Request for review: do not use --sync with df during sanity check

[Erik Trimble]

On 5/4/2011 3:37 PM, David Katleman wrote:
> Would be interesting to know the original objection to Andrew's change 
> last year.
>
> Absent that, I see no reason this could not be removed and the change 
> looks fine.
>
> When originally implemented 10+ years ago, disk space was considerably 
> smaller, as were buffers, so an accurate count was more relevant.
>
> Today multiple builds on the same disk are fairly common, making the 
> check itself even less useful.
>
> Add to that, the information being gathered is just for a WARNING, the 
> build will continue, regardless.  Hardly the need for exacting 
> accuracy, especially since the df information is then compared to 
> static estimates of how much space a build will consume.
>
>             Dave
>             (katleman)
>

That, and particularly with ZFS (either on FreeBSD or Solaris), 'df' is 
NOT reliable for accurate disk space consumption.

I'd suggest that we simply obsolete that check, as it's not really very 
useful anymore, on any platform.

-Erik




>
> On 5/4/2011 2:43 PM, Dr Andrew John Hughes wrote:
>> On 17:35 Wed 04 May     , Omair Majid wrote:
>>> Hi,
>>>
>>> Can someone please review the following webrev?
>>>
>>> http://cr.openjdk.java.net/~omajid/webrevs/no-sync/01/
>>>
>>> It modifies the sanity check (under linux) to use df without --sync 
>>> when
>>> it computes the free space. This is what is done in the solaris case.
>>> The accuracy gained by using --sync is probably not worth the
>>> performance loss caused by flushing disk buffers.
>>>
>>> The patch was originally written by Andrew John Hughes
>>> (ahughes at redhat.com) and we have had this in icedtea6 for almost a 
>>> year now.
>>>
>>> Thanks,
>>> Omair
>> Let's see if you do better than I did...
>>
>> http://mail.openjdk.java.net/pipermail/build-dev/2010-June/003056.html


-- 
Erik Trimble
Java System Support
Mailstop:  usca22-123
Phone:  x17195
Santa Clara, CA