This page describes the architecture of MyTardis.
MyTardis is built on the Django web framework, which itself is built on Python, thus MyTardis follows the architectural model of Django.
- Web Server
- MyTardis is typically deployed with the standard Nginx + Gunicorn + Django + Python stack.
- Ideally, use PostgreSQL. MySQL or MariaDB may be usable but it is not as well tested.
- Data Storage
- Local disk, network mounted disks as well as S3 and SFTP are supported.
- Ingestion is normally site specific. A desktop application is available (MyData).
MyTardis follows the standard Django architecture.
The major functional modules in MyTardis are:
- A RESTful API provides anonymous and authenticated access to most of the stored data.
- Authentication and Authorisation.
- Data download functionality.
- Processing of metadata as data is uploaded, e.g. extract EXIF metadata from JPEG files.
- Additional commands for the Django CLI. The backup utility is implemented as a Django command.
- Database migration code.
- RIF-CS metadata publication.
- Provides a searchable and authorisation enabled index for quickly finding data.
- SFTP server
- Provides read access to data in MyTardis via the SFTP protocol
Information on the individual modules is available from the Module Index.
At the component / module level, performance and Scalability within MyTardis is achieved by:
- Allowing long lived or compute intensive operations to be hosted on separate
machines from the main web server.
- E.g. by running multiple web servers and sending long lived operations, such as ingestion, to a server reserved for such operations.
- Performance and Scalability of the database is achieved by a combination of 1) optimising the SQL requests issued by MyTardis, and 2) database specific scalability, please refer to the appropriate database documentation.
- Performance and Scalability of the web server is provided through the normal
mechanisms for nginx / wsgi hosted python applications, e,g,:
- Increasing individual Server capability
- Individual Server performance / utilization may be managed by controlling the number of python (django) processes active at any one time.
- Deploying multiple web servers
- Multiple web servers may be deployed using standard mechanisms, e.g. load balancers. State (session) information is distributed using Django’s standard session model.
- Increasing individual Server capability
- The Data Archive is a normal file system, e.g. NFS, SAMBA, etc., with performance and scalability dependent on the implementation and deployment.
- Extraction and formatting of metadata for ingestion is up to the client and may be distributed across any number of machines.
SQL Scalability Notes¶
The datafileparameter table is expected to be the single largest table by an order of magnitude, in the hundreds of millions of records (at the Australian Synchrotron).
For Postgres, this will probably be addressed by table partioning. In this case using a date field to partition is often recommended. Dates may be introduced as part of the support for anotations, raw data, versioned data and derived data.
Persistance and Data Architecture¶
Django provides an Object-Relational-Model that maps Django Models to the underlying relational database.
The Django Models are defined in
Access to data catalogued by MyTardis may be either public, i.e. anonymous access allowed, or private, requiring authentication. The Authentication Framework is documented in ref-authframework.
Unless the application has a very rich business model (which MyTardis does not), a class diagram doesn’t convey much information in Django applications (it ends up being a flat list of classes). To understand the context in which the MyTardis classes are created, please refer to the Django documentation, http://docs.djangoproject.com/ on the responsibilities of models, views, managers and templates.
This section describes the top level layout of the source code in the GitHub repository.
- MyTardis User and Administrative documentation
- Contains all the optional functionality and installation specific
functionality. Examples include the optional ANDS Register (ands_register) and Related Info (related_info) tabs, and the ANSTO (mecat-ansto) and Australian Synchrotron (mecat-as) installation specific modules.
- Some applications are part of the main code base, others can be added
at installation time.
- Search related code
- This is the main django application that contains most of the core MyTardis functionality