The data format: interconnected tables

The ALICE O2 data model is a collection of tables. Each table can be seen as a collection of objects, where each column represents a property of the objects, and each row is an object. When the users write their analysis, they need to provide at least two things: a query which selects rows of one or more tables and a function which gets invoked on the results of the query. When specifying the query, the users can perform typical database operations between them. For example, they can only select rows matching a given query, or they can put two tables one on the side of the other and select the rows of the resulting table in what it's called a "join".

Tables can be read from the input files AO2D files and or they can be created by tasks. To produce some tables with commonly used quantities there exist a set of predefined helper tasks which can be included in user analysis work flows to produce tables with commonly used information

The data model also provides a bunch of pre-defined joins and iterators.

Table relations

Information contained in different tables can be related. E.g. a track belongs to a given collision, or signals in the FIT or Zdc detectors belong to a bunch crossing.

Hence the dependent tables need to hold an index which points to a specific row of the master table. For this the dependent table (e.g. table Tracks) has an index column [master]Id (in this case CollisionsId) which points to the related information in table master. See also e.g. master=BCs and dependent=CaloTriggers and many more.


For each index in a table, there is a method to get the corresponding object. For example in the table aod::Tracks there is an index collisionId. So if you have an object aod::Track called track, you can directly access the collision of this track by calling track.collision(). You can also check before doing so that the track you are interested in has a collision, by using the templated function .has_smthg(), in this example using track.has_collision(), which returns a boolean . The type of the object track.collision() is o2::aod::Collision. This is true for every index in every table, so calling for instance collision.bc() gives you the bc of the collision you are working on. You can then call for example collision.bc().globalBC() .


Be aware that tables can be joined and be extended with extra colums.

In order to process the data, analysis task objects have to be written and have to subscribe to tables - i.e., they have to tell the framework that a certain table is of interest and will be processed by the task. In what follows, we will explain the basics regarding analysis tasks and the subscription to data. If you are interested in a reference guide that covers all the basic content of the tables that are currently in O2Physics, please refer to the automatically generated Appendix: Data model reference.