Files
tryton/doc/topics/start_server.rst
T
2019-02-05 19:09:53 +01:00

70 lines
2.2 KiB
ReStructuredText

.. _topics-start-server:
=======================
How to start the server
=======================
Web service
===========
You can start the default web server bundled in Tryton with this command line::
trytond -c <config file>
The server will wait for client connections on the interface defined in the
`web` section of the :ref:`configuration <topics-configuration>`.
.. note:: When using multiple config files the order is importart as last
entered files will override the items of first files
WSGI server
-----------
If you prefer to run Tryton inside your own WSGI server instead of the simple
server of Werkzeug, you can use the application `trytond.application.app`.
Following environment variables can be set:
* `TRYTOND_CONFIG`: Point to :ref:`configuration <topics-configuration>` file.
* `TRYTOND_LOGGING_CONFIG`: Point to :ref:`logging <topics-logs>` file.
* `TRYTOND_COROUTINE`: Use coroutine for concurrency.
* `TRYTOND_DATABASE_NAMES`: A list of database names in CSV format, using
python default dialect.
.. warning:: You must manage to serve the static files from the web root.
Coroutine server
----------------
The Werkzeug server uses thread for concurrency. This is not optimal for the
long-polling request on the :ref:`bus <ref-bus>` as each client consumes
permanently one thread.
You can start the server with coroutine using the option `--coroutine`.
Cron service
============
If you want to run some scheduled actions, you must also run the cron server
with this command line::
trytond-cron -c <config file> -d <database>
The server will wake up every minutes and preform the scheduled actions defined
in the `database`.
Worker service
==============
If you want to use a pool of workers to run :ref:`asynchronously some tasks
<topics-task-queue>`, you must activate the worker in the `queue` section of
the :ref:`configuration <topics-configuration>` and run the worker manager with
this command line::
trytond-worker -c <config file> -d <database>
The manager will dispatch tasks from the queue to a pool of worker processes.
Services options
================
You will find more options for those services by using `--help` arguments.