A brief description of the architecture of Gunicorn.
Gunicorn is based on the pre-fork worker model. This means that there is a central master process that manages a set of worker processes. The master never knows anything about individual clients. All requests and responses are handled completely by worker processes.
The master process is a simple loop that listens for various process signals and reacts accordingly. It manages the list of running workers by listening for signals like TTIN, TTOU, and CHLD. TTIN and TTOU tell the master to increase or decrease the number of running workers. CHLD indicates that a child process has terminated, in this case the master process automatically restarts the failed worker.
The most basic and the default worker type is a synchronous worker class that handles a single request at a time. This model is the simplest to reason about as any errors will affect at most a single request. Though as we describe below only processing a single request at a time requires some assumptions about how applications are programmed.
sync worker does not support persistent connections - each connection is
closed after response has been sent (even if you manually add
Connection: keep-alive header in your application).
The asynchronous workers available are based on Greenlets (via Eventlet and Gevent). Greenlets are an implementation of cooperative multi-threading for Python. In general, an application should be able to make use of these worker classes with no changes.
For full greenlet support applications might need to be adapted. When using, e.g., Gevent and Psycopg it makes sense to ensure psycogreen is installed and setup.
Other applications might not be compatible at all as they, e.g., rely on the original unpatched behavior.
The worker gthread is a threaded worker. It accepts connections in the main loop. Accepted connections are added to the thread pool as a connection job. On keepalive connections are put back in the loop waiting for an event. If no event happens after the keepalive timeout, the connection is closed.
There’s also a Tornado worker class. It can be used to write applications using the Tornado framework. Although the Tornado workers are capable of serving a WSGI application, this is not a recommended configuration.
Choosing a Worker Type¶
The default synchronous workers assume that your application is resource-bound in terms of CPU and network bandwidth. Generally this means that your application shouldn’t do anything that takes an undefined amount of time. An example of something that takes an undefined amount of time is a request to the internet. At some point the external network will fail in such a way that clients will pile up on your servers. So, in this sense, any web application which makes outgoing requests to APIs will benefit from an asynchronous worker.
This resource bound assumption is why we require a buffering proxy in front of a default configuration Gunicorn. If you exposed synchronous workers to the internet, a DOS attack would be trivial by creating a load that trickles data to the servers. For the curious, Hey is an example of this type of load.
Some examples of behavior requiring asynchronous workers:
- Applications making long blocking calls (Ie, external web services)
- Serving requests directly to the internet
- Streaming requests and responses
- Long polling
- Web sockets
How Many Workers?¶
DO NOT scale the number of workers to the number of clients you expect to have. Gunicorn should only need 4-12 worker processes to handle hundreds or thousands of requests per second.
Gunicorn relies on the operating system to provide all of the load balancing
when handling requests. Generally we recommend
(2 x $num_cores) + 1 as the
number of workers to start off with. While not overly scientific, the formula
is based on the assumption that for a given core, one worker will be reading
or writing from the socket while the other worker is processing a request.
Obviously, your particular hardware and application are going to affect the optimal number of workers. Our recommendation is to start with the above guess and tune using TTIN and TTOU signals while the application is under load.
Always remember, there is such a thing as too many workers. After a point your worker processes will start thrashing system resources decreasing the throughput of the entire system.
How Many Threads?¶
Since Gunicorn 19, a threads option can be used to process requests in multiple threads. Using threads assumes use of the gthread worker. One benefit from threads is that requests can take longer than the worker timeout while notifying the master process that it is not frozen and should not be killed. Depending on the system, using multiple threads, multiple worker processes, or some mixture, may yield the best results. For example, CPython may not perform as well as Jython when using threads, as threading is implemented differently by each. Using threads instead of processes is a good way to reduce the memory footprint of Gunicorn, while still allowing for application upgrades using the reload signal, as the application code will be shared among workers but loaded only in the worker processes (unlike when using the preload setting, which loads the code in the master process).
Under Python 2.x, you need to install the ‘futures’ package to use this feature.