync() system call would block until any previously scheduled sync
> had completed.  Thus, the completion of the SECOND sync command
> guarantees that the FIRST sync completed flushing the buffers to disk.
>
> Now, I certainly have not spent the time to look at the code to any
> antiquated Unix kernels to confirm whether this was ever actually true,
> anywhere.  And I don't intend to.  But it's at least plausible that it
> was true at one point in some popular Unix.  As you yourself said, for
> quite a long while now on Linux (since August of 1995), sync() actually
> does wait until the buffers are flushed.  But even that is mostly
> irrelevant as the kernel forces the buffers to be flushed periodically
> and flushes them prior to system shutdown (assuming it can, of
> course).
I agree with your explanation. In any case, the halt system call at that
time did not wait. Also, system calls were generally single threaded.

-- 
Jerry Feldman <gaf at blu.org>
Boston Linux and Unix
PGP key id:3BC1EB90 
PGP Key fingerprint: 49E2 C52A FC5A A31F 8D66  C0AF 7CEA 30FC 3BC1 EB90