Take this summary with a mountain of salt, the grades are based on how each implementation is performing relative to the others, test by test. It does not take into account that some implementations have deal-breaker performance.
For example, most reads are too slow with treebeard AL to make it a serious choice for anything other than a toy project.
In the same way, MPTT can be extremely slow at writing data, making it unusable on a wide range of projects.
Note as well that despite the extremely low grade of django-tree in terms of storage, it is absolutely not a deal breaker. By the way, the amount of indexes from django-tree is based on a parameter that can be customized field by field.
[An HTML table with grades should be displayed just above, but it might not always work on GitHub.]
- MPTT: stores a
parentforeign key andleft,right,tree_id&levelfields to represent the node position in the tree. Can be very slow at writing data, which can be blocking for a table of > 100 000 rows as you need to rebuild it from time to time. The only alternative to django-tree where tree structures are safely stored. - treebeard AL: Adjacency List. Very basic, stores a foreign key only. Very fast to write, extremely slow to read.
- treebeard MP: Materialized Path. Stores the position in the tree
using a string in the form
000100010002for the equivalent to[0, 0, 1]in django-tree. Very fast for reading, good at writing, but very brittle: works only using an annoying Python API and enforces no database constraint. - treebeard NS: Nested Sets. Equivalent to MPTT, but without a parent foreign key. As slow as MPTT and as brittle as treebeard MP.
.svg)
Note that deleting with django-tree, treebeard AL and MPTT is slower than MP and NS due to Django itself.
Since these 3 implementations rely on a ForeignKey usually with on_delete=CASCADE,
the Collector from Django tries to find related data through these foreign keys,
even though we already send all descendants of the deleted node to Django.
The only way to speed this up would be to specify DO_NOTHING, but that
would be misleading since the descendants are still getting deleted.
