[squid-users] force full download and limit bandwith

Alex Rousskov rousskov at measurement-factory.com
Wed Jan 17 19:27:14 UTC 2018

On 01/17/2018 09:15 AM, Heiler Bemerguy wrote:
> Everytime I enable this:
>     range_offset_limit -1 fullDLext
>     quick_abort_min 0 KB
>     quick_abort_max 0 KB
>     quick_abort_pct 95
> Our link is fully utilized for a whole day if at least 2 users tries to
> get the same file with range (206) requests (Windows Update does this)
> Rockstore with SMP and 3.5.27
> The file will be simultaneously downloaded from the beginning by each of
> workers that got the request, and in the end it won't be cached, giving
> a SWAPFAIL_MISS message.. This easily TOPS our link for a whole day
> (maybe till the users turn their computers off..?)
> It's a known bug, as I filled a bug report years ago.. it was never
> fixed...

To avoid misundertanding, there are at least three overlapping issues
here. Only one of them is a bug:

* The fact that Squid may fetch more data than is necessary when
configured to prefetch is not a bug. Both range_offset_limit and
quick_abort_* are essentially prefetching options -- they tell Squid to
fetch more than what the asking client needs.

* The fact that multiple workers may attempt concurrent downloads of the
same object is not necessarily a bug. Squid behavior in such cases is
controlled by collapsed_forwarding (and various HTTP rules).

* The fact that the concurrently fetched cachable object may never be
cached is a bug. It may be fixed as a side effect of the pending pull
request #46 changes, but I am not sure.


> Em 17/01/2018 12:36, Alex Rousskov escreveu:
>> On 01/17/2018 05:47 AM, joseph wrote:
>>> Amos Jeffries wrote
>>>> "range_offset_limit N" only affects the initial starting point for a 
>>>> transaction. If a client wants to start reading a range somewhere in the 
>>>> first N bytes Squid will request the full file in order to cache it for 
>>>> future requests. Otherwise only the range the client wants will be 
>>>> requested - and cannot (yet) be cached.
>>> meaning it will continue  dose not restart 
>> You have misinterpreted Amos's response: Whether Squid continues after
>> the client disappears depends on quick_abort_* settings, not
>> range_offset_limit settings.
>> HTH,
>> Alex.
>> _______________________________________________
>> squid-users mailing list
>> squid-users at lists.squid-cache.org
>> http://lists.squid-cache.org/listinfo/squid-users
> _______________________________________________
> squid-users mailing list
> squid-users at lists.squid-cache.org
> http://lists.squid-cache.org/listinfo/squid-users

More information about the squid-users mailing list