ElasticSearch needs configuration settings for analyzers

[SacNaturalFoods]

The default EdgeNgram analyzer tokenizer is "lowercase", which seems to throw out digits, so I'm screwed if I want to search zip codes or employee numbers with partial strings.

Setting the tokenizer to "edgeNGram" and filter to ["edgeNGram", "lowercase"] allows me to partially search both letters and numbers, but these settings are in haystack.backends.elasticsearch_backend. Shouldn't this be a haystack configuration setting, or am I doing something wrong?

[racedo]

+1

[speedplane]

You are correct that elastic search settings are hard-coded into the backend. I was able to modify these settings by subclassing the elastic search backend and modifying DEFAULTSETTINGS in __init__. It would be nice if there was a getter/setter for the settings, but the subclassing solution works just fine. Note that you'll have to also modify settings.py to point to your new backend.

[racedo]

I modify directly the haystack/backends/elasticsearch_backend.py to achieve the same. It's not only the DEFAULT_SETTINGS that need to be modified, it's hardcoded to use the snowball analyzer for anything that's not a NgramField or EdgeNgramField and there's no setting for the default analyzer either, I guess an ideal solution would be:

1 - Allow configuration for the settings
2 - Allow configuration for the mapping
3 - Allow configuration for the default analyzer

[speedplane]

Right, I ran into that issue too. In your subclass, you need to override the build_schema method:

    def build_schema(self, fields):
        # Convert all "snowball" analyzers into the DEFAULT_ANALYZER analyzer set in __init__
        content_field_name, mapping = \
            super(ElasticsearchSearchBackendSubclass, self).build_schema(fields)
        for field_name, field_mapping in mapping.items():
            if "analyzer" in field_mapping and field_mapping["analyzer"] == "snowball":
                field_mapping["analyzer"] = self.DEFAULT_ANALYZER
        return content_field_name, mapping

Of course, a real solution with getter and setters would be preferable.

[bitcity]

👍 on ability to override backend settings for analyzers without monkey-patching *_backend.py files.