[Varnish] #1772: first_byte_timeout is ignored when re-using a backend connection (HTTP Keep Alive)

[Varnish]

#1772: first_byte_timeout is ignored when re-using a backend connection (HTTP Keep
Alive)
--------------------+------------------------
 Reporter:  tnt     |       Owner:
     Type:  defect  |      Status:  new
 Priority:  normal  |   Milestone:
Component:  build   |     Version:  4.1.0-TP1
 Severity:  normal  |  Resolution:
 Keywords:          |
--------------------+------------------------

Comment (by Dridi):

 My comments from the IRC:

 > my understanding is the following
 > we should pass a timeout to VBT_Wait, to be used with the underlying
 cond wait
 > the cond wait being done in a loop, that means more system calls to re-
 compute the deadline on each iteration
 > (which means it would probably not be done this way)
 > after a vbt wait, we could compute the remaining timeout and pass it to
 V1F_FetchRespHdr
 > I don't remember how vcl_pipe deals with timeouts, so that has to be
 considered too

 Then I wrote a test case:
 {{{
 varnishtest "honour first-byte timeout on a stolen connection"

 server s1 {
         rxreq
         expect req.url == "/first"
         txresp -hdr "Connection: keep-alive"

         rxreq
         expect req.url == "/stolen"
         txresp

         delay 1.5

         rxreq
         expect req.url == "/second"
         txresp
 } -start

 varnish v1 -vcl {
         backend s1 {
                 .host = "${s1_addr}";
                 .port = "${s1_port}";
                 .max_connections = 1;
                 .first_byte_timeout = 1s;
         }
 } -start

 client c1 {
         txreq -url "/first"
         rxresp
         expect resp.status == 200

         sema r1 sync 2
         sema r2 sync 2

         txreq -url "/second"
         rxresp
         expect resp.status == 503
 } -start

 client c2 {
         sema r1 sync 2

         txreq -url "/stolen"
         rxresp
         expect resp.status == 200

         sema r2 sync 2
 } -start

 client c1 -wait
 }}}

 And the thing that is puzzling me is that I couldn't make the test pass by
 moving the '''delay''' right before the last '''txresp''' in s1.

-- 
Ticket URL: <https://www.varnish-cache.org/trac/ticket/1772#comment:1>
Varnish <https://varnish-cache.org/>
The Varnish HTTP Accelerator