elasticstack 0.0.4 (experimental)

Configurable indexing and other extras for Haystack (with ElasticSearch biases)

Configurable index mapping

The search mapping provided by Haystack's ElasticSearch backend includes brief but sensible defaults for nGram analysis. You can add change these settings or add your own mappings by providing a mapping dictionary using ELASTICSEARCH_INDEX_SETTINGS in your settings file. This example takes the default mapping and adds a synonym analyzer:

ELASTICSEARCH_INDEX_SETTINGS = {
    'settings': {
        "analysis": {
            "analyzer": {
                "synonym_analyzer" : {
                    "type": "custom",
                    "tokenizer" : "standard",
                    "filter" : ["synonym"]
                },
                "ngram_analyzer": {
                    "type": "custom",
                    "tokenizer": "lowercase",
                    "filter": ["haystack_ngram", "synonym"]
                },
                "edgengram_analyzer": {
                    "type": "custom",
                    "tokenizer": "lowercase",
                    "filter": ["haystack_edgengram"]
                }
            },
            "tokenizer": {
                "haystack_ngram_tokenizer": {
                    "type": "nGram",
                    "min_gram": 3,
                    "max_gram": 15,
                },
                "haystack_edgengram_tokenizer": {
                    "type": "edgeNGram",
                    "min_gram": 2,
                    "max_gram": 15,
                    "side": "front"
                }
            },
            "filter": {
                "haystack_ngram": {
                    "type": "nGram",
                    "min_gram": 3,
                    "max_gram": 15
                },
                "haystack_edgengram": {
                    "type": "edgeNGram",
                    "min_gram": 2,
                    "max_gram": 15
                },
                "synonym" : {
                    "type" : "synonym",
                    "ignore_case": "true",
                    "synonyms_path" : "synonyms.txt"
                }
            }
        }
    }
}

The synonym filter is ready for your index, but will go unused yet.

The default analyzer for non-nGram fields in Haystack's ElasticSearch backend is the snowball analyzer. A perfectly good analyzer but not necessarily what you need. It's also language specific (English by default).

Specify your analyzer with ELASTICSEARCH_DEFAULT_ANALYZER in your settings file:

ELASTICSEARCH_DEFAULT_ANALYZER = 'synonym_analyzer'

Now all your analyzed fields, except for nGram fields, will be analyzed using synonym_analyzer.

Field based analysis

Even with a new default analyzer you may want to change this on a field by field basis as fits your needs. To do so, use the fields from elasticstack.fields to specify your analyzer with the analyzer keyword argument:

from haystack import indexes
from elasticstack.fields import CharField
from myapp.models import MyContent

class MyContentIndex(indexes.SearchIndex, indexes.Indexable):
    text = CharField(document=True, use_template=True,
            analyzer='synonym_analyzer')

def get_model(self):
    return MyContent

Django CBV style views

Haystacks's class based views predate the inclusion of CBVs into the Django core and so the paradigms are different. This makes it harder to impossible to make use of view mixins.

The bundled SearchView and FacetedSearchView classes are based on django.views.generic.edit.FormView using the SearchMixin and FacetedSearchMixin, respectively. The SearchMixin provides the necessary search related attributes and overloads the form processing methods to execute the search.

The SearchMixin adds a few search specific attributes:

• load_all - a Boolean value for specifying database lookups
• queryset - a default SearchQuerySet. Defaults to EmtpySearchQuerySet
• search_field - the name of the form field used for the query. This is added to allow for views which may have more than one search form. Defaults to q.

Management commands

python manage.py show_mapping

python manage.py show_document myapp.MyModel 19181

Why not add this stuff to Haystack?

This project first aims to solve problems related specifically to working with ElasticSearch. Haystack is 1) backend agnostic (a good thing), 2) needs to support existing codebases, and 3) not my project. Most importantly, adding these features through a separate Django app means providing them without needing to fork Haystack. Hopefully some of the features here, once finalized and tested, will be suitable to add to Haystack.