Caching is a common techneque to achieve performance goals, when a web application has to perform some operation that could take a long time. There are two major types of caching used in Web Applications:
- Whole-page caching – works at the HTTP protocol level to avoid entire requests to the server by having either the user’s browser, or an intermediate proxy server (such as Squid) intercept the request and return a cached copy of the file.
- Application-level caching – works within the application server to cache computed values, often the results of complex database queries, so that future requests can avoid needing to re-caculate the values.
Most web applications can only make very selective use of HTTP-level caching, such as for caching generated RSS feeds, but that use of HTTP-level caching can dramatically reduce load on your server, particularly when using an external proxy such as Squid and encountering a high-traffic event (such as the Slashdot Effect).
For web applications, application-level caching provides a flexible way to cache the results of complex queries so that the total load of a given controller method can be reduced to a few user-specific or case-specific queries and the rendering overhead of a template. Even within templates, application-level caching can be used to cache rendered HTML for those fragments of the interface which are comparatively static, such as database-configured menus, reducing potentially recursive database queries to simple memory-based cache lookups.
TurboGears comes with application-level caching middleware enabled by default in QuickStarted projects. The middleware, Beaker is the same package which provides Session storage for QuickStarted projects. Beaker is the standard cache framework of the Pylons web framework, on which TurboGears 2.1 is based.
Beaker supports a variety of backends which can be used for cache or session storage:
Each of these backends can be configured from your application’s configuration file, and the resulting caches can be used with the same API within your application.
The configured Beaker cache is provided by the pylons module. This is more properly thought of as a CacheManager, as it provides access to multiple independent cache namespaces. To access the cache from within a controller module:
from pylons import cache @expose() def some_action(self, day): # hypothetical action that uses a 'day' variable as its key def expensive_function(): # do something that takes a lot of cpu/resources return expensive_call() # Get a cache for a specific namespace, you can name it whatever # you want, in this case its 'my_function' mycache = cache.get_cache('my_function') # Get the value, this will create the cache copy the first time # and any time it expires (in seconds, so 3600 = one hour) cachedvalue = mycache.get_value( key=day, createfunc=expensive_function, expiretime=3600 ) return dict(myvalue=cachedvalue)
The Beaker cache is a two-level namespace, with the keys at each level being string values. The call to cache.get_cache() retrieves a cache namespace which will map a set of string keys to stored values. Each value that is stored in the cache must be pickle-able.
Pay attention to the keys you are using to store your cached values. You need to be sure that your keys encode all of the information that the results being cached depend upon in a unique manner. In the example above, we use day as the key for our cached value, on the assumption that this is the only value which affects the calculation of expensive_function, if there were multiple parameters involved, we would need to encode each of them into the key.
The Beaker API exposed here requires that your functions for calculating complex values be callables taking 0 arguments. Often you will use a nested function to provide this interface as simply as possible. This function will only be called if there is a cache miss, that is, if the cache does not currently have the given key recorded (or the recorded key has expired).
In templates, the cache namespace will automatically be set to the name of the template being rendered. Nothing else is required for basic caching, unless the developer wishes to control for how long the template is cached and/or maintain caches of multiple versions of the template.
The cache also supports the removal values from the cache, using the key(s) to identify the value(s) to be removed and it also supports clearing the cache completely, should it need to be reset.
# Clear the cache mycache.clear() # Remove a specific key mycache.remove_value('some_key')
Beaker is configured in your QuickStarted application’s main configuration file in the app:main section.
To use memory-based caching:
[app:main] beaker.cache.type = memory
To use file-based caching:
[app:main] beaker.cache.type = file beaker.cache.data_dir = /tmp/cache/beaker beaker.cache.lock_dir = /tmp/lock/beaker
To use DBM-file-based caching:
[app:main] beaker.cache.type = dbm beaker.cache.data_dir = /tmp/cache/beaker beaker.cache.lock_dir = /tmp/lock/beaker
To use SQLAlchemy-based caching you must provide the url parameter for the Beaker configuration. This can be any valid SQLAlchemy URL, the Beaker storage table will be created by Beaker if necessary:
[app:main] beaker.cache.type = ext:database beaker.cache.url = sqlite:///tmp/cache/beaker.sqlite
Memcached allows for creating a pool of colaborating servers which manage a single distributed cache which can be shared by large numbers of front-end servers (i.e. TurboGears instances). Memcached can be extremely fast and scales up very well, but it involves an external daemon process which (normally) must be maintained (and secured) by your sysadmin.
Memcached is a system-level daemon which is intended for use solely on “trusted” networks, there is little or no security provided by the daemon (it trusts anyone who can connect to it), so you should never run the daemon on a network which can be accessed by the public! To repeat, do not run memcached without a firewall or other network partitioning mechanism! Further, be careful about storing any sensitive or authentication/authorization data in memcache, as any attacker who can gain access to the network can access this information.
Ubuntu/Debian servers will generally have memcached configured by default to only run on the localhost interface, and will have a small amount of memory (say 64MB) configured. The /etc/memcached.conf file can be edited to change those parameters. The memcached daemon will also normally be deactivated by default on installation. A basic memcached installation might look like this on an Ubuntu host:
sudo aptitude install memcached sudo vim /etc/default/memcached # ENABLE_MEMCACHED=yes sudo vim /etc/memcached.conf # Set your desired parameters... sudo /etc/init.d/memcached restart # now install the Python-side client library... # note that there are other implementations as well... easy_install python-memcached
You then need to configure TurboGears/Pylon’s beaker support to use the memcached daemon in your .ini files:
[app:main] beaker.cache.type = ext:memcached beaker.cache.url = 127.0.0.1:11211 # you can also store sessions in memcached, should you wish # beaker.session.type = ext:memcached # beaker.session.url = 127.0.0.1:11211
You can have multiple memcached servers specified using ; separators. Usage, as you might imagine is the same as with any other Beaker cache configuration (that is, to some extent, the point of the Beaker Cache abstraction, after all):
- Beaker Caching – discussion of use of Beaker’s caching services
- Beaker Configuration – the various parameters which can be used to configure Beaker in your config files
- Memcached – the memcached project
- Python Memcached – Python client-side binding for memcached
- Caching for Performance – Stephen Pierzchala’s general introduction to the concept of caching in order to improve web-site performance
HTTP supports caching of whole responses (web-pages, images, script-files and the like). This kind of caching can dramatically speed up web-sites where the bulk of the content being served is largely static, or changes predictably, or where some commonly viewed page (such as a home-page) requires complex operations to generate.
HTTP-level caching is handled by external services, such as a Squid proxy or the user’s browser cache. The web application’s role in HTTP-level caching is simply to signal to the external service what level of caching is appropriate for a given piece of content.
If any part of you page has to be dynamically generated, even the simplest fragment, such as a user-name, for each request HTTP caching likely will not work for you. Once the page is HTTP-cached, the application server will not recieve any further requests until the cache expires, so it will not generally be able to do even minor customizations.
HTTP/1.1 supports the ETag caching system that allows the browser to use its own cache instead of requiring regeneration of the entire page. ETag-based caching avoids repeated generation of content but if the browser has never seen the page before, the page will still be generated. Therefore using ETag caching in conjunction with one of the other types of caching listed here will achieve optimal throughput and avoid unnecessary calls on resource-intensive operations.
Caching via ETag involves sending the browser an ETag header so that it knows to save and possibly use a cached copy of the page from its own cache, instead of requesting the application to send a fresh copy.
The etag_cache() function will set the proper HTTP headers if the browser doesn’t yet have a copy of the page. Otherwise, a 304 HTTP Exception will be thrown that is then caught by Paste middleware and turned into a proper 304 response to the browser. This will cause the browser to use its own locally-cached copy.
etag_cache() returns pylons.response for legacy purposes (pylons.response should be used directly instead).
ETag-based caching requires a single key which is sent in the ETag HTTP header back to the browser. The RFC specification for HTTP headers indicates that an ETag header merely needs to be a string. This value of this string does not need to be unique for every URL as the browser itself determines whether to use its own copy, this decision is based on the URL and the ETag key.
from pylons.controllers.util import etag_cache def my_action(self): etag_cache('somekey') return render('/show.myt', cache_expire=3600)
Or to change other aspects of the response:
from pylons.controllers.util import etag_cache from tg import response def my_action(self): etag_cache('somekey') response.headers['content-type'] = 'text/plain' return render('/show.myt', cache_expire=3600)
In this example that we are using template caching in addition to ETag caching. If a new visitor comes to the site, we avoid re-rendering the template if a cached copy exists and repeat hits to the page by that user will then trigger the ETag cache. This example also will never change the ETag key, so the browsers cache will always be used if it has one.
The frequency with which an ETag cache key is changed will depend on the web application and the developer’s assessment of how often the browser should be prompted to fetch a fresh copy of the page.
Add links to Beaker region (task-specific caching mechanisms) support.
Document what the default Beaker cache setup is for TG 2.1 quickstarted projects (file-based, likely).
Provide code-sample for use of cache within templates