[Varnish] 4.x Hit / Miss header

With version 4.x the classic snippet:

sub vcl_deliver {
        if (obj.hits > 0) {
                set resp.http.X-Cache = "HIT";
        } else {
                set resp.http.X-Cache = "MISS";
        }
}

doesn’t work anymore, because obj.hits no longer reports the number of hits for a cached object. In order to add a custom header indicating object origin, you must using a little trick with the X-Varnish header value.

A non-cachable or freshly added object will have only one hash. An object retrieved from the cache several hashed. This snippet use this property:

sub vcl_deliver {
  if (resp.http.X-Varnish ~ "[0-9]+ +[0-9]+") {
    set resp.http.X-Cache = "HIT";
  } else {
    set resp.http.X-Cache = "MISS";
  }
}

[Varnish] 4.x HTTP Authentication

With the 4.x branch, the vcl syntax have change significantly, rendering this post obsolete.

Now the vcl_recv block, must look like this:

if (!req.http.Authorization ~ "Basic XXXXXXXXXXXXX"){
   return(synth(401, "Authentication required"));
}

Also vcl_error doesn’t exist anymore. It has been splitted into vcl_backend_error for as it name implies and vcl_synth for all “custom-made” error code. The block to add into vcl_synth look like this:

if (resp.status == 401) {
   set resp.status = 401;
   set resp.http.WWW-Authenticate = "Basic";
   return(deliver);
}

As usual you can add your own HTML code inside this function.

[Varnish] 4.x health command

With previous varnish version, the “undocumented-command-to-know” to check backend health was:

varnishadm debug.health

A return code 200 was great and any other value bad. Problem: this command basically give no details on the probe setting and it success rate.

With version 4.x varnish the command has changed. Now it’s:

varnishadm backend.list

and this time it not only give you a list of each backend but also the associated probe success rate.

[Varnish] varnishstats

varnishstat is a tool used to monitor the basic health of Varnish. Unlike others it doesn’t read log entries, but displays statistics from a running varnishd instance. It can be used to determine request rate, memory and thread usage.

Output

Like top the command output is in real-time. Data are displayed in a table form. The first column is the raw data of the counter. For example, in case of the ‘cache_hit’ counter, this is the total number of cache hits since varnishd was started. The second column is the counter change per second. The third is the average change per second since varnishd was started. The three next are the same except with larger time scale (10, 100 and 1000 seconds respectively).

Note that you can use the option -1 for non ‘interactive’ use. In this case varnishstat list all stats and quit immediately.

Interesting counters

There is a lot of counters to look after, but keys metrics can be divided into 4 categories:

  • Client: client connections and requests
  • Cache: cache hits, evictions
  • Thread: thread creation, failures, queues
  • Backend: success, failure and health

Client metrics

client_req: this counter display the number of requests you’re receiving per unit of time. Monitoring this metric can alert you of spikes in incoming web traffic, whether legitimate or nefarious.

sess_dropped: once Varnish is out of worker threads, each new request is queued up and this counter is incremented. When the queue is full, new incoming requests will be dropped and this counter will also be incremented. If sess_dropped isn’t equal to zero, either your varnish is overloaded or it thread pool is too small.

Cache performance metrics

Using the cache_hit and cache_miss counter, you can calculate the cache hit ratio:

ratio = cache_hit / (cache_hit + cache_miss)

This derived metric provides visibility into the effectiveness of the cache. The higher the ratio the better. A ratio above 0.7 is considered as ‘good’. If your ratio is ‘bad’ you should check which objects aren’t cached and why.

n_lru_nuked: the LRU (Least Recently Used) nuked objects counter should be watch closely. If the counter value increase a lot that probably means varnish is evicting objects at a faster rate then usual because of memory shortage. In this case you may want to increase the cache size if possible.

Thread metrics

threads_failed: the number of times varnishd unsuccessfully tried to create a thread. A value greater then zero likely indicate you reach the server limits. It could also append if you try to spawn a huge number of thread in a short time. The latter case usually occurs right after varnish is started, and can be corrected by increasing the thread_pool_add_delay value.

threads_limited: number of times a thread needed to be created but couldn’t because varnishd already maxed out its capacity. If you have a value greater the zero and still have available resources left, you should increase the thread_pool_max value.

Backend metrics

backend_fail: number of backend connection failures. This counter should be very close to zero. If it’s not the case, it could means you have:

  • network issues
  • overloaded/laggy backend (time to first byte or between bytes exceeded)
  • unresponsive backend

Further Reading and sources

Generality on PHP

The content of this post was written by myself as an entry at my work wiki. It’s intended for new administrator with few knowledge of web environment.

What is PHP ?

PHP is a server-side scripting language designed for web development. PHP code may be embedded into HTML code, or it can be used in combination with various template systems, content management system and frameworks. PHP code is usually processed by an interpreter implemented as a module in the web server (like apache mod_php) or as a separated executable (PHP-FPM).

PHP versions

PHP core was rewritten several time, breaking compatibility each time. Since PHP4 the official interpreter is named “Zend Engine”. PHP4/ZendEngine 1.0 was first released on May 2000 and deprecated in August 2008. PHP5/ZendEngine 2.0 was first released in 2004 and is still actively maintained. PHP6 was an experimental branch for implementing unicode support. It’s no longer in use. The next branch is called PHP7/ZendEngine 3.0 and was first released on December 2015.

PHP and webservers

In order to “connect” PHP with your webserver you can use two different system :

  • an “embedded” interpreter: mod_php for Apache HTTPd, isapi for IIS, etc..
  • an external fast-cgi process: php-fpm

Embedded interpreter was the historical choice but since PHP5.5 it’s recommended to use the fast-cgi approach. In each case you need to load the appropriate module on your webserver.

+ Embedded interpreter: mod_php

mod_php is an apache module who load the official interpreter (Zend Engine) inside it. With mod_php the code is executed by apache (user www-data for Debian). It’s not possible to use several unix user for each application. Setting files are loaded only once, at startup. Any change in configuration imply restarting apache. mod_php can also be memory hungry because each new apache child process load the PHP interpreter even if no code is executed.

+ External fast-cgi process: PHP-FPM

PHP-FastCGI Process Manager is a daemon implementing the fast-cgi protocol for PHP. PHP-FPM is the official implementation since PHP5.3.3, superseding other fast-cgi implementation like FCGI, SpawnFCGI, etc…

PHP-FPM is more efficient than mod_php because spawning process is adaptative. FPM can start workers with different uid/gid and different setting files. This allows much greater security and scalability because the webserver and the code interpreter can be split into their own individual server environments if necessary. PHP-FPM can also shared opcode cache across multiple processes.

PHP-FPM runs as a standalone daemon, but you still need module to connect it to your webserver:

  • for nginx : ngx_http_fastcgi_module
  • for apache 2.4 and greater : mod_proxy_fcgi
  • for older apache version : mod_fastcgi (this setup is not recommended)

HHVM

Beside the official interpreter (Zend Engine) several implementation of PHP exist : Pipp, Phalanger, HHVM. These implementations are partial! There is no warranty your application can work with them.

The most interesting alternative implementation is HHVM aka. HipHop Virtual Machine. HHVM actually works on the same principle as the JVM (Java Virtual Machine). HHVM not only translate PHP code into an high-level bytecode (more or less like other opcodes solutions) but also execute it on an JIT compiler. PHP code performance can be increase by a factor 2 to 5 when using HHVM.

PHP accelerator

In order to improve PHP performance several “accelerator” extension have been made. These extensions all work on the same principle: they store the already parsed PHP code into a pseudo-bytecode, called opcode, generally keep in memory. This technique reduce web applications response’s times drastically.

+ APC

For PHP version upto 5.3 APC is the recommended accelerator.
You can adjust the quantity of memory used by apc into /etc/php5/apache2/conf.d/apc.ini. A value of 256Mb is recommended.

It possible but not advised to install APC on PHP 5.4. It has been reported that such configuration can lead to execution errors that break whole applications. APC can’t be install on newer PHP version.

+ opCache

Since version 5.5 PHP has it own built-in accelerator: opCache. You can install and configure OpCache on PHP 5.4 manually. It’s also possible to fine-tune opCache. We usually use these values:

opcache.memory_consumption=256
opcache.interned_strings_buffer=8
opcache.max_accelerated_files=4000
opcache.revalidate_freq=60
opcache.fast_shutdown=1
opcache.enable_cli=1

APCu

In addition to opcode caching APC also offer a way to cache serialized object, pretty much like redis does. This functionality isn’t provided by opCache so a new extension was created: APCu. APCu only provide object caching.

[Varnish] Beware of Vary header

Vary is one of the most powerful HTTP response headers. Used correctly it’s possible to do wonderful things with it. Unfortunately most people used it so badly, that today most CDN don’t use it anymore and simply avoid caching if it used for anything else than HTTP compression.

Using it for HTTP compression

Most people use the Vary header exclusively with the Accept-Encoding value in order to cache several version of the same URI depending on the Accept-Encoding field value. Usual values for Accept-Encoding are gzip, deflate and more rarely sdch, but many more exist. Also keep in mind that any combination of these individual values is valid !

To keep the number of cache version at minimum, you can use the following VCL snippet to normalize the Accept-Encoding header, only keeping gzip and deflate as acceptable values:

if (req.http.Accept-Encoding) {
   if (req.http.Accept-Encoding ~ "gzip") {
      set req.http.Accept-Encoding = "gzip";
   } elsif (req.http.Accept-Encoding ~ "deflate") {
      set req.http.Accept-Encoding = "deflate";
   } else {
      remove req.http.Accept-Encoding;
   }
}

Using it with other field

It possible to specify any other field into the Vary value and also multiple fields by separating their names with a comma. But i don’t recommend this setup because of how CDN treat it.

Anyways, if you want to use Vary correctly you must always keep in mind these two cardinal rules: never use the raw values of the indicated field and never choose a field with too much variant.

For example the following fields should NEVER be used into a Vary value:

  • Referer: because that utterly stupid
  • Cookie: because that probably the most unique request header ever
  • User-Agent: because with more then 8000 usual variants, you pretty to sure to never use your cache anymore

[Varnish] Stripping tracking cookies

Tracking tool, like google analytics, set cookies that are only used by their client side javascript. Theses cookies are no interest for your server, and depending your configuration their can event prevent cache usage. So it’s better to simply strip them.

To do that, add into the vcl_recv this block:

# Remove Google Analytics, Piwik and other tracking cookies
if (req.http.Cookie) {
   set req.http.Cookie = regsuball(req.http.Cookie, "(^|;\s*)(__[a-z]+|has_js)=[^;]*", "");
   set req.http.Cookie = regsuball(req.http.Cookie, "(^|;\s*)(_pk_(ses|id)[\.a-z0-9]*)=[^;]*", "");
   set req.http.Cookie = regsuball(req.http.Cookie, "(^|; ) *__utm.=[^;]+;? *", "\1");
}
# Remove cookies when empty
if (req.http.Cookie == "") {
   remove req.http.Cookie;
}

Further Reading and sources