Architecture

Processes

gws server start configures the application, writes the configuration of the embedded servers to /gws-var/server and starts them:

nginx and the web application communicate over a unix socket. The number of workers and timeouts are set in the server section of the configuration, see Config.

Configuration lifecycle

Configuring runs once, in the process that starts the server:

1. The specs are generated from the source code.
2. The configuration files are read and validated against the specs.
3. The object tree is built: each node runs pre_configure and configure, and creates its children.
4. When the tree is complete, each node runs post_configure.
5. The tree is pickled to /gws-var/config/config.pickle.

Each web and spool worker then loads the pickled tree and calls activate on every node. The code for these steps is in loader.

Since activate runs in every worker, it is the place for per-process resources, like database connections or file watchers. configure and post_configure run only once and must not rely on such resources.

Request flow

1. nginx applies rewrite rules and serves static files. Everything else goes to the web application.
2. The web application parses the request into a WebRequester. Paths starting with /_ are commands, other paths fail with 404.
3. Middleware runs: the database manager, then the authentication manager, which reads the session and sets the user.
4. The action manager finds the command, validates the request parameters against the specs and checks permissions.
5. The command method runs and returns a response.
6. The response is encoded as JSON, msgpack or sent as raw content.

See Actions and requests for details.

Background jobs

Long-running tasks, like printing or exporting, run as jobs. A job is stored in a sqlite database in /gws-var and queued to the uWSGI spooler, which runs it in a spool worker. The client polls the job status with a separate command. The job API is in job.

Monitor

The monitor runs in the spool process. It watches configuration files and reconfigures the server when they change. It also runs periodic tasks registered by nodes. Set server.withMonitor false to disable it.