Optimize your eZ Publish with Varnish

GLOBAL AGENCY : e-communication e-technology e-systems

eZ Summer Camp 2012
eZ Publish Varnish Integration
How to optimize your eZ Publish with Varnish Reverse
Proxy

eZ Summer Camp – eZ Publish Varnish Integration – Sébastien Morel 06/09/2012 1

Novactive
About us


Novactive
PARIS OFFICE
NOVACTIVE, in a nutshell
42/44 rue du Paradis
Nicolas GENTY Bruno BASTET Damien LHOUILLIER 75010 Paris
Director VP of WW Sales CTO France

MONTREAL OFFICE
TEAM (75 employees) Key Figures 360 rue Saint Jacques
!  Top management : 3 !  Design : 5 !  Creation date: 1996 Ouest - Suite 1805
!  Sales : 9 !  Net Worth (Dec10): $486K Montréal, QC H2Y 1P5
!  Consulting & Projet management : 19
!  Engineering and
!  Sales (2011): aprox: $5M Canada
Development : 38
!  Workforce 2012: 75
3 locations : Paris, Montréal et Tunis PARTNERSHIP & EXPERTISE
Platinum eZPublish
Partner

MAIN CUSTOMERS WHAT WE DO?
Varnish partner
Consulting & Expertise: Consulting,
support design and implementation for
Emailvision partner
Web & Mobile, Design, Online Marketing
certified ISO 9001 v.
Web and mobile development: Web:
2008
Expertise in cutting-edge Open Source In 2009, 2010 and 2011
technologies, mobile iPhone and Android
Development Expertise on mobile
(Web and native Apps)


About me

Engineering and Innovation Director @Novactive

!   Quick summary
! Since 2003 at Novactive
! Since 2005 on eZ Publish
! Since 2010 on iOS
! Since 2010 with Varnish

! Twitter : Plopix
!   Blog : http://blog.plopix.net
!   Mail : seb@novactive.com


!   Novactive
!   Reverse Proxy
! Varnish eZ Publish Architecture
! Varnish Basics
! Varnish Advanced #1 Summary
!   ESI Basics
!   ESI with eZ Publish
! Recursive ESI
! Varnish purge on eZ Publish publication
! Varnish Advanced #2


Summary

Reminder of basics
-  Architecture
!   Novactive -  Varnish
-  VCL
!   Reverse Proxy -  Varnish Flow
! Varnish eZ Publish Architecture -  MISS / HIT definition
! Varnish Basics -  Installation and Configuration

!   ESI Basics
! Recursive ESI


Summary

VCL tips and tricks
-  Debug
!   Novactive -  ACL
!   Reverse Proxy -  C code
-  Implementations
! Varnish Basics
!   ESI Basics
! Recursive ESI


Summary

Edge Side Include
-  Defintion
!   Novactive -  Tips and tricks
-  ESI content view
!   Reverse Proxy
-  Recursive usage
! Varnish eZ Publish Architecture -  Implementations
! Varnish Basics
!   ESI Basics
! Recursive ESI


Summary

Instant publishing
-  Tricks
!   Novactive -  Unleash the TTL
!   Reverse Proxy -  Varnish purge
-  Implementation
! Varnish Basics
!   ESI Basics
! Recursive ESI


Summary

Varnish daemon
-  Tips and Tricks
!   Novactive -  VCL advanced
!   Reverse Proxy -  Performance optimization
! Varnish eZ Publish Architecture -  Troubeshoot your instance
-  Implementation
! Varnish Basics
!   ESI Basics
! Recursive ESI


!   Novactive
!   Reverse Proxy
! Varnish Basics
!   ESI Basics
! Recursive ESI


Reverse Proxy
General


Reverse Proxy features

Features
! Handles a HTTP Cache
! Handles ESI (Edge Side Include)

Advantages
! Optimize the usage of your webserver
! Therefore of your database
! Allows us to handle high peaks of traffic

Disadvantages
!   Works on a short TTL (Time To Live)
! Delaying the publication with 2xTTL time
! Induced time lag of contents
!   The permanent dilemma:
!   The greater the TTL is, the older the information could be
!   The more you reduce the TTL, the less you handle high traffic


Reverse Proxy

Main Open Source Reverse Proxy in the world

! Squid

!   Apache

! Lighthttpd

! Nginx

! Varnish


Reverse Proxy architecture

Architecture

!   Clients
! Resources who issues the request

! Backends
!   The different webservers behind Varnish

! Backend storage
!   The cache storage method

!   Configuration
!   Set of rules to manage application



Basic platform integration

Web Server

ReverseProxy Web Server

Visitors
Web Server

!   You can see a reverse proxy as a shield of your web server
farm.



Generic diagram (MISS and HIT)
Request If not in the cache

backend response backend response

Request

backend response

Visitors Reverse Backend
Browsers Proxy Apache
Clients Web server


!   Novactive
!   Reverse Proxy
! Varnish Basics
!   ESI Basics
! Recursive ESI


eZ Publish without Varnish
Classic eZ Publish Architecture


Classic architecture

Visitors

eZ
Contributors

MySQL


eZ Publish with Varnish
Classic eZ Publish with Varnish


Classic Architecture + Varnish

Visitors

Varnish

eZ
Contributors

MySQL


eZ Publish Cluster without Varnish
eZ DFS Classic eZ Publish Architecture


eZ DFS Classic architecture

Visitors

Load Balancer

eZ eZ eZ eZ

Contributors

eZ DFS

NFS MySQL


eZ Publish Cluster with Varnish
eZ DFS eZ Publish with Varnish


eZ DFS Architecture + Varnish

Load Balancer

Visitors
Varnish Varnish

eZ eZ eZ eZ

Contributors

eZ DFS

NFS MySQL


!   Novactive
!   Reverse Proxy
! Varnish Basics
!   ESI Basics
! Recursive ESI


Varnish basics
Configuration


Varnish Basics

Installation

Configuration in service mode
!   /etc/default/varnish
!   Service global configuration

!   /etc/varnish/yourconfiguration.vcl
!   All your rules to handle HTTP cache in your application

Global configuration
!   Socket port and socket admin port
!   Memory
! Process limitation
!   Default TTL
! VCL file path


Varnish Basics

VCL : Varnish Configuratin Language
! Syntax from C and Perl
!   No loops, no variables
=> It describes the caching policy !

Syntax
!   # and /* foo */ for comments
!   sub $name functions

Functions
! regsub(str, regex, sub) / regsuball(str, regex, sub)
! ban_hash(regex) / ban_url(regex) / ban (expression)
!   restart
!   return()

Based on a Request Flow


Simplified Varnish request flow


Varnish flow in diagram

Macro Simplified Varnish Flow : MISS
Request
vcl_recv

vcl_hash

Is this request in
my cache ?

Request page
generation

vcl_fetch

vcl_deliver

backend response
Visitors Varnish Backend
Browser Apache
Clients Web server


Varnish flow in diagram

Macro Simplified Varnish Flow : HIT

Request
vcl_recv

vcl_hash

Is this request in
my cache ?

vcl_deliver

backend response

Visitors Varnish Backend
Browsers Apache
Clients Web server


Varnish Flow

Recv
! Called at the beginning of a request, after the complete request
has been received

Fetch
! Called when the requested object has been retrieved from the
backend

Deliver
! Called before a cached object is delivered to the client

Hash
! Called to define the unicity of a cached object

Error
! Called when an error occurs


!   Novactive
!   Reverse Proxy
! Varnish Basics
!   ESI Basics
! Recursive ESI


Varnish advanced #1
VCL: Varnish Configuration Language


Varnish advanced #1

Debug information
!   Usually added in the headers
!   But, in the response headers !

HIT or MISS ?
!   Most important information
!   In the vcl_deliver
# Called before a cached object is delivered to the client !
sub vcl_deliver {!
if (obj.hits > 0) {!
set resp.http.X-Cache = "HIT”;!
set resp.http.X-Cache-Hits = obj.hits;!
} else {!
set resp.http.X-Cache = "MISS";!
}!
set resp.http.WhoisCache = "eZ Summer Camp Conf – Varnish 1”!
return(deliver);!
}!

!   You should also add
!   the “hits” counter
!   the name of the Varnish server ( if you are in a multi server architecture)


Varnish advanced #1

MISS ? Why ?

!   Different ways to have a MISS
!   It’s the first call after an expiration
!   Because it’s a rule in the vcl_recv
!   return(pass)
!   Because it’s a rule in the vcl_fetch
!   return(hit_for_pass)

!   How to find out why you have a MISS
!   In the recv
!   Only access to the request (req)
!   In the fetch
!   You have access to the response (beresp)

!   Solution ?
!   Add Header into the request in the recv function
!   Copy these headers into the response in the fetch function


Varnish advanced #1

MISS ? Why ?
!   In the vcl_recv

# called at the beginning of a request, after the complete request has been received!
sub vcl_recv {!
!
… some rules …!
!
if (req.request == "POST") {!
set req.http.X-Debug = "Not Cached according to configuration (POST)"; }!
return(pass);!
} !

!   In the vcl_fetch
# Called when the requested object has been retrieved from the backend!
sub vcl_fetch {!
!
… some rules …!
!
# show our debug "!
set beresp.http.X-Debug-recv = req.http.X-Debug;!
!
… some rules …!
!
# You are respecting the Cache-Control=private header from the backend!
if ( beresp.http.Cache-Control ~ "private") {!
set beresp.http.X-Debug-fetch = "Not Cached according to configuration (Cache-Control)”;!
set beresp.ttl = 0s;!
return(hit_for_pass);!
}!
!


Varnish advanced #1

Add ACL
!   Everybody can see these headers
!   Not very secure
# ACL for debug IP!
acl debuggers {!
"localhost";!
"192.168.0.0"/16;!
}!
!

!   Usage

# pass mode can't handle POST (yet) so use pipe!
if (req.request == "POST") {!
if (client.ip ~ debuggers) {!
set req.http.X-Debug = "Not Cached according to configuration (POST)”;!
return(pass);!
}!
} !


Varnish advanced #1

Add C in your VCL
!   Useful to do very specific stuff

C{!
You C code!
}C !

!   You have to see into the source code of Varnish to know the valid
method

if( beresp.http.X-ttl ~ "s$"){ # seconds!
C{!
char *ttl;!
ttl = VRT_GetHdr(sp, HDR_BERESP, "06X-ttl:");!
VRT_l_beresp_ttl(sp, atoi(ttl));!
}C!
}!


!   Novactive
!   Reverse Proxy
! Varnish Basics
!   ESI Basics
! Recursive ESI


Edge Side Include
Definition


Edge Side Include

Definition

! markup language
!   not interpreted by the web server

!
<esi:include src="http://example.com/1.html"/>

!   Inclusion of pages fragments
!   Assembled on the edges
!   On reverse proxy (Varnish, Akamai, or other CDNs)

!   Enable ESI in the VCL fetch method

set beresp.do_esi = true; !


Edge Side Include

Be careful

!   Do not parse ESI in images and binary content

!   Self-closing XML element

!   ESI will be ignored in pipe (see Varnish Flow)

!   ESI requests are considered as isolated requests
!   Don’t forget that !


!   Novactive
!   Reverse Proxy
! Varnish Basics
!   ESI Basics
! Recursive ESI


ESI with eZ Publish
A cache-block replacement


ESI with eZ Publish

Objectives

!   Managing cache-block

!   Varnish became an applicative part of the platform

!   ESI System
!   Deleting cache-block and using benefit of ESI views

!   Update these caches (old cache -block) with the ezpublish view
caching system.

!   Let’s introduce the ESI content view...


ESI with eZ Publish

Standard pagelayout : With cache-block
Pagelayout

Header (cache-block)

Menu Sidebar
(cache-block) (cache-block)

View

Footer (cache-block)


ESI with eZ Publish

ESI pagelayout: Without cache-block
Pagelayout

Header (/content/view/header/2)

Menu Sidebar
(/content/view/menu/N) (/content/view/sidebar/
N)

View

Footer (/content/view/footer/2)


ESI with eZ Publish

Cache-block reminder performance and usage

!   ignore_content_expiry
!   WITHOUT : each publication expires all your cache-block
!   WITH : you have to wait the TTL (expiry)

!   subtree_expire
!   Not very useful in most case
!   Needs cronjob

!   keys
!   two templates with same keys will generate at least 2 cache-block

!   In general, cache-block can be a performance killer !


ESI with eZ Publish

ESI content view

!   Forget the key concept of the cache-block

!   Think about
!   content view
!   view cache system
!   smart view cache system

!   For now, each ESI content view will have short TTL
!   But keep in mind, that it won’t always be the case


ESI with eZ Publish

ESI content view VS cache-block

! When the TTL is expired, there is no generation
!   You get the view cache

!   You can easily clear a view
!   For now you have to wait the Varnish TTL

!   You can use these views everywhere
!   In another view, or in the pagelayout !

!   And if you want a TTL regeneration, don’t add the view in your
CacheViewModes


How to do better ?


!   Novactive
!   Reverse Proxy
! Varnish Basics
!   ESI Basics
! Recursive ESI


Recursive ESI with eZ Publish
Simplify your code


Recursive ESI with eZ Publish

Concept

! Varnish supports recursive ESI inclusion

!   Limited by the max_esi_includes parameter

! Very useful when you have to browse a tree

! It’s the one way to use ESI content view


Can we do better ? Again?
Yes !


!   Novactive
!   Reverse Proxy
! Varnish Basics
!   ESI Basics
!   ESI for eZ Publish
! Recursive ESI


Varnish purge on eZ Publish
publication
Unleash your TTL


Varnish purge on publication

Concept

! When eZ Publish expires a content view, forward this expiration to
Varnish

!   How ? 2 ways to send a purge request
!   Use the ezpEvent « content/cache »
!   Use a StaticCacheHandler

!   Configure the vcl to handle this request and expire the cache
object.

!   Use the Smart View Cache to connect the different ESI view purge

! After that, you can increase your TTL !
!   To an unllimited value


Varnish purge on publication

First, how to handle a purge request
!   In the vcl_recv

if (req.request == "BAN") {!
# ACL check and quick return (error)!
if (!client.ip ~ purgers) {!
error 405 "Method not allowed";!
}!
" "!
if (req.url ~ "^/url/") {!
ban_url(regsub(req.url, "^/url/", ""));!
error 200 "Purge of objects with url ("+regsub(req.url, "^/url/", "")+") done.";!
}!
}!


!   Novactive
!   Reverse Proxy
! Varnish Basics
!   ESI Basics
!   ESI for eZ Publish
! Recursive ESI


Varnish advanced #2
VCL advanced


Varnish advanced #2

Pool of backend
!   random
director www_director random {!
{!
.backend = web1;!
.weight = 1;!
} "!
{!
.backend = web2;!
.weight = 1;!
}!
}!
!

!   Don’t forget to add the second backend

backend web2 {!
".host = "localhost";!
".port = "80";!
} !

!   Change the backend

set req.backend = www_director;
! !

!


Varnish advanced #2

Check the heath of your backend

!   Poke your backend every N seconds
!   Affects backend selection
backend web1 {!
.host = "localhost";!
.port = "80";!
.probe = {!
.url = "/";!
.interval = 3s;!
.window = 5;!
.threshold = 2;!
}!
}!

!   Varnish doesn’t send Host with health checks.
.probe = { !
.request =!
"GET / HTTP/1.1"!
"Host: www.foo.bar"!
"Connection: close";
!

!   Don’t set the eZ Publish home page for the probe.


Varnish advanced #2

Check the heath of your backend

!   req.grace (in the vcl_recv)
!   accept serving 30 minutes old object
# set the grace!
set req.grace = 30m;
!

!   beresp.grace (in the vcl_fetch)
!   keep all objects for 30 minutes beyond their TTL

# set the grace!
Set beresp.grace = 30m;!

!   It’s like the “Stale cache”


Varnish advanced #2

RewriteRule
!   Varnish can do it

sub vcl_recv {!
!
… some rules …!
!
if ( req.url ~ "^/layout/(.[^/]*)/([0-9]*)(.*)$" ) {!
set req.url = regsub( !
" " "req.url, !
" " ""^/layout/(.[^/]*)/([0-9]*)(.*)$", !
" " ""/layout/set/1/content/view/full/23”!
" ");!
}!
!
… some rules …!
!
} !


Varnish advanced #2

Do a 301 HTTP redirection

sub vcl_recv {!
!
… some rules …!
!
if (req.http.host ~ "agreatbutlongdns.ezsc" ) {!
error 301 "ezpublish4.ezsc"; #see vcl_error!
}!
… some rules …!
!
}!
!
sub vcl_error {!
… some rules …!
if (obj.status == 301) {!
set obj.http.Location = "http://" + obj.response + ":" + server.port + req.url;!
} else {!
… some rules …!
}!
}!
!
!


Varnish advanced #2
Usage


Varnish advanced #2

varnishlog


Varnish advanced #2

varnishlog

!   quite extensive

!   filter information you need

!   Useful to debug what happens
!   And valid your configuration

Best options
!   –o : group by request
!   -b : only show traffic related to a backend
!   -c : only show traffic related to a client
!   -i <tag> : only show one tag


Varnish advanced #2

varnishtop

!   Useful to know some real-time information

!   top URLs sent to the client

!   top URLs sent to the backend


Varnish advanced #2

varnishncsa

!   It allows you to log client access as Apache do

!   NCSA log file example

127.0.0.1 - - [29/Aug/2012:03:50:02 +0200] "GET http://ezpublish4.ezsc:8080/Getting-Started/Selected-
Features HTTP/1.1" 200 4314 "-" "JoeDog/1.00 [en] (X11; I; Siege 2.70)"!
127.0.0.1 - - [29/Aug/2012:03:50:02 +0200] "GET http://ezpublish4.ezsc:8080/content/view/tagcloud/2
HTTP/1.1" 200 3371 "-" "JoeDog/1.00 [en] (X11; I; Siege 2.70)"!
127.0.0.1 - - [29/Aug/2012:03:50:02 +0200] "GET http://ezpublish4.ezsc:8080/content/view/sitemap/2
HTTP/1.1" 200 3187 "-" "JoeDog/1.00 [en] (X11; I; Siege 2.70)"!
127.0.0.1 - - [29/Aug/2012:03:50:02 +0200] "GET http://ezpublish4.ezsc:8080/content/advancedsearch
HTTP/1.1" 200 4505 "-" "JoeDog/1.00 [en] (X11; I; Siege 2.70)"}!
!
!


Varnish advanced #2

varnishstat

!   Perfect tool to get a good representation of your Varnish health

!   Some useful statistic
!   Uptime
!   HIT rate
!   Client connections accepted
!   Client connections received
!   Cache HIT
!   Cache MISS
!   Failed backend connections


Varnish advanced #2

varnishstat


Varnish advanced #2

! Find the URL the most sent to the backend

Varnistop –i TxURL!
!
!

! Find the URL the most asked by the client

Varnistop –i TxURL!
!
!

!   Have you try the « -j » option of varnishstats
!   new in 3.03 since 08/20/12


Varnish advanced #2
Daemon configuration


Varnish advanced #2

BAN Console
!   Access
!   varnishadm
!   telnet localhost 6082

!   Tunable paramaters
!   param.show –l

!   Be careful about the thread parameters
! 
root@ezsc ~ # varnishadm param.show | grep thread!
cc_command "exec gcc -std=gnu99 -g -O2 -pthread -fpic -shared -Wl,-x -o %o %s"!
thread_pool_add_delay 2 [milliseconds]!
thread_pool_add_threshold 2 [requests]!
thread_pool_fail_delay 200 [milliseconds]!
thread_pool_max 2000 [threads]!
thread_pool_min 1 [threads]!
thread_pool_purge_delay 1000 [milliseconds]!
thread_pool_stack 65536 [bytes]!
thread_pool_timeout 120 [seconds]!
thread_pool_workspace 16384 [bytes]!
thread_pools 2 [pools]!
thread_stats_rate 10 [requests]!


Varnish advanced #2

Threading parameters
!   Number of threads
!   how many requests Varnish can serve concurrently

!   Minimum threads running = thread_pools * thread_pool_min
!  2 * 1 = 2
!   Maximum threads running = thread_pool_max
!   2000

!   Don’t forget to add your configuration in the default configuration
!   /etc/default/varnish


Varnish advanced #2

Real-time configuration changes

!   you can change a param with varnishadm
!   but don’t forget to add the new parameter value in the daemon file configuration

!   About the VCL, be careful because a restart empties the cache
!   Varnish don’t keep cache
!   very bad idea in a production environment !

!   So let’s use varnishadm in order to reload the VCL


To sum up…


To sum up

On the network administration side, you know

!   How to set architecture of your plateform

!   How to configure your Varnish

!   How to optimize the daemon

!   How to optimize your VCL

!   How to troubleshoot your instance


To sum up

On the developer side, you know

!   How to make a good conception with ESI content view

!   How to remove cache-block and unleash your instance

!   How to purge cache on publication


Conclusion


Conclusion

Achieved objectives
!   You can now have a production website with
!   98% of requests cached by Varnish
!   2 % of traffic handle by your webserver farm

!   Set a architecture with
!   A database dedicated to the backoffice and to the generation on new content
!   A very low usage of your NFS

! Beyond speed, more security
! Your website is all the time available due to the shield protection of Varnish
! even if we have server problems


eZ Accelerator

To go further, discover eZ Accelerator available on the eZ Market

Key Features
!   Full support of URLs
!   URL Translator, Siteaccess, PathPrefix, StaleCache etc..
! Asynchronous purge
!   Multi varnish architecture
!   Cache purge on content publication
! Dedicated interface to manage your Varnish servers


What about eZ Publish 5

eZ Publish 5 with Varnish
! Varnish on Symfony 2
! Totally integrated with the HTTP Cache system
!   ESI are directly handled by the template and cache system

! Two caching models
!   Expiration model
! Based on the TTL
!   Validation model
! Based on eTag

! Always the same rule : Never generate the same response twice !

!   So all the concept seen on this workshop are relevant with eZ Publish 5


Questions ?


Thank you!


Adresse : 42-44, rue de Paradis
75010 PARIS

Téléphone : 01.48.24.33.60
Fax : 01.48.24.33.54

Email : info@novactive.com
Site : www.novactive.com

SARL au capital de 132.576 euros
RCS Paris B 408 999 233


Optimize your eZ Publish with Varnish

More Related Content

Similar to Optimize your eZ Publish with Varnish (20)

Recently uploaded (20)

Optimize your eZ Publish with Varnish