Configuring Varnish 6 with grace mode, vcl_hit is never executed - varnish

Question

I'm trying to configure varnish 6.0.0 as a cache server when the webserver is going down for maintanance but I can't get the expected behavior.
I have the next configuration (using grace mode):
vcl 4.0;
import directors;
import std;
backend default {
.host = "some.server.com";
.port = "80";
.probe = {
.url = "/health_check.php";
.interval = 5s;
.timeout = 1s;
.window = 5;
.threshold = 3;
}
}
sub vcl_recv {
std.syslog(180, "RECV: recv");
#Cache - grace mode
set req.http.grace = "none";
}
sub vcl_backend_response {
std.syslog(180, "RECV: backend");
#Cache - grace mode
set beresp.ttl = 10s;
set beresp.grace = 1h;
#set beresp.keep = 24h;
}
sub vcl_deliver {
std.syslog(180, "RECV: deliver");
#Cache
set resp.http.grace = req.http.grace;
}
sub vcl_hit {
std.syslog(180, "RECV: hit************************");
if (obj.ttl >= 0s) {
# normal hit
return (deliver);
}
# We have no fresh fish. Lets look at the stale ones.
if (std.healthy(req.backend_hint)) {
# Backend is healthy. Limit age to 10s.
if (obj.ttl + 10s > 0s) {
set req.http.grace = "normal(limited)";
return (deliver);
} else {
# No candidate for grace. Fetch a fresh object.
return(miss);
}
} else {
# backend is sick - use full grace
if (obj.ttl + obj.grace > 0s) {
set req.http.grace = "full";
return (deliver);
} else {
# no graced object.
return (miss);
}
}
}
Then, when I get the log messages:
tail -f /var/log/messages
I just get the next steps:
varnishd[11801]: RECV: recv
varnishd[11801]: RECV: hash
varnishd[11801]: RECV: backend
varnishd[11801]: RECV: deliver
For this I know that the subroutine "vcl_hit" never is executed, so when the webserver is gone down, immediately I get an error messsage from varnish instead the cache from varnish:
Error 503 Backend fetch failed
Backend fetch failed
Guru Meditation:
XID: 164960
Varnish cache server
Any idea to get the expected behavior?

Answer 1

The webpage that we use have a session for the users, so we need identify when a user login or not.
With the previous configuration even if the cookie is empty, the default varnish configuration doesn't make cache for requests with cookies.
So I added some lines to make content cacheable (explained in code):
sub vcl_recv {
# intial state
set req.http.grace = "none";
# If our backend is down, unset all cookies and serve pages from cache.
if (!std.healthy(req.backend_hint)) {
unset req.http.Cookie;
}
# Are there cookies left with only spaces or that are empty?
if (req.http.cookie ~ "^\s*$") {
unset req.http.cookie;
}
# NO cache si NO es get o head --> ejm, post, read, delete, etc
if (req.method != "GET" && req.method != "HEAD") {
return (pass);
}
# If server is healthy && user is in session --> no cache
if (std.healthy(req.backend_hint)) {
if (req.http.Authorization || req.http.cookie ~ "some_cookie=") {
/* Not cacheable by default */
return (pass);
}
}
return(hash);
}
sub vcl_backend_response {
#Cache - grace mode
set beresp.ttl = 10s;
#set beresp.ttl = 1h;
set beresp.grace = 1h;
set beresp.keep = 24h;
set beresp.http.Cache-Control = "max-age=3600";
}
The others subroutines have the same configuration.

Answer 2

It could be that the response from the backend is "uncacheable".
Does it have any Cache-Control header set by default? or cookies?
You can unset that header or set it to another value that allows caching:
unset beresp.http.Cache-Control
If you want that header to be private, no-cache when sent to the client to prevent upstream caching, then reset it in vcl_deliver