How to make multiple clients can get the response at the same time by stream.

[Per Buer]

Hi,

I was a bit quick and I didn't read the whole email the first time. Sorry
about that. You're actually using the streaming branch, already I see. What
you're writing is really, really odd. There is a slight lock while the
"first" object is being fetched where other requests will be put on the
waiting list. However, when the hit-for-pass object is created these should
be released and pass'ed to the clients.

If the backend takes forever coming back with the response headers then the
situation would be something like what you describe. However, that would be
odd and doesn't make much sense.

PS: The streaming branch was renamed "plus" when it got other experimental
features. You'll find source
https://github.com/mbgrydeland/varnish-cacheand packages at
repo.varnish-cache.org/test if I recall correctly.




On Wed, Jan 30, 2013 at 3:32 AM, Xianzhe Wang <wxz19861013 at gmail.com> wrote:

>
> Hi,
> Thanks a lot.
>
> I tried option
> "set req.hash_ignore_busy = true;"
> in vlc_recv.
> I think it works. But there are side effects: it would increase backend
> load.
>
> I have an idea about it in my previous email. what do you think about it?
>
> Another question is that where can I find the "plus" branch of Varnish
> which matches this issue.
>
>  Any suggestions  will be appreciate.
> Thanks again for help.
>
> Regards,
> --
> Shawn Wang
>
>
> ---------- Forwarded message ----------
> From: Xianzhe Wang <wxz19861013 at gmail.com>
> Date: 2013/1/30
> Subject: Re: How to make multiple clients can get the response at the same
> time by stream.
> To: Jakub Słociński <kuba at ovh.net>
>
>
> Hi  Jakub S.
> Thank you very much.
> I tried, and take a simple test, two client request the big file at the
> same time, they get the response stream immediately, so  it works.
> In that case, multiple requests will go directly to "pass", they do not
> need to wait, but it would increase backend load.
> We need to balance the benefits and drawbacks.
>
> I wanna is that:
>     Client 1 requests url /foo
>     Client 2..N request url /foo
>     Varnish tasks a worker to fetch /foo for Client 1
>     Client 2..N are now queued pending response from the worker
>     Worker fetch response header(just header not include body) from
> backend, and find it  non-cacheable, then  make the remaining
> requests(Client 2..N) go directly to "pass". And creat the hit_for_pass
> object synchronously in the first request(Client 1).
>     Subsequent requests are now given the hit_for_pass object instructing
> them to go to the backend as long as the hit_for_pass object exists.
>
> As I mentioned below, is it feasible? Or do you have any Suggestions?
>
> Thanks again for help.
>
> Regards,
> --
> Shawn Wang
>
>
>
> 2013/1/29 Jakub Słociński <kuba at ovh.net>
>
>> Hi Xianzhe Wang,
>> you should try option
>> "set req.hash_ignore_busy = true;"
>> in vlc_recv.
>>
>> Regards,
>> --
>> Jakub S.
>>
>>
>> Xianzhe Wang napisał(a):
>> > Hello everyone,
>> >     My varnish version is 3.0.2-streaming release.And I set
>> > "beresp.do_stream  = true" in vcl_fetch in order to "Deliver the object
>> to
>> > the client directly without fetching the whole object into varnish";
>> >
>> > This is a part of my *.vcl file:
>> >
>> >  sub vcl_fetch {
>> >     set beresp.grace = 30m;
>> >
>> >     set beresp.do_stream = true;
>> >
>> >     if (beresp.http.Content-Length && beresp.http.Content-Length ~
>> > "[0-9]{8,}") {
>> >        return (hit_for_pass);
>> >     }
>> >
>> >      if (beresp.http.Pragma ~ "no-cache" || beresp.http.Cache-Control ~
>> > "no-cache" || beresp.http.Cache-Control ~ "private") {
>> >            return (hit_for_pass);
>> >        }
>> >
>> >      if (beresp.ttl <= 0s ||
>> >          beresp.http.Set-Cookie ||
>> >          beresp.http.Vary == "*") {
>> >
>> >                 set beresp.ttl = 120 s;
>> >                 return (hit_for_pass);
>> >      }
>> >
>> >     return (deliver);
>> >  }
>> >
>> > Then I request a big file(about 100M+) like "xxx.zip" from
>> clients.There is
>> > only one client can access the object.because "the object will marked as
>> > busy as it is delivered."
>> >
>> > But if  the request goes directly to “pass” ,multiple clients can get
>> the
>> > response at the same time.
>> >
>> > Also if I remove
>> >   if (beresp.http.Content-Length && beresp.http.Content-Length ~
>> > "[0-9]{8,}") {
>> >        return (hit_for_pass);
>> >     }
>> > to make the file cacheable,multiple clients can get the response at the
>> > same time.
>> >
>> > Now I want "multiple clients can get the response at the same time." in
>> all
>> > situations("pass","hit","hit_for_pass").
>> >
>> > What can I do for it?
>> > Any suggestions  will be appreciate.
>> > Thank you.
>> >
>> >  -Shawn Wang
>>
>> > _______________________________________________
>> > varnish-misc mailing list
>> > varnish-misc at varnish-cache.org
>> > https://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc
>>
>>
>
>


-- 
 <http://www.varnish-software.com/> *Per Buer*
CEO | Varnish Software AS
Phone: +47 958 39 117 | Skype: per.buer
We Make Websites Fly!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.varnish-cache.org/lists/pipermail/varnish-misc/attachments/20130130/85221a26/attachment.html>