Backend health polling with client or round-robin

[Justin Davis]

Hello all.

Is "/server-status?auto" an appropriate URL to use for backend health polling with a client or round-robin director?

We recently had one of our two load balanced servers go belly-up, during which Varnish continued attempting to server content from the failed node.

Would a small file be preferable? It's my understanding that a backend node is marked as unhealthy if the health check times out or returns a status other than 200 more than the configured threshhold times per configured window.


My backend declaration:

backend web1 {
   .host   = "XX.XX.XX.XX";
   .port   = "http";
  .probe  = {
    .url  = "/server-status?auto";
    .timeout  = 34 ms;
    .interval  = 3 s;
    .window = 10;
    .threshold  = 8;
  }
}

backend web2 {
   .host   = "XX.XX.XX.XX";
   .port   = "http";
  .probe  = {
    .url  = "/server-status?auto";
    .timeout  = 34 ms;
    .interval  = 3 s;
    .window = 10;
    .threshold  = 8;
  }
}


director web client {
   {
     .backend  = web1;
     .weight = 1;
   }
   {
     .backend  = web2;
     .weight = 1;
   }
}


Thank you all for any assistance, direction, pointers, consideration or words of encouragement.

Cheers,
Justin


--
Justin Davis
Liberal Arts ITS
Senior Systems Administrator
justindavis at mail.utexas.edu