I need to create an irule that will reject a specified percnetage of traffic.

Hi,

How about something like this?

when HTTP_REQUEST {

 Check the path
if { [HTTP::path] eq "/build/apps/medrec/jaml" }{

 For 50% of traffic send a 502 response
if { rand() > .50 }{
HTTP::respond 502 content {html...}
}
}
}

Aaron

Hi kimjim,

 

 

Have you thought of using Ratio Load Balancing with Connection Limits set on the Servers?

 

 

Or you could even use SNMP Monitor to monitor the server and Dynamic Ratio load balancing (you would use the SNMP DCA or SNMP DCA Base Templates). That is described in detail here if you are interested.

 

 

Hope this helps.

Nice idea Michael.

 

 

You could also set connection limits on the pool members and/or virtual server to protect them.

 

 

Aaron

Hoolio Thank You very much for all your help.

 

Your irule worked like a charm

 

but i have one question will rand add any latency if so can you suggest me any other alternative.

 

 

Micheal Thanks for your answer , i am not sure how to do it ,really appreciate your help.

 

rand should add very little CPU load or latency.

 

 

Aaron

Why not tinker to figure out the high water mark for your application and use something like this... https://devcentral.f5.com/wiki/iRules.HTTPSessionLimit.ashx

 

 

This rule "Limits total concurrent HTTP sessions to a pre-defined threshold, allowing those clients with a session cookie to continue once the limit is reached, while redirecting new clients until concurrent sessions are again below the theshold."

 

 

That was you can have some clients have a decent experience instead of randomly getting bumped

For the 50% case, you could optimise the rule as follows:

 

 

when HTTP_REQUEST {

 

set flag 0

 

Check the path

 

if { [HTTP::path] eq "/build/apps/medrec/jaml" }{

 

 

For 50% of traffic send a 502 response

 

if { $flag == 0 }{

 

HTTP::respond 502 content {html...}

 

set flag 1

 

}

 

else {

 

set flag 0

 

}

 

}

 

}

I like your thinking TDK. But using a local variable for this would be specific to that TCP connection. So if you use that logic, every other HTTP request to the specific URI on each TCP connection would be dropped. I'd guess this wouldn't lead to dropping 50% of requests as some (or most?) clients would not reuse their TCP connections to make multiple requests to that specific URI.

 

 

You could use a session table entry to create a flag that you increment and then reset to 0 (https://devcentral.f5.com/wiki/iRules.CMP_v10_compatible_counters_using_the_session_table.ashx).

 

 

However, using expr will give a much more even distribution compared with a local variable, regardless of the connection reuse behavior of clients. And it doesn't require any session tracking complexity.

 

 

Aaron