| Version: | 0.5.0 |
|---|---|
| Author: | Ben Lopatin (http://benlopatin.com) |
Configurable indexing and other extras for Haystack (with ElasticSearch biases).
Full documentation is on Read the Docs.
- Django: tested against Django 1.8, and 1.9
- Haystack: tested against Haystack 2.4.0, it should work with any combination of Haystack and Django that work
- ElasticSearch: presumably any newish version will do, however the only version tested against so far is 0.19.x
Some of these features are backend agnostic but most features have ElasticSearch in mind.
For more background see the blog post Stretching Haystack's ElasticSearch Backend.
The search mapping provided by Haystack's ElasticSearch backend includes brief but sensible defaults for nGram analysis. You can globaly 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.
Before your new analyzer can be used you will need to change your Haystack engine and rebuild/update your index. In your settings.py modify HAYSTACK_CONNECTIONS accordingly:
HAYSTACK_CONNECTIONS = {
'default': {
'ENGINE': 'elasticstack.backends.ConfigurableElasticSearchEngine',
'URL': env_var('HAYSTACK_URL', 'http://127.0.0.1:9200/'),
'INDEX_NAME': 'haystack',
},
}
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.
If you want to specify a custom search_analyzer for nGram/EdgeNgram fields, define it with the ELASTICSEARCH_DEFAULT_NGRAM_SEARCH_ANALYZER settings:
ELASTICSEARCH_DEFAULT_NGRAM_SEARCH_ANALYZER = 'standard'
Alternatively you can configure index mapping per index. This is usefull for multilanguage index settup. In this case HAYSTACK_CONNECTION contains key SETTINGS_NAME have to match with name in ELASTICSEARCH_INDEX_SETTINGS:
HAYSTACK_CONNECTIONS = {
'default': {
'ENGINE': 'elasticstack.backends.ConfigurableElasticSearchEngine',
'URL': env_var('HAYSTACK_URL', 'http://127.0.0.1:9200/'),
'INDEX_NAME': 'haystack',
'SETTINGS_NAME': 'cs',
'DEFAULT_ANALYZER': 'czech_hunspell',
'DEFAULT_NGRAM_SEARCH_ANALYZER': 'standard',
},
}
ELASTICSEARCH_INDEX_SETTINGS = {
'cs': {
"settings": {
"analysis": {
"analyzer": {
"czech_hunspell": {
"type": "custom",
"tokenizer": "standard",
"filter": ["stopwords_CZ", "lowercase", "hunspell_CZ", "stopwords_CZ", "remove_duplicities"]
}
},
"filter": {
"stopwords_CZ": {
"type": "stop",
"stopwords": ["právě", "že", "test", "_czech_"],
"ignore_case": True
},
"hunspell_CZ": {
"type": "hunspell",
"locale": "cs_CZ",
"dedup": True,
"recursion_level": 0
},
"remove_duplicities": {
"type": "unique",
"only_on_same_position": True
},
}
}
}
},
}
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
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.
Note
The SearchMixin uses the attribute named queryset for the resultant SearchQuerySet. Naming this attribute searchqueryset would make more sense semantically and hew closer to Haystack's naming convention, however by using the queryset attribute shared by other Django view mixins it is relatively easy to combine search functionality with other mixins and views.
Make a change and wonder why your results don't look as expected? The management command show_mapping will print the current mapping for your defined search index(es). At the least it may show that you've simply forgotten to update your index with new mappings:
python manage.py show_mapping
By default this will display the existing_mapping which shows the index, document type, and document properties.:
{
"haystack": {
"modelresult": {
"properties": {
"is_active": {
"type": "boolean"
},
"text": {
"type": "string"
},
"published": {
"type": "date",
"format": "dateOptionalTime"
}
}
}
}
}
If you provide the --detail flag this will return only the field mappings but including additional details, such as boost levels and field-specific analyzers.:
{
"is_active": {
"index": "not_analyzed",
"boost": 1,
"store": "yes",
"type": "boolean"
},
"text": {
"index": "analyzed",
"term_vector": "with_positions_offsets",
"type": "string",
"analyzer": "custom_analyzer",
"boost": 1,
"store": "yes"
},
"pub_date": {
"index": "analyzed",
"boost": 1,
"store": "yes",
"type": "date"
}
}
Provided the name of an indexed model and a key it generates and prints the generated document for this object:
python manage.py show_document myapp.MyModel 19181
The JSON document will be formatted with 'pretty' indenting.
The form, view, and backend functionality in this project is considered stable. Test coverage is not substantial, but is run against Django 1.8 through Django 1.10 on Python 2.7, 3.4, and 3.5.
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.