Varnish does not appear to be configured for a public folder web root (at least, with NGINX)

[andrewhowdencom]

Preconditions

1. Magento 2.4-develop
2. Varnish 4/5/6

Steps to reproduce

1. Set up a server according to the provided recommendations, with NGINX as the HTTP host and using the public folder
2. Export varnish as shown by Magento
3. Install varnish configuration

Expected result

1. Server works

Actual result

1. Varnish doesn't work (specifically, it returns a 503 and complains about upstream being down). Additionally, there are a bunch of messages in the NGINX error log about "404 /pub/health_check.php not found"

File is here: https://github.com/magento/magento2/blob/2.2-develop/app/code/Magento/PageCache/etc/varnish5.vcl

As you can see, it has the pub hardcoded. I might suggest doing detection in the application, and adjusting the probe root accordingly.

[orlangur]

I believe it's handled properly in other lines of the same file: if (req.url ~ "^/(pub/)?(media|static)/") {

Not sure why root index.php it still not removed in favor of pub/index.php though.

[andrewhowdencom]

I'm not sure -- I don't know if Varnish caches it's own probes? To me that wouldn't make sense. Docs aren't helpful here:

https://varnish-cache.org/docs/4.0/reference/vcl.html#reference-vcl-probes

Additionally, tailing nginx logs seems to support this (with the pass directive disabled):

127.0.0.1 - - [01/Nov/2017:17:23:30 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"
127.0.0.1 - - [01/Nov/2017:17:23:35 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"
127.0.0.1 - - [01/Nov/2017:17:23:40 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"
127.0.0.1 - - [01/Nov/2017:17:23:45 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"
127.0.0.1 - - [01/Nov/2017:17:23:50 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"
127.0.0.1 - - [01/Nov/2017:17:23:55 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"
127.0.0.1 - - [01/Nov/2017:17:24:00 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"
127.0.0.1 - - [01/Nov/2017:17:24:05 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"
127.0.0.1 - - [01/Nov/2017:17:24:10 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"
127.0.0.1 - - [01/Nov/2017:17:24:15 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"
127.0.0.1 - - [01/Nov/2017:17:24:20 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"
127.0.0.1 - - [01/Nov/2017:17:24:25 +0100] "GET /health_check.php HTTP/1.1" 200 5 "-" "-" "-"

[andrewhowdencom]

Oh also I patched this to remove the pub. Specifically, we route based on virtualhosts rather than ports:

backend default {
    .host = "127.0.0.1";
    .port = "80";
    .first_byte_timeout = 600s;
    .probe = {
        # SITEWARDS CHANGES START
        # The normal probe configuration does not allow setting specific headers for request[1]. So, the below snippet
        # manually constructors the request to include the Host header.
        #
        # [1] https://varnish-cache.org/lists/pipermail/varnish-misc/2011-October/021298.html
        .request =
            "GET /health_check.php HTTP/1.1"
            "Host: magento.local"
            "Connection: close";
        # SITEWARDS CHANGES END
        .timeout = 2s;
        .interval = 5s;
        .window = 10;
        .threshold = 5;
   }
}

[magento-engcom-team]

Hi @andrewhowdencom
Could you please let us know if the issue is still actual? Thank you

[andrewhowdencom]

Looks like it. As mentioned, i customised the VCL in production for us as above, for various reasons. But the 2.2 develop branch still has the path hard-coded.

I imagine the rule "if you need two versions of index.php you'll probably need two versions of health_check.php" holds true.

https://github.com/magento/magento2/blob/2.2-develop/app/code/Magento/PageCache/etc/varnish5.vcl#L13