Advanced Varnish configuration
Varnish provides several features that prevent customers from experiencing long delays and timeouts when the Commerce server is not functioning properly. These features can be configured in the default.vcl
file. This topic describes the additions that Commerce provides in the VCL (Varnish Configuration Language) file you download from the Admin.
See the Varnish Reference Manual for details about using the Varnish Configuration Language.
Health check
The Varnish health check feature polls the Commerce server to determine whether it is responding in a timely manner. If it is responding normally, fresh content is regenerated after the Time to Live (TTL) period expires. If not, Varnish always serves stale content.
Commerce defines the following default health check:
.probe = {
.url = "/pub/health_check.php";
.timeout = 2s;
.interval = 5s;
.window = 10;
.threshold = 5;
}
Every 5 seconds, this health check calls the pub/health_check.php
script. This script checks the availability of the server, each database, and Redis (if installed). The script must return a response within 2 seconds. If the script determines that any of these resources are down, it returns a 500 HTTP error code. If this error code is received in six out of ten attempts, the backend is considered unhealthy.
The health_check.php
script is located in the pub
directory. If your Commerce root directory is pub
, then be sure to change the path in the url
parameter from /pub/health_check.php
to health_check.php
.
For more information, see the Varnish health checks documentation.
Grace mode
Grace mode enables Varnish to keep an object in cache beyond its TTL value. Varnish can then serve the expired (stale) content while it fetches a new version. This improves the flow of traffic and decreases load times. It is used in the following situations:
- When the Commerce backend is healthy, but a request is taking longer than normal
- When the Commerce backend is not healthy.
The vcl_hit
subroutine defines how Varnish responds to a request for objects that have been cached.
When the Commerce backend is healthy
When the health checks determine that the Commerce backend is healthy, Varnish checks whether time remains in the grace period. The default grace period is 300 seconds, but a merchant can set the value from the Admin as described in Configure Commerce to use Varnish. If the grace period has not expired, Varnish delivers the stale content, and asynchronously refreshes the object from the Commerce server. If the grace period has expired, Varnish serves the stale content and synchronously refreshes the object from the Commerce backend.
The maximum amount of time that Varnish serves a stale object is the sum of the grace period (300 seconds by default) and the TTL value (86400 seconds by default).
To change the default grace period from within the default.vcl
file, edit the following line in the vcl_hit
subroutine:
if (obj.ttl + 300s > 0s) {
When the Commerce backend is not healthy
If the Commerce backend is not responsive, Varnish serves stale content from cache for three days (or as defined in beresp.grace
) plus the remaining TTL (which by default is one day), unless the cached content is manually purged.
Saint mode
Saint mode excludes unhealthy backends for a configurable amount of time. As a result, unhealthy backends cannot serve traffic when using Varnish as a load balancer. Saint mode can be used with grace mode to allow for complex handling of unhealthy backend servers. For example, if one backend server is unhealthy, retries can be routed to another server. If all other servers are down, then serve stale cached objects. The saint mode backend hosts and blackout periods are defined in the default.vcl
file.
Saint mode can also be used when Commerce instances are individually taken offline to perform maintenance and upgrade tasks without affecting the availability of the Commerce site.
Saint mode prerequisites
Designate one machine as the primary installation. On this machine, install the main instance of Commerce, mySQL database, and Varnish.
On all other machines, the Commerce instance must have access the primary machine’s mySQL database. The secondary machines should also have access to the files of the primary Commerce instance.
Alternatively, static files versioning can be turned off on all machines. This can be accessed from the Admin under Stores > Settings > Configuration > Advanced > Developer > Static Files Settings > Sign Static Files = No.
Finally, all Commerce instances must be in production mode. Before Varnish starts, clear the cache on each instance. In the Admin, go to System > Tools > Cache Management and click Flush Magento Cache. You can also run the following command to clear the cache:
bin/magento cache:flush
Installation
Saint mode is not part of the main Varnish package. It is a separately versioned vmod
that must be downloaded and installed. As a result, you must recompile Varnish from source, as described in the installation instructions for your version of Varnish.
After you recompile, you can install the Saint mode module. In general, follow these steps:
-
Obtain the source code from Varnish modules. Clone the Git version (master version) since the 0.9.x versions contain a source code error.
-
Build the source code with autotools:
code language-bash sudo apt-get install libvarnishapi-dev || sudo yum install varnish-libs-devel ./bootstrap # If running from git. ./configure make make check # optional sudo make install
See Varnish module collection for information about installing the Saint mode module.
Sample VCL file
The following code example shows the code that must be added to your VCL file to enable saint mode. Place the import
statements and backend
definitions at the top of the file.
import saintmode;
import directors;
backend default1 {
.host = "192.168.0.1";
.port = "8080";
.first_byte_timeout = 600s;
.probe = {
.url = "/pub/health_check.php";
.timeout = 2s;
.interval = 5s;
.window = 10;
.threshold = 5;
}
}
backend default2 {
.host = "192.168.0.2";
.port = "8080";
.first_byte_timeout = 600s;
.probe = {
.url = "/pub/health_check.php";
.timeout = 2s;
.interval = 5s;
.window = 10;
.threshold = 5;
}
}
sub vcl_init {
# Instantiate sm1, sm2 for backends tile1, tile2
# with 10 blacklisted objects as the threshold for marking the
# whole backend sick.
new sm1 = saintmode.saintmode(default1, 10);
new sm2 = saintmode.saintmode(default2, 10);
# Add both to a director. Use sm0, sm1 in place of tile1, tile2.
# Other director types can be used in place of random.
new magedirector = directors.random();
magedirector.add_backend(sm1.backend(), 1);
magedirector.add_backend(sm2.backend(), 1);
}
sub vcl_backend_fetch {
# Get a backend from the director.
# When returning a backend, the director will only return backends
# saintmode says are healthy.
set bereq.backend = magedirector.backend();
}
sub vcl_backend_response {
if (beresp.status >= 500) {
# This marks the backend as sick for this specific
# object for the next 20s.
saintmode.blacklist(20s);
# Retry the request. This will result in a different backend
# being used.
return (retry);
}
}