Indexes

Sheraf provides a way to quickly access your data matching some criterias. This section shows how to take advantage of sheraf indexes.

Faster filtering and ordering

With large collections, searching for models with filter() can be very long. Indeed, the default behavior of filter() is to scan every available model, and tests its values.

>>> # The following statements have equal complexity
>>> MyModel.filter(my_attribute="something") 
>>> (model for model in MyModel.all() if model.my_attribute == "something") 

Of course all the models are not loaded in memory at once due to the lazy behavior of QuerySet, but the iteration still need to make numerous accesses to the database, and test every model instance, thus severely degrading the performances.

A solution to keep good performances is to use attribute indexation with Index. Creating an Index object in a IndexedModel creates a new index table in the database. This table matches the model instances with their attribute values.

>>> class Cowboy(sheraf.Model):
...     table = "simple_cowboy"
...     name = sheraf.SimpleAttribute()
...     nameindex = sheraf.Index("name")
...
>>> with sheraf.connection(commit=True) as conn:
...     george = Cowboy.create(name="George Abitbol")
...
...     # A dedicated table has been created for the 'nameindex' index
...     assert george.mapping in conn.root()["simple_cowboy"]["nameindex"]["George Abitbol"].values()
...
...     # Filtering over names is a lot faster!
...     assert [george] == Cowboy.filter(nameindex="George Abitbol")

A shortcut for this is to call the index() method on attributes. It takes the very same arguments as the Index object, but the index will have the same name as the attribute. Any piece of code that worked with a non-indexed attribute will have the very same behavior when the attribute is indexed, but faster. So migrations are painless.

>>> class Cowboy(sheraf.Model):
...     table = "simple_cowboy_shortcut"
...     name = sheraf.SimpleAttribute().index()
...
>>> with sheraf.connection(commit=True) as conn:
...     george = Cowboy.create(name="George Abitbol")
...     assert [george] == Cowboy.filter(name="George Abitbol")

Attribute indexation also hugely improves the order() performances.

>>> import uuid
>>> with sheraf.connection(): 
...     # Even with a lot of cowboy...
...     for i in range(10000):
...         Cowboy.create(name=str(uuid.uuid4()))
...
...     # ... ordering on names is very fast
...     assert george in Cowboy.order(name=sheraf.DESC)

Values unicity or multiplicity

>>> class Cowboy(sheraf.Model):
...     table = "unique_cowboy"
...     name = sheraf.SimpleAttribute()
...     email = sheraf.SimpleAttribute().index(unique=True)

By default, indexed attributes of different model instances can have the same value. There is no issue for two cowboys to be called George Abitbol. But in our example, the email attribute has a unique flag. This means that only one person can have a specific email at a time. Trying to create a second person with george@abitbol.com will result in a UniqueIndexException exception.

>>> with sheraf.connection(commit=True):
...     george = Cowboy.create(name="The true George", email="george@abitbol.com")
...
>>> with sheraf.connection():
...     Cowboy.create(name="The fake George", email="george@abitbol.com")
Traceback (most recent call last):
    ...
sheraf.exceptions.UniqueIndexException

Note that when an attribute is unique, you can use the read() method.

>>> with sheraf.connection():
...     assert george == Cowboy.read(email="george@abitbol.com")

Multiple indexes

What if we want to index birth years and birth months? This is quite straightforward, index() calls can be chained to describe different indexes, and the key parameter can be used to identify them.

>>> class Cowboy(sheraf.Model):
...     table = "multiple_cowboy"
...     birth = sheraf.DateTimeAttribute() \
...         .index(key="year", index_keys_func=lambda birth: birth.year) \
...         .index(key="month", index_keys_func=lambda birth: birth.month)
...
>>> from datetime import datetime
>>> with sheraf.connection():
...     peter = Cowboy.create(birth=datetime(1989, 4, 13))
...     assert [peter] == Cowboy.filter(year=1989)
...     assert [peter] == Cowboy.filter(month=4)
...     assert [peter] == Cowboy.search(year=datetime(1989, 4, 13))
...     assert [peter] == Cowboy.search(month=datetime(1989, 4, 13))

Custom values in the index

Sometimes you may want to transform a value before indexation, or before querying the database.

Choose how to record data in the index

For instance, what if we would like to index cowboy not its name, but on its initials?

index() takes a index_keys_func argument that is a function taking the attribute value, and returning a collection of keys on which the model instance should be indexed.

>>> def initials(name):
...     return "".join(word[0] for word in name.split(" "))
...
>>> class Cowboy(sheraf.Model):
...     table = "valuable_cowboy"
...     name = sheraf.StringAttribute().index(index_keys_func=initials)
...
>>> with sheraf.connection(commit=True):
...     george = Cowboy.create(name="George Abitbol")

Here we pass the a function that returns the initials of a. Now it is possible to search for someone only knowing its initials.

>>> with sheraf.connection():
...     assert [george] == Cowboy.filter(name="GA")
...     assert [] == Cowboy.filter(name="George Abitbol")

Note that the filter() name parameter will not be transformed into initials. It search for the exact data in the index.

Note

index_keys_func functions can return either a single element or a collection of elements. Depending on the noneok and nullok Index parameters, None and falsy index keys might be ignored.

Choose how to search data in the index

Now what if you need to search for the initials of a cowboy based on another cowboy’s name? You could just use the search() method to do that.

>>> with sheraf.connection():
...     assert [george] == Cowboy.search(name="Gerard Amsterdam")
...     assert [george] == Cowboy.search(name="Geoffrey Abitbol")

You may want to be able to edit the values you pass to name. For instance, you may want your users to be able to search for initials in whatever order they have been passed.

index() takes a search_keys_func argument that is a function taking the data you want to search, and return a collection of keys to search in the index. search() will search for all the keys in the index, and will return the matching model instances. By default the search_keys_func argument takes the same argument than the index() index_keys_func argument.

>>> from itertools import permutations
>>> class Cowboy(sheraf.Model):
...     table = "invaluable_cowboy"
...     name = sheraf.StringAttribute().index(
...         index_keys_func=initials,
...         search_keys_func=lambda name: {
...             "".join(p) for p in permutations(initials(name))
...         },
...     )
...
>>> with sheraf.connection(commit=True):
...     george = Cowboy.create(name="George Abitbol")
...
...     assert [george] == Cowboy.search(name="Amsterdam Gerard")

Now we index the initials of cowboys, but we search for all the combinations of initials with the words that are passed to the search argument.

Note

search_keys_func functions can return either a single element or a collection of elements. If the collection is ordered as in a list, then the index will be searched in the order of the list. If the list contains a same element several times, it will only be returned once.

Make custom searchs and recording the default behavior

This name attribute and its indexation seems very convenient, so you would like to use it in other models. Luckily sheraf offers you a way to do this, and cut the boilerplate. If a Attribute defines some methods called index_keys or search_keys, they will be used by default if the index() index_keys_func and search_keys_func are not provided:

>>> class NameAttribute(sheraf.StringAttribute):
...     def index_keys(self, name):
...         return initials(name)
...
...     def search_keys(self, name):
...         return {"".join(p) for p in permutations(initials(name))}
...
>>> class Cowboy(sheraf.Model):
...     table = "clean_cowboy"
...     name = NameAttribute().index()
...
>>> with sheraf.connection(commit=True):
...     george = Cowboy.create(name="George Abitbol")
...
...     assert [george] == Cowboy.search(name="Amsterdam Gerard")

NameAttribute can now be used in other models (and it does not need to be indexed, it just can be).

Some attributes like ModelAttribute or collections like ListAttribute take benefit of this. They allow complex types like models or collections to be indexed. Generally models are indexed on their identifier, and every component of a collection is indexed.

>>> class Horse(sheraf.Model):
...     table = "horse"
...     name = sheraf.StringAttribute()
...
>>> class Cowboy(sheraf.Model):
...     table = "horsed_cowboy"
...     horses = sheraf.LargeListAttribute(
...         sheraf.ModelAttribute(Horse)
...     ).index()
...
>>> with sheraf.connection(commit=True):
...     jolly = Horse.create(name="Jolly Jumper")
...     george = Cowboy.create(horses=[jolly])
...
...     assert [george] == Cowboy.search(horses=jolly)

Dig a bit deeper

We could easilly use this to create a simple full-text search engine on a model attribute with only a few lines:

>>> from itertools import combinations
>>> def substrings(string):
...     return {
...         word[x:y]
...         for word in string.split(" ")
...         for x, y in combinations(range(len(word)+1), r=2)
...     }
...
>>> class Cowboy(sheraf.Model):
...     table = "deeper_cowboy"
...     biography = sheraf.SimpleAttribute().index(index_keys_func=substrings)
...
>>> with sheraf.connection():
...     george = Cowboy.create(
...         biography="He is 50, he is a cowboy and he is the most classy man on the world."
...     )
...     assert [george] == Cowboy.filter(biography="boy")

The substrings function extracts all the possible substring from all the words in a string. Now you can find a cowboy by searching for any piece of word in his biography.

To see how indexes can be used to build a full-text search engine, you can check the Advanced string indexes section.

Indexes over multiple attributes

It is possible for an index to watch several attributes. To do this you cannot use the index() shortcut, so you need to define the index with a Index object.

Here both first_name and last_name are indexed in the same place:

>>> class Cowboy(sheraf.Model):
...     table = "common_cowboys"
...     first_name = sheraf.StringAttribute()
...     last_name = sheraf.StringAttribute()
...
...     name = sheraf.Index(first_name, last_name)
...
>>> with sheraf.connection():
...     george = Cowboy.create(first_name="George", last_name="Abitbol")
...     assert george in Cowboy.search(name="George")
...     assert george in Cowboy.search(name="Abitbol")

When an index has several attributes, it can have a different indexation methods for each attribute, and a default one:

>>> class Cowboy(sheraf.Model):
...     table = "advanced_common_cowboys"
...     first_name = sheraf.StringAttribute()
...     last_name = sheraf.StringAttribute()
...     surname = sheraf.StringAttribute()
...
...     name = sheraf.Index(first_name, last_name, surname)
...
...     @name.index_keys_func
...     def default_name_indexation(self, value):
...         return value.lower()
...
...     @name.index_keys_func(first_name, last_name)
...     def full_name_indexation(self, first_name, last_name):
...         return f"{first_name} {last_name}".lower()
...
>>> with sheraf.connection():
...     george = Cowboy.create(first_name="George", last_name="Abitbol", surname="Georgy")
...     assert george in Cowboy.search(name="George Abitbol")
...     assert george in Cowboy.search(name="Georgy")
...     assert george not in Cowboy.search(name="Abitbol")

Here we used the index_keys_func() decorator to define a default_name_indexation method. As we did not pass any argument to the decorator, this method is the default indexation method for the index name. We also defined a full_name. By passing the first_name and last_name attributes to the index_keys_func() decorator, we assigned this method to both the attributes, and thus those very attributes can be indexed at the same time using this method.

Using indexation methods common to several attributes is very useful if you need conditionnal indexation.

>>> class Cowboy(sheraf.Model):
...     table = "little_big_cowboys"
...     name = sheraf.StringAttribute()
...     sherif = sheraf.BooleanAttribute()
...
...     sherif_names = sheraf.Index(name, sherif)
...
...     @sherif_names.index_keys_func(name, sherif)
...     def sherif_names_indexation(self, name, sherif):
...         return {name} if sherif else {}
...
>>> with sheraf.connection():
...     george = Cowboy.create(name="George", sherif=True)
...     peter = Cowboy.create(name="Peter", sherif=False)
...     assert george in Cowboy.search(sherif_names="George")
...     assert peter not in Cowboy.search(sherif_names="George")

The sherif_names index is updated each time a cowboy name or sherif attribute is edited, and it only contains the names of the sherifes.

Index inheritance

Index are inherited the most transparently as you can expect. You can overwrite a parent index, or even create an index on a parent attribute:

>>> class Cowboy(sheraf.Model):
...     table = "legacy_cowboys"
...     first_name = sheraf.StringAttribute()
...     last_name = sheraf.StringAttribute()
...
...     last_name_index = sheraf.Index(last_name, index_keys_func=lambda x: x.lower())
...
>>> class UpperCowboy(Cowboy):
...     table = "upper_cowboys"
...     last_name_index = sheraf.Index("last_name", index_keys_func=lambda x: x.upper())
...     first_name_index = sheraf.Index("first_name", index_keys_func=lambda x: x.upper())
...
>>> with sheraf.connection():
...     george = UpperCowboy.create(first_name="george", last_name="abitbol")
...     assert george in UpperCowboy.filter(first_name_index="GEORGE")
...     assert george in UpperCowboy.filter(last_name_index="ABITBOL")

In the Cowboy model the last_name_index stores the names in lowercase, but in the inherited UpperCowboy model the index has been overwritten so names are stored in the index in uppercase. UpperCowboy also defines a first_name_index on the first_name attribute, that is defined in its parent model class.

Health checks and fixes

Now you are convinced that indexes are awesome and you want to add some in your models. You can totally just add a .index() on your attributes, and everything will go fine…

…except that things may not be faster. This is because indexation is disabled for already populated model tables. If your database is empty, indexation will work out of the box, but if you already have some instances you will get a IndexationWarning when you will create or edit model instances.

>>> class Cowboy(sheraf.Model):
...     table = "future_cowboys"
...     name = sheraf.StringAttribute()
...
>>> with sheraf.connection(commit=True):
...     george = Cowboy.create(name="George")
...     peter = Cowboy.create(name="Peter")
...
>>> # Now you decide to add indexes in your code
>>> class Cowboy(sheraf.Model):
...     table = "future_cowboys"
...     name = sheraf.StringAttribute().index()
...
>>> import warnings
>>> with sheraf.connection(commit=True):
...     with warnings.catch_warnings(record=True) as warns:
...         steven = Cowboy.create(name="Steven")
...         assert warns[0].category is sheraf.exceptions.IndexationWarning

Sheraf provides tools to check the health of your model tables. So now, let us check how things are going for cowboys:

>>> from sheraf import print_health
>>> with sheraf.connection(): 
...     print_health(Cowboy, attribute_checks=["index"])
             _                     __        _               _
=========== | | ================= / _| ==== | | =========== | | ===============
         ___| |__   ___ _ __ __ _| |_    ___| |__   ___  ___| | _____
        / __| '_ \ / _ \ '__/ _` |  _|  / __| '_ \ / _ \/ __| |/ / __|
        \__ \ | | |  __/ | | (_| | |   | (__| | | |  __/ (__|   <\__ \
        |___/_| |_|\___|_|  \__,_|_|    \___|_| |_|\___|\___|_|\_\___/
===============================================================================
index                                                         OK       KO
- __main__.Cowboy_____________________________________ TOTAL: 0_______ 3_______
  - name_____________________________________________________ 0_______ 3_______

You can see here that the indexation table name is absent. You can call index_table_rebuild() to create and populate it.

>>> with sheraf.connection(commit=True):
...     Cowboy.index_table_rebuild("name")

Now that your index table is created and filled, you won’t be bothered by an IndexationWarning anymore.

>>> with sheraf.connection(commit=True):
...     with warnings.catch_warnings(record=True) as warns:
...         boss = Cowboy.create(name="Boss")
...         assert not warns