| Author |
Message |
< Yaws mailing list ~ Proposal: throttle incoming requests |
| tobbe |
 Posted: Wed Nov 01, 2006 12:50 pm |
 |
|
|
User
Joined: 19 Jan 2005
Posts: 274
Location: Stockholm, Sweden
|
If Yaws deliver requests faster than what the rest
of the system can handle them, it would be nice
to be able to throttle the amount of incoming requests.
I was thinking of adding a global parameter:
REQUESTS_PER_SECOND=<integer between 1 and 1000>
The yaws_server.erl could be modified like this:
--------------------------------------------------
do_accept(GS) when GS#gs.ssl == nossl ->
?Debug("wait in accept ... ~n",[]),
+ sleep(GS#gs.req_per_sec),
gen_tcp:accept(GS#gs.l);
do_accept(GS) when GS#gs.ssl == ssl ->
+ sleep(GS#gs.req_per_sec),
ssl:accept(GS#gs.l,10000).
+%% Other values make no sense
+sleep(I) when I>1,I=<1000 ->
+ Timeout = trunc(1000/I),
+ receive after Timeout -> true end;
+sleep(_) ->
+ true.
-------------------------------------------------
So what do you think about this?
Cheers, Tobbe
-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
Erlyaws-list mailing list
Erlyaws-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/erlyaws-list
Post recived from mailinglist |
|
|
| Back to top |
   
|
| noss |
| Posted: Wed Nov 01, 2006 1:04 pm |
|
|
|
User
Joined: 09 Oct 2005
Posts: 290
|
On 11/1/06, Torbjorn Tornkvist <tobbe@tornkvist.org> wrote:
> If Yaws deliver requests faster than what the rest
> of the system can handle them, it would be nice
> to be able to throttle the amount of incoming requests.
>
> I was thinking of adding a global parameter:
How about an interface to change this value while running? So one could
regulate it based on application-specific load measurements.
-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
Erlyaws-list mailing list
Erlyaws-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/erlyaws-list
Post recived from mailinglist |
|
|
| Back to top |
|
| Guest |
| Posted: Wed Nov 01, 2006 1:04 pm |
|
|
|
Guest
|
Torbjorn Tornkvist <tobbe@tornkvist.org> wrote:
> If Yaws deliver requests faster than what the rest
> of the system can handle them, it would be nice
> to be able to throttle the amount of incoming requests.
>
> I was thinking of adding a global parameter:
>
> REQUESTS_PER_SECOND=<integer between 1 and 1000>
>
> The yaws_server.erl could be modified like this:
> --------------------------------------------------
> do_accept(GS) when GS#gs.ssl == nossl ->
> ?Debug("wait in accept ... ~n",[]),
> + sleep(GS#gs.req_per_sec),
> gen_tcp:accept(GS#gs.l);
> do_accept(GS) when GS#gs.ssl == ssl ->
> + sleep(GS#gs.req_per_sec),
> ssl:accept(GS#gs.l,10000).
>
> +%% Other values make no sense
> +sleep(I) when I>1,I=<1000 ->
> + Timeout = trunc(1000/I),
> + receive after Timeout -> true end;
> +sleep(_) ->
> + true.
> -------------------------------------------------
>
> So what do you think about this?
Without looking at how the Yaws request receiving code looks like, it
appears to me that REQUESTS_PER_SECOND is rather SLEEPTIME_PER_REQUEST?
Besides that, couldn't such a rate limiting be made more dynamic if it
was based on the run queue length as reported by erlang:statistics/1?
Ulf Wiger indicated to me that that could be a good source of indication
that you are overloaded, and should start backing off. I've been
planning on implementing it in YXA but haven't found the time yet so I
cannot tell you if it works in practice.
/Fredrik
-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
Erlyaws-list mailing list
Erlyaws-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/erlyaws-list
Post recived from mailinglist |
|
|
| Back to top |
|
| tobbe |
| Posted: Wed Nov 01, 2006 1:23 pm |
|
|
|
User
Joined: 19 Jan 2005
Posts: 274
Location: Stockholm, Sweden
|
Christian S wrote:
> On 11/1/06, Torbjorn Tornkvist <tobbe@tornkvist.org> wrote:
>> If Yaws deliver requests faster than what the rest
>> of the system can handle them, it would be nice
>> to be able to throttle the amount of incoming requests.
>>
>> I was thinking of adding a global parameter:
>
> How about an interface to change this value while running? So one could
> regulate it based on application-specific load measurements.
Yes, it could be a good idea.
--Tobbe
-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
Erlyaws-list mailing list
Erlyaws-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/erlyaws-list
Post recived from mailinglist |
|
|
| Back to top |
|
| tobbe |
| Posted: Wed Nov 01, 2006 1:28 pm |
|
|
|
User
Joined: 19 Jan 2005
Posts: 274
Location: Stockholm, Sweden
|
Fredrik Thulin wrote:
>
> Without looking at how the Yaws request receiving code looks like, it
> appears to me that REQUESTS_PER_SECOND is rather SLEEPTIME_PER_REQUEST?
Hm..yes..or perhaps MAX_REQUESTS_PER_SECOND.
>
> Besides that, couldn't such a rate limiting be made more dynamic if it
> was based on the run queue length as reported by erlang:statistics/1?
But if you have a large system, involving heavy use of mnesia etc, it
is hard to measure this. This is an easy way (perhaps too easy?) to
protect the rest of the system from being overloaded.
--Tobbe
> Ulf Wiger indicated to me that that could be a good source of indication
> that you are overloaded, and should start backing off. I've been
> planning on implementing it in YXA but haven't found the time yet so I
> cannot tell you if it works in practice.
>
> /Fredrik
>
>
> -------------------------------------------------------------------------
> Using Tomcat but need to do more? Need to support web services, security?
> Get stuff done quickly with pre-integrated technology to make your job easier
> Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
> http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
Erlyaws-list mailing list
Erlyaws-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/erlyaws-list
Post recived from mailinglist |
|
|
| Back to top |
|
| noss |
| Posted: Wed Nov 01, 2006 1:41 pm |
|
|
|
User
Joined: 09 Oct 2005
Posts: 290
|
On 11/1/06, Torbjorn Tornkvist <tobbe@tornkvist.org> wrote:
> Christian S wrote:
> > How about an interface to change this value while running? So one could
> > regulate it based on application-specific load measurements.
>
> Yes, it could be a good idea.
More thinking about the issue.
To make a working automatic controller one also needs
information about current request frequency.
Otherwise it is hard to know how sleep time should be modified
to modify throttling in the desired direction.
-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
Erlyaws-list mailing list
Erlyaws-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/erlyaws-list
Post recived from mailinglist |
|
|
| Back to top |
|
|
|
Digg It
Del.icio.us
Reddit
Facebook
Stumble Upon
Technorati
