Description
Varnish is a caching front-end server. This document has notes on how to use Varnish with Plone.
This chapter contains information about using the Varnish caching proxy with Plone.
To use Varnish with Plone
Backporting examples (Ubuntu 8.04 Hardy Heron)
Buildout examples
You can access Varnish admin console on your server by:
# Your system uses a secret handshake file
varnishadm -T localhost:6082 -S /etc/varnish/secret
(Ubuntu/Debian installation)
The telent management console is available on some configurations where varnishadm cannot be used. The functionality is the same.
Example:
ssh yourhost
# Your system does not have a secret handshake file
telnet localhost 6082
Note
Port number depends on your Varnish settings.
More often than not, it is beneficial to load new configuration without bringing the cache down for maintenance. Using this method also checks the new VCL for syntax errors before activating it. Logging in to Varnish CLI requires the varnishadm tool, the address of the management interface, and the secret file for authentication.
See the varnishadm man-page for details.
Opening a new CLI connection to the Varnish console, in a buildout-based Varnish installation:
parts/varnish-build/bin/varnishadm -T localhost:8088
Port 8088 is defined in buildout.cfg:
[varnish-instance]
telnet = localhost:8088
Opening a new CLI connection to the Varnish console, in a system-wide Varnish installation on Ubuntu/Debian:
varnishadm -T localhost:6082 -S /etc/varnish/secret
You can dynamically load and parse a new VCL config file to memory:
vcl.load <name> <file>
For example:
vcl.load newconf_1 /etc/varnish/newconf.vcl
vcl.load will load and compile the new configuration. Compilation will fail and report on syntax errors. Now that the new configuration has been loaded, it can be activated with:
vcl.use newconf_1
Note
Varnish remembers <name> in vcl.load, so every time you need to reload your config you need to invent a new name for vcl.load / vcl.use command pair.
To see a real-time log dump (in a system-wide Varnish configuration):
varnishlog
By default, Varnish does not log to any file and keeps the log only in memory. If you want to extract Apache-like logs from varnish, you need to use the varnishncsa utility.
Check live "top-like" Varnish statistics:
parts/varnish-build/bin/varnishstat
Use the admin console to print stats for you:
stats
200 2114
95717 Client connections accepted
132889 Client requests received
38638 Cache hits
21261 Cache hits for pass
...
The following snippet will restart a varnishd instance which has been started from the plone.recipe.varnish buildout directly invoking bin/varnish-instance command.
It will also create an Apache-compatible log file which you can examine using text editing tools by running the varnishncsa command which will read log data from the Varnish memory-mapped file, and write it to a text file in Apache format.
Example:
#!/bin/sh
# Varnish restart script
sudo killall varnishd
sudo bin/varnish-instance
# Create Apache compatible log file
sudo kill `cat var/varnishncsa.pid`
sudo parts/varnish-build/bin/varnishncsa -D -d -a -w var/log/varnish.log -P var/varnishncsa.pid
When Varnish has been set-up you need to include Plone virtual hosting rule in its configuration file.
If you want to map Varnish backend directly to Plone-as-a-virtualhost (i.e. Zope's VirtualHostMonster is used to map site name to Plone site instance id) use req.url mutating.
The following maps the Plone site id plonecommunity to the plonecommunity.mobi domain. Plone is a single Zope instance, running on port 9999.
Example:
backend plonecommunity {
.host = "127.0.0.1";
.port = "9999";
}
sub vcl_recv {
if (req.http.host ~ "^(www.)?plonecommunity.mobi(:[0-9]+)?$"
|| req.http.host ~ "^plonecommunity.mfabrik.com(:[0-9]+)?$") {
set req.backend = plonecommunity
set req.url = "/VirtualHostBase/http/" req.http.host ":80/plonecommunity/VirtualHostRoot" req.url;
set req.backend = plonecommunity;
}
}
You give IP address(s) and ports to Varnish to listen to on the varnishd command line using -a switch. Edit /etc/default/varnish:
DAEMON_OPTS="-a 192.168.1.1:80 \
-T localhost:6082 \
-f /etc/varnish/default.vcl \
-s file,/var/lib/varnish/$INSTANCE/varnish_storage.bin,1G"
You can provide an uncached version of the site for editors:
Please see cache issues related to LinguaPlone.
Any cookie set on the server side (session cookie) or on the client-side (e.g. Google Analytics Javascript cookies) is poison for caching the anonymous visitor content.
HTTP caching needs to deal with both HTTP request and response cookie handling
Example how remove all Plone related cookies besides ones dealing with the logged in users (content authors):
sub vcl_recv {
if (req.http.Cookie) {
# (logged in user, status message - NO session storage or language cookie)
set req.http.Cookie = ";" req.http.Cookie;
set req.http.Cookie = regsuball(req.http.Cookie, "; +", ";");
set req.http.Cookie = regsuball(req.http.Cookie, ";(statusmessages|__ac|_ZopeId|__cp)=", "; \1=");
set req.http.Cookie = regsuball(req.http.Cookie, ";[^ ][^;]*", "");
set req.http.Cookie = regsuball(req.http.Cookie, "^[; ]+|[; ]+$", "");
if (req.http.Cookie == "") {
remove req.http.Cookie;
}
}
...
# Let's not remove Set-Cookie header in VCL fetch
sub vcl_fetch {
# Here we could unset cookies explicitly,
# but we assume plone.app.caching extension does it jobs
# and no extra cookies fall through for HTTP responses we'd like to cache
# (like images)
if (!beresp.cacheable) {
return (pass);
}
if (beresp.http.Set-Cookie) {
return (pass);
}
set beresp.prefetch = -30s;
return (deliver);
}
The snippet for stripping out non-Plone cookies comes from http://www.phase2technology.com/node/1218/
That article notes that "this processing occurs only between Varnish and the backend [...]; the client, typically a user’s browser, still has all the cookies. Nothing is happening to the client’s original request." While it's true that the browser still has the cookies, they never reach the backend and are therefor ignored.
Another example how to purge Google cookies only and allow other cookies by default:
sub vcl_recv {
# Remove Google Analytics cookies - will prevent caching of anon content
# when using GA Javascript. Also you will lose the information of
# time spend on the site etc..
if (req.http.cookie) {
set req.http.Cookie = regsuball(req.http.Cookie, "__utm.=[^;]+(; )?", "");
if (req.http.cookie ~ "^ *$") {
remove req.http.cookie;
}
}
....
Use the following snippet to set a HTTP response debug header to see what Varnish sees as cookie after vcl_recv clean-up:
sub vcl_fetch {
set beresp.http.X-Varnish-Cookie-Debug = "Request cookie: " req.http.Cookie;
...
}
And then test with wget:
cd /tmp # wget wants to save files...
wget -S http://www.site.fi
--2011-11-16 11:28:37-- http://www.site.fi/
Resolving www.site.fi (www.site.fi)... xx.20.128.xx
Connecting to www.site.fi (www.site.fi)|xx.20.128.xx|:80... connected.
HTTP request sent, awaiting response...
HTTP/1.1 200 OK
Server: Zope/(2.12.17, python 2.6.6, linux2) ZServer/1.1
X-Cache-Operation: plone.app.caching.noCaching
Content-Language: fi
Expires: Sun, 18 Nov 2001 09:28:37 GMT
Cache-Control: max-age=0, must-revalidate, private
X-Cache-Rule: plone.content.folderView
Content-Type: text/html;charset=utf-8
Set-Cookie: I18N_LANGUAGE="fi"; Path=/
Content-Length: 23836
X-Cookie-Debug: Request cookie: (null)
Date: Wed, 16 Nov 2011 09:28:37 GMT
X-Varnish: 1562749485
Age: 0
Via: 1.1 varnish
This cookie could be removed in vcl_fetch response post-processing (how?). However, a better solution is to disable this cookie in the backend itself: in this case in Plone portal_languages backend tool. Disable it by Use cookie for manual override setting in portal_languages.
You can make sure that Varnish does not accidentally cache error pages. E.g. it would cache front page when the site is down:
sub vcl_fetch {
if ( beresp.status >= 500 ) {
set beresp.ttl = 0s;
set beresp.cacheable = false;
}
...
}
More info
Below is an example how to create an action to purge the whole Varnish cache.
First you need to allow HTTP PURGE request in default.vcl from localhost. We'll create a special PURGE command which takes URLs to be purged out of the cache in a special header:
acl purge {
"localhost";
# XXX: Add your local computer public IP here if you
# want to test the code against the production server
# from the development instance
}
...
sub vcl_recv {
...
# Allow PURGE requests clearing everything
if (req.request == "PURGE") {
if (!client.ip ~ purge) {
error 405 "Not allowed.";
}
# Purge for the current host using reg-ex from X-Purge-Regex header
purge("req.http.host == " req.http.host " && req.url ~ " req.http.X-Purge-Regex);
error 200 "Purged.";
}
}
Then let's create a Plone view which will make a request from Plone to Varnish (upstream localhost:80) and issue PURGE command. We do this using Requests Python lib.
Example view code:
import requests
from Products.CMFCore.interfaces import ISiteRoot
from five import grok
from requests.models import Request
class Purge(grok.CodeView):
"""
Purge upstream cache from all entries.
This is ideal to hook up for admins e.g. through portal_actions menu.
You can access it as admin::
http://site.com/@@purge
"""
grok.context(ISiteRoot)
# Onlyl site admins can use this
grok.require("cmf.ManagePortal")
def render(self):
"""
Call the parent cache using Requets Python library and issue PURGE command for all URLs.
Pipe through the response as is.
"""
# This is the root URL which will be purged
# - you might want to have different value here if
# your site has different URLs for manage and themed versions
site_url = self.context.portal_url() + "/"
headers = {
# Match all pages
"X-Purge-Regex" : ".*"
}
resp = requests.request("PURGE", site_url + "*", headers=headers)
self.request.response["Content-type"] = "text/plain"
text = []
text.append("HTTP " + str(resp.status_code))
# Dump response headers as is to the Plone user,
# so he/she can diagnose the problem
for key, value in resp.headers.items():
text.append(str(key) + ": " + str(value))
# Add payload message from the server (if any)
if hasattr(resp, "body"):
text.append(str(resp.body))
More info
Varnish can do round robin load balancing internally. You want to distribute CPU intensive load between several ZEO front end client instances each listeting to its own port.
Example:
# Round-robin between two ZEO front end clients
backend app1 {
.host = "localhost";
.port = "8080";
}
backend app2 {
.host = "localhost";
.port = "8081";
}
director app_director round-robin {
{
.backend = app1;
}
{
.backend = app2;
}
}
sub vcl_recv {
if (req.http.host ~ "(www\.|www2\.)?app\.fi(:[0-9]+)?$") {
set req.url = "/VirtualHostBase/http/www.app.fi:80/app/app/VirtualHostRoot" req.url;
set req.backend = app_director;
}
The source code of this file is hosted on GitHub. Everyone can update and fix errors in this document with few clicks - no downloads needed.
For basic information about updating this manual and Sphinx format please see Writing and updating the manual guide.