svn commit: r283136 - head/sys/netinet

[Lawrence Stewart]

On 05/20/15 23:19, Eric van Gyzen wrote:
> On 05/20/2015 02:33, Lawrence Stewart wrote:
>> On 05/20/15 14:24, Hiren Panchasara wrote:
>>> On 05/20/15 at 02:13P, Lawrence Stewart wrote:
>>>> Hi Hiren,
>>>>
>>>> On 05/20/15 11:08, Hiren Panchasara wrote:
>>>>> Author: hiren Date: Wed May 20 01:08:01 2015 New Revision:
>>>>> 283136 URL: https://svnweb.freebsd.org/changeset/base/283136
>>>>>
>>>>> Log: Add a new sysctl net.inet.tcp.hostcache.purgenow=1 to
>>>>> expire and purge all entries in hostcache immediately.
>>>>>
>>>>> In collaboration with:	bz, rwatson MFC after:	1 week Relnotes:
>>>>> yes Sponsored by:	Limelight Networks
>>>>
>>>> Why introduce a new sysctl and not change the existing behaviour
>>>> of net.inet.tcp.hostcache.purge?
>>>
>>> I thought it'd make more sense to keep the existing behavior as is
>>> and provide new knob for the new behavior.
>>
>> Don't think so - why would deferring a purge to the next purge run be
>> useful compared to purging immediately? I'd strongly suggest you adapt
>> this change to the existing purge sysctl. I can't see why anyone would
>> miss the old functionality.
> 
> I am generally wary of a question such as "Why would anyone want...", because as soon as the code is released, someone answers it.
> 
> That being said, I have always wanted Hiren's purgenow behavior, and I've always been annoyed by the lazy-purge behavior.  I would suggest implementing Lawrence's suggestion, but NOT MFC'ing it, since that would be a disruptive change.
> 
> Thanks for your work, Hiren.

I see no reason not to MFC it - it's not a POLA violation for a stable
branch. When the user requests a purge, it's surely equally as good (and
I think anyone of right mind would argue better ;) to purge immediately
than some number of seconds "n" in the future, where "n" is between 1
and the value of net.inet.tcp.hostcache.prune.

Cheers,
Lawrence