accelerate your web app with a layer of varnish

Download Accelerate your web app with a layer of Varnish

If you can't read please download the document

Upload: jeremy-cook

Post on 16-Apr-2017

3.896 views

Category:

Technology


0 download

TRANSCRIPT

Accelerate your app with a layer of Varnish

Who?

Ex-pat Englishman, now living in Southern Ontario.

Web developer for 5 years, mostly PHP.

(Almost) Senior Software Engineer at TribeHR.

Co-organiser of Guelph PHP User Group.

Ex-professional musician.

Varnish can do incredible things for your app...

Mention jaw dropping nature of it...Talk is an intro to what varnish is, what it can do and when it is (as well as is not) the right tool for the job.

What is Varnish?

Open source caching reverse proxy or HTTP accelerator.

Mature software:Current version: 3.0.5

Varnish 4 currently in tech preview.

Used by 5% (500) of the top 10,000 sites on the internet.

Explain what a caching reverse proxy is.Varnish users: BBC, Synacor, Slideshare, Facebook(?), Twitter(search), Forbes.com and NY Times.Mention that I have stories about how these organisations are using Varnish but no time...

Varnish basics

Listens on a port (normally 80).

When a request arrives:If content is already in cache it is served directly.

If not Varnish forwards the request to a 'backend' server, delivering the content when it is received.

Only caches GET and HEAD requests.

Varnish's definition of a backend: any http server that can produce content that can potentially be cached.Safe http methods.

Varnish features

Can cache any kind of content delivered via HTTP.

Blazingly fast with incredible performance.

Support for multiple backends and basic load balancing.

Mention html, json, xml, css, javascript, images, media files, etc.Developers claim speed improvement of 300-1000xDev's also claim seeing Varnish deliver 20GBPS on off the shelf hardware...Claim that you will typically max out your internet connection before exhausting Varnish's ability to serve content from cache.

Varnish features (cont)

Configuration through Varnish Configuration Language (VCL).

Support for Edge Side Includes (ESI).

Behaviour can be extended through Varnish Mods (VMODs).

Includes tools to monitor and tune (varnishtop, varnishlog, varnishhist, varnishstat, etc).

Show us the demo!

https://github.com/JCook21/VarnishDemo

Mention some people have had problems under OSX.

Apache results

Varnish Results

Is Varnish always the right tool?

What kind of content are you serving?

Cookies.

SSL.

Varnish is like a layer of... varnish.

Do you really need Varnish?

Getting started with Varnish

Setting cache times

Two possible methodsHTTP cache headers

Directly in VCL

Prefer HTTP cache headers wherever possible.

Varnish looks for value in s-maxage or maxage in cache-control. If neither set looks for Expires header. Then looks in VCL. If nothing set then uses the default_ttl runtime option.Can be overridden in VCL.Varnish ignores request cache headers.Will only cache:200: OK203: Non-Authoritative Information300: Multiple Choices301: Moved Permanently302: Moved Temporarily307: Temporary Redirect410: Gone404: Not Found

Supported OS's

Fully supported:Linux

FreeBSD

'Try not to break':NetBSD

OpenBSD

OS X

Needs a 64 bit OS ideally.

Mention that 32 bit installs of supported versions are possible but limited memory.If your app runs on another OS there's nothing to stop you putting Varnish in front of it on a supported OS...

Installing Varnish

Use system package manager.For Debian, Ubuntu and RedHat/CentOS Varnish software maintains repositories with up to date versions.

To obtain up to date version to install on non-core supported platforms compile from source.

More info at the Varnish website.

Basic configuration

Ships with an excellent, 'sane' default VCL configuration.

To begin using Varnish:Configure runtime options (/etc/default/varnish).

Add one or more backends to VCL file (/etc/varnish/default.vcl).

Lego analogy.-a port Varnish listens on.-f VCL file-s type and size of storageMention that runtime options are passed to varnishd whent started, normally set in defaults file.

Adding VCL backends

Must add at least one backend in VCL.

If more than one, first will be the default.

Backends can be grouped into 'directors' to apply load balancing.

Mention that VCL allows you to decide which backend to use when a request is received.Also mention Varnish supports several types of load balancing.

Sample backend definition

backend default { .host = "127.0.0.1"; .port = "8080";}

Advanced configuration

Introduction to VCL

Provides a window into Varnish's state during the request/response cycle.

Not a full programming language.

C/Perl-like syntax.

Compiled into a C shared object at runtime and loaded into Varnish.

Mention that issuing Varnish reload simply reloads the config.Server not stopped so cache not purged.Existing items in cache will keep the existing cache time!Reload does nothing if VCL invalid, which is good...Mention no support for user variables or loops.

VCL Subroutines

Dispatched at different times in the request/response cycle.

Each routine has default code:Varnish appends this to the end of any code you add.

Can be short-circuited by using the return keyword. Be careful!

Allowed return values documentation.

You can also define your own.

Note that you can't pass arguments to subroutines in VCL.

VCL Subroutines (cont)

vcl_init

vcl_recv

vcl_pipe

vcl_pass

vcl_hash

vcl_hit

vcl_miss

vcl_fetch

vcl_deliver

vcl_error

vcl_fini

Only give an overview!!

VCL Processing Cycle

Also mention pipe along with pass.

VCL Functions

Functions Varnish makes available to you:hash_data(str)

regsub(str, regex, sub)

regsuball(str, regex, sub)

ban(ban expression)

ban_url(regex) (Deprecated)

purge()

return()

Also have set and unset keywords to manipulate variables and call to dispatch your own subroutines.

hash_data: adds a string to the hash input.regsub: replaces first regex match with the subregsuball: replaces all regex matches with subban: bans all objects in cache that match the expression.ban_url: bans all objects in cache that match the regex. DEPRECATED.Ban is a filter on objects in the cache. Ban things from being served from cache but doesn't stop new items entering the cache.

vcl_recv

Called once the OS kernel has given the complete request to Varnish.

Typical uses:Set the backend to use.

Amend the request.

Add extra information to the request.

Force a bypass of cache lookups (pass or pipe).

vcl_fetch

Called when a response is received from a backend, before it is considered for inclusion in the cache.

Typical uses:Amend or set a TTL.

Add or remove HTTP headers.

Detect errors (more later).

Turn on ESI processing.

Mention that Varnish will set a ttl before this subroutine is called.

Select a backend

backend default { .host = "192.168.2.1"; .port = "8080"; }

backend bar { .host = "192.168.2.2"; .port = "8080":}

sub vcl_recv { if (req.http.host ~ "bar.com") { set req.backend = bar; } // etc}

Remove cookies

sub vcl_recv { if (req.url !~ "^/admin/.*") { unset req.http.cookie; }}

sub vcl_fetch { if (req.url !~ "^/admin/.*") { unset beresp.http.set-cookie; }}

Set default TTL

sub vcl_fetch { if (! beresp.http.cache-control && beresp.http.content-type ~ "text/html|text/xml|application/json") { set beresp.ttl = 5m; }}

Note that this example is overly simplistic!

VMOD's

Allow you to extend Varnish's features.

Written in C.

Community VMOD's athttps://www.varnish-cache.org/vmods

Examples:Geo-IP lookup (BBC)

Query string normalisation

Memcache

Grace mode

Graced content is an object that has expired but is kept in cache.

Grace mode is when a graced object is used.

When this can happen:No healthy backends are available.

Deliver old content while fetching new content is pending.

Varnish queues multiple requests for the same content.Allows Varnish to stop requests for content piling up while fetching content from a backend.To use graced content for unhealthy backends health checks must be set.

Configuring grace

sub vcl_recv { if (req.backend.healthy) { set req.grace = 30s; } else { set req.grace = 1h; }}

sub vcl_fetch { set beresp.grace = 1h;}

Configure req.grace in vcl_recv and beresp.grace in vcl_fetch.req.grace can be set dynamically depending on the health of a backend.Typically store an object for several hours past ttl.Only use it for a few seconds past the ttl unless the backend is sick.

Health checks

backend foo { .host = "192.168.2.1"; .port = "8080"; .max_connections = 10; .probe = { .url = "/test"; .interval = 5s; .timeout = 1s; .window = 5; .threshold = 3; } }

Configured with url, interval, window, threshold.Interval is how often to probe.Threshold and window set limits, eg. Threshold probes must pass in window probes.

Saint mode

Allows you to say that a backend is 'sick' for one url.

Varnish will not make further requests to that backend for the url for the time specified.

Gives you a chance to detect errors in responses.

Saint mode example

sub vcl_fetch { if (beresp.status ~ ^5\d{2}$) { set beresp.saintmode = 10s; return(restart); } }

Used to mark a single object on a single backend as sick for a period of time.If n (default 10) objects are marked as sick whole backend will be marked as sick.Can trigger grace mode.

Edge Side Includes

Allows different content on the same page to be cached for different times.

Request flow:External request returns a page with special ESI tags in it.

For each ESI tag Varnish issues a sub request which generates the content.

Both 'master' and sub requests can be cached.

Note that Varnish does not implement the complete ESI specification.

Configuring ESI's

Done in vcl_fetch:set beresp.do_esi = true;

Need to configure app to support 'layouts' and 'components'.

Various frameworks have support for ESI's.

Mention SF2 native support for ESI.

ESI Example

ESI example

Thanks for listening!

Varnish docs: https://www.varnish-cache.org/docs/3.0/tutorial/index.html

Any questions?

Feel free to contact me:@JCook21

[email protected]