[Varnish] #749: 503 if web server closes connection too soon.

[Varnish]

#749: 503 if web server closes connection too soon.
-------------------------+--------------------------------------------------
 Reporter:  Dan          |       Owner:  phk  
     Type:  enhancement  |      Status:  new  
 Priority:  normal       |   Milestone:       
Component:  varnishd     |     Version:  trunk
 Severity:  normal       |    Keywords:       
-------------------------+--------------------------------------------------

Comment(by Dan):

 Replying to [comment:1 phk]:

 > For one thing, we should only retry if we got a recycled backend
 connection.
 > But once we retry, we may get another recycled backend connection,
 How about forcing a new connection on-retry (hence not recycling) to avoid
 retrying a lot of dead connections in case the webserver had a general
 problem?

 > so how many times do we retry ?

 Make it configurable and default to 1 or 2 retries.

 Another thing is whether there should be a small sleep before retry, and
 if there should be "incremental backoff" on failed retries ? So each retry
 attempt sleeps a couple of miliseconds * retry-attempt-count.

 Eg. "10ms * (retry_count-1)":
  * 1st retry: 0*10 = sleep 0ms
  * 2nd retry: 1*10 = sleep 10ms
  * 3nd retry: 2*10 = sleep 20ms

 Dont know if thats a good idea, but im thinking it will give whatever
 problem occured a little time to solve itself.

-- 
Ticket URL: <http://varnish-cache.org/ticket/749#comment:2>
Varnish <http://varnish-cache.org/>
The Varnish HTTP Accelerator