configure_settings.md

Configuration of the settings INI file

The settings INI file is the primary project property access point for every AIDE module. It contains parameters, addresses and some passwords and must therefore never be exposed to the public!

The settings file is divided into the following categories:

[Project]

In the latest version of AIDE, this section only contains the credentials for the so-called "super user" (who has full permission in every project).

Name	Values	Default value	Required	Comments
adminName	(string)		YES	Name of the AIDE administrator (super user) account.
adminEmail	(e-mail string)		YES	E-mail address of the AIDE administrator account.
adminPassword	(string)		YES	Plain text password of the AIDE administrator account.

[Server]

This section contains parameters for all the individual instances' addresses.

Name	Values	Default value	Required	Comments
host	(IP address or hostname)	0.0.0.0	YES	This is the host IP address of the current instance. As such, it might need to be set differently for every machine taking part in AIDE. Be sure to change the individual addresses below to make the machines reachable to each other, whenever necessary.
port	(numeric)	80	YES	Network port of the current instance. Again, you might want to specify custom values depending on the machine here. For example, the frontend (LabelUI) might run on HTTP's standard port 80, but you can e.g. route the AIWorker instances through different ports here. Be sure to change the individual addresses below to make the machines reachable to each other, whenever necessary.
numWorkers	(numeric)	6	NO	Number of Gunicorn server threads to launch. More threads can serve more requests in parallel, but might also cause a computational overhead and use up more database connections.
index_uri	(URI)	/		URL snippet under which the index page can be found. By default this can be left as "/", but may be changed if AIDE is e.g. deployed under a sub-URL, such as "http://www.mydomain.com/aide", in which case it would have to be changed to "/aide".
dataServer_uri	(URI)		YES	URI, resp. URL of the FileServer instance. Note that the instance needs to be accessible to both the users accessing the LabelUI webpage, as well as to any running AIWorker instance. In URL format this may include the port number and the FileServer's "staticfiles_uri" parameter too (see below); for example: http://fileserver.domain.com:67742/files.
aiController_uri	(URI)			The same for the AIController instance. This must primarily be accessible to running AIWorker instances, but the value of it is also used in the frontend to determine whether AI support is enabled or not. In URL format this may include the port number of the AIController too; for example: http://aicontroller.domain.com:67743.

[UserHandler]

Name	Values	Default value	Required	Comments
time_login	(numeric > 0)	600	YES	Time (in seconds) for a session to last if the user is inactive. Upon exceeding the threshold specified here, the user is either asked to re-type their password, or else redirected to the index page.
create_account_token	(string)			A custom string of (preferably) random characters required to be known to users who would like to create a new account on the project page. This is to make the project semi-secret. If this value is set, the webpage to create a new account can be accessed as follows: http://<hostname>/createAccount?t=<create_account_token>, substituting the expressions in angular brackets accordingly. If left out, a new account can be created by simply visiting: http://<hostname>/createAccount.

[AIController]

Name	Values	Default value	Required	Comments
broker_URL	(URL)	amqp://localhost	YES	URL under which the message broker (RabbitMQ, Redis, etc.) can be reached. This might include an access username, password, port and trailing specifier (e.g. queue). Refer to the individual frameworks for details.
result_backend	(URL)	redis://localhost:6379/0	YES	Backend URL under which status updates and results are fetched. Important: it is required to use a persistent backend for the message store (do not use rpc). The recommended backend is Redis. See details here.
maxNumWorkers_train	(numeric)	-1		Maximum number of AIWorker instances to consider when training. -1 means that all available AIWorkers will be involved in training, and that the images will be distributed evenly across them. If > 1 or = -1, the training images will be distributed evenly over the number of AIWorkers specified, and the model's 'average_model_states' function will be called once all workers have finished training to generate a new, holistic model state. Note that this might not always be preferred (some models might not allow to be averaged). In this case, set this number to 1 to limit training (on all training images) to just one AIWorker.
maxNumWorkers_inference	(numeric)	-1		Maximum number of AIWorker instances to involve when doing inference on images. -1 means that all available AIWorkers will be involved, and that the images will be distributed evenly across them.
max_num_concurrent_tasks	(numeric)	2		Maximum number of tasks that can be launched at a time per project. Individual projects may provide their own override, but the value provided here serves as an absolute upper ceiling. Auto-launched tasks are always limited to one at a time; the number of tasks here takes these into account, even for user-launched tasks (i.e., if the number set here is 2 and there is already a task running, whether auto-launched or not, at most one additional task can be run at a time). Set to <= 0 to allow an unlimited number of tasks to be executed at a time per project.

[AIWorker]

Name	Values	Default value	Required	Comments
inference_batch_size_limit	(numeric)	-1	YES	Number of images to perform inference on at a time. If this value is smaller than the designated number of images for inference in a job, the total number of images will be split into chunks of this size and processed in order, on each AIWorker. This is especially important for data-intensive annotation types, such as segmentation masks, where all the annotations are loaded into system memory prior to calling the inference job. By limiting the number of images to be processed at once, pressure on system RAM can be relieved. Set to a reasonable value if you encounter out of memory issues on AIWorkers. For annotation types that generate less data (labels, points, bounding boxes), this parameter can generally be ignored (resp. set to the default of -1, i.e. "unlimited").

[FileServer]

Name	Values	Default value	Required	Comments
staticfiles_dir	(path)		YES	Root directory on the local disk of the file server to serve files from.
staticfiles_uri_addendum	(URI string)		NO	Optional snippet to append after the file server's host name. For example, if set to aide, the file server provides files through http(s)://host:port/aide.
tempfiles_dir	(path)	OS temp dir	NO	Directory where files like data download request results are stored. Defaults to the OS' temporary files directory (i.e., /tmp on Unix or Linux, ~/APPDATA/Local/Temp on Windows, or others).
watch_folder_interval	(float)	60	NO	Interval (in seconds) for periodic project folder watch functionality. If project are configured to automatically watch their image folder for changes, those tasks will be carried out on the file server in a combined way every number of seconds specified here. Set to 0 (zero) or a negative value to globally disable folder watching for all projects. Default is 60 (one minute).

[Database]

Name	Values	Default value	Required	Comments
name	(string)		YES	Name of the Postgres database on the server.
schema	(string)		YES	Schema within the database to store the project data in.
host	(URL)		YES	URL under which the database can be accessed (without the port). Can be set to localhost if and only if all AIDE modules are to be launched on the same server the database is hosted on.
port	(numeric)		YES	Port the database listens to. Note: Postgres' default port is 5432; unless the database instance is solely connected to LAN (and not WAN), it is advised to change the Postgres port to another, free value. The database installation instructions will automatically consider the custom port.
user	(string)		YES	Name of the user that is given access to the database.
password	(string)		YES	Password (in clear text) for the Postgres user. NOTE: unlike all other database fields, the password is case-sensitive.
credentials	(string)			Alternative form of providing database username and password through a text file, similar to credentials e.g. in Linux's fstab entries. Text file may contain lines like username=user and password=pass, without whitespaces. Parameters user and password that are directly provided in the .ini file have precedence over the credentials file.
max_num_connections	(numeric)	16		Maximum number of connections to the database per server running an AIDE module. This number, multiplied by the number of server instances running AIDE, must not exceed the maximum number of connections defined in Postgres' configuration file. A minimum number of 2 is required and will always be enforced at runtime.