Searchlight API

Searchlight’s API adds authentication and Role Based Access Control in front of Elasticsearch’s query API.

Authentication

Searchlight, like other OpenStack APIs, depends on Keystone and the OpenStack Identity API to handle authentication. You must obtain an authentication token from Keystone and pass it to Searchlight in API requests with the X-Auth-Token header.

See Keystone Authentication for more information on integrating with Keystone.

Using v1

For the purposes of examples, assume a Searchlight server is running at the URL http://searchlight.example.com on HTTP port 80. All queries are assumed to include an X-Auth-Token header. Where request bodies are present, it is assumed that an appropriate Content-Type header is present (usually application/json).

Searches use Elasticsearch’s query DSL.

Elasticsearch stores each ‘document’ in an ‘index’, which has one or more ‘types’. Searchlight’s indexing service stores all resource types in their own document type, grouped by service into indices. For instance, the OS::Glance::Image and OS::Glance::Metadef types both reside in the searchlight index. type is unique to a resource type.

Document access is defined by each document type, for instance for glance images:

  • If the current user is the resource owner OR

  • If the resource is marked public

Some resources may have additional rules. Administrators have access to all resources, though by default searches are restricted to the current tenant unless all_projects is set in the search request body.

Querying available plugins

Searchlight indexes OpenStack resources as defined by installed plugins. In general, a plugin maps directly to an OpenStack resource type. For instance, a plugin might index nova instances, or glance images. There may be multiple plugins related to a given OpenStack project (an example being glance images and metadefs).

A given deployment may not necessarily expose all available plugins. Searchlight provides a REST endpoint to request a list of installed plugins. A GET request to http://searchlight.example.com/v1/search/plugins might yield:

{
    "plugins": [
        {
             "type": "OS::Glance::Image",
             "alias-searching": "searchlight-search"
             "alias-indexing": "searchlight-listener"
        },
        {
             "type": "OS::Glance::Metadef",
             "alias-searching": "searchlight-search"
             "alias-indexing": "searchlight-listener"
        }

    ]
}

This response shows the plugin information associated with the Glance image and metadef resources.

  • type: the resource group, which is used as the document type in Elasticsearch.

  • alias-searching: the Elasticsearch alias used for querying.

  • alias-indexing: the Elasticsearch alias used for indexing.

If desired, all indexed Glance images can be queried directly from Elasticsearch, rather than using Searchlight. Assuming an Elasticsearch server running on localhost, the following request can be made:

curl http://localhost:9200/searchlight-search/OS::Glance::Image/_search

More example searches

Results are shown here only where it would help illustrate the example. The query parameter supports anything that Elasticsearch exposes via its query DSL. There are normally multiple ways to represent the same query, often with some subtle differences, but some common examples are shown here.

Administrators - search all resources

By default, all users see search results restricted by access control; in practice, this is a combination of resources belonging to the user’s current tenant/project, and any fields that are restricted to administrators.

Administrators also have the option to view all resources, by passing all_projects in the search request body. For instance, a POST to http://searchlight.example.com/searchlight/v1/search:

{
    "query": {
        "match_all": {}
    },
    "all_projects": true
}

Restricting document index or type

To restrict a query to Glance image and metadef information only (both index and type can be arrays or a single string):

{
    "query": {
        "match_all": {}
    },
    "type": ["OS::Glance::Image", "OS::Glance::Metadef"]
}

If index or type are not provided they will default to covering as wide a range of results as possible. Be aware that it is possible to specify combinations of index and type that can return no results. In general type is preferred since type is unique to a resource.

Retrieving an item by id

To retrieve a resource by its OpenStack ID (e.g. a glance image), we can use Elasticsearch’s term query:

{
  "index": "searchlight",
  "query": {
    "term": {
      "id": "79fa243d-e05d-4848-8a9e-27a01e83ceba"
    }
  }
}

Limiting and paging results

Elasticsearch (and Searchlight) support paging through the size and from parameters (Searchlight also accepts limit and offset respectively as synonyms). from is zero-indexed. If size is zero, no results will be returned. This can be useful for retrieving the total number of hits for a query without being interested in the results themselves, or for aggregations:

{
  "query": {"match_all": {}},
  "size": 0
}

Gives:

{
  "hits": {
    "hits": [],
    "max_score": 0.0,
    "total": 40
  }
}

Limiting the fields returned

To restrict the source to include only certain fields using Elasticsearch’s source filtering:

{
  "type": "OS::Glance::Image",
  "_source": ["name", "size"]
}

Gives:

{
  "_shards": {
    "failed": 0,
    "successful": 1,
    "total": 1
  },
  "hits": {
    "hits": [
      {
        "_id": "76580e9d-f83d-49d8-b428-1fb90c5d8e95",
        "_index": "searchlight",
        "_score": 1.0,
        "_source": {
          "name": "cirros-0.3.2-x86_64-uec",
          "size": 3723817
        },
        "_type": "OS::Glance::Image"
      },
      ...
    ],
    "max_score": 1.0,
    "total": 4
  },
  "timed_out": false,
  "took": 1
}

Versioning

Internally an always-incrementing value is stored with search results to ensure that out of order notifications don’t lead to inconsistencies with search results. Normally this value is not exposed in search results, but including a search parameter version: true in requests will result in a field named _version (note the underscore) being present in each result:

{
  "index": "searchlight",
  "query": {"match_all": {}},
  "version": true
}

{
  "hits": {
    "hits": [
      {
        "_id": "76580e9d-f83d-49d8-b428-1fb90c5d8e95",
        "_index": "searchlight",
        "_version": 462198730000000000,
        ....
      },
      ....
    ]
  },
  ...
}

Sorting

Elasticsearch allows sorting by single or multiple fields. See Elasticsearch’s sort documentation for details of the allowed syntax. Sort fields can be included as a top level field in the request body. For instance:

{
  "query": {"match_all": {}},
  "sort": {"name": "desc"}
}

You will see in the search results a sort field for each result:

...
{
   "_id": "7741fbcc-3fa9-4ace-adff-593304b6e629",
   "_index": "glance",
   "_score": null,
   "_source": {
       "name": "cirros-0.3.4-x86_64-uec",
       "size": 25165824
   },
   "_type": "image",
   "sort": [
       "cirros-0.3.4-x86_64-uec",
       25165824
   ]
},
...

Wildcards

Elasticsearch supports regular expression searches but often wildcards within query_string elements are sufficient, using * to represent one or more characters or ? to represent a single character. Note that starting a search term with a wildcard can lead to extremely slow queries:

{
  "query": {
    "query_string": {
      "query": "name: ubun?u AND mysql_version: 5.*"
    }
  }
}

Highlighting

A common requirement is to highlight search terms in results:

{
  "type": "OS::Glance::Metadef",
  "query": {
    "query_string": {
      "query": "database"
    }
  },
  "_source": ["namespace", "description"],
  "highlight": {
    "fields": {
      "namespace": {},
      "description": {}
    }
  }
}

Results:

{
  "hits": {
    "hits": [
      {
        "_id": "OS::Software::DBMS",
        "_index": "searchlight",
        "_type": "OS::Glance::Metadef",
        "_score": 0.56079304,
        "_source": {
          "description": "A database is an organized collection of data. The data is typically organized to model aspects of reality in a way that supports processes requiring information. Database management systems are computer software applications that interact with the user, other applications, and the database itself to capture and analyze data. (http://en.wikipedia.org/wiki/Database)"
        },
        "highlight": {
          "description": [
            "A <em>database</em> is an organized collection of data. The data is typically organized to model aspects of",
            " reality in a way that supports processes requiring information. <em>Database</em> management systems are",
            " computer software applications that interact with the user, other applications, and the <em>database</em> itself",
            " to capture and analyze data. (http://en.wikipedia.org/wiki/<em>Database</em>)"
          ],
          "display_name": [
            "<em>Database</em> Software"
          ]
        }
      }
    ],
    "max_score": 0.56079304,
    "total": 1
  },
  "timed_out": false,
  "took": 3
}

Faceting

Searchlight can provide a list of field names and values present for those fields for each registered resource type. Exactly which fields are returned and whether values are listed is up to each plugin. Some fields or values may only be listed for administrative users. For some string fields, ‘facet_field’ may be included in the result and can be used to do an exact term match against facet options.

To list supported facets, issue a GET to http://searchlight.example.com/v1/search/facets:

{
  "OS::Glance::Image": [
    {
      "name": "status",
      "type": "string"
    },
    {
      "name": "created_at",
      "type": "date"
    },
    {
      "name": "virtual_size",
      "type": "long"
    },
    {
      "name": "name",
      "type": "string",
      "facet_field": "name.raw"
    },
    ...
  ],
  "OS::Glance::Metadef": [
    {
      "name": "objects.description",
      "type": "string"
    },
    {
      "name": "objects.properties.description",
      "type": "string",
      "nested": true
    },
    ...
  ],
  "OS::Nova::Server": [
    {
      "name": "status",
      "options": [
        {
          "doc_count": 1,
          "key": "ACTIVE"
        }
      ],
      "type": "string"
    },
    {
      "name": "OS-EXT-SRV-ATTR:host",
      "type": "string"
    },
    {
      "name": "name",
      "type": "string",
      "facet_field": "name.raw"
    },
    {
      "name": "image.id",
      "type": "string",
      "nested": false
    },
    {
      "name": "OS-EXT-AZ:availability_zone",
      "options": [
        {
          "doc_count": 1,
          "key": "nova"
        }
      ],
      "type": "string"
    }
    ...
  ]
}

Facet fields containing the ‘nested’ (boolean) attribute indicate that the field mapping type is either ‘nested’ or ‘object’. This can influence how a field should be queried. In general ‘object’ types are queried as any other field; ‘nested’ types require some additional complexity.

It’s also possible to request facets for a particular type by adding a type query parameter. For instance, a GET to http://searchlight.example.com/v1/search/facets?type=OS::Nova::Server:

{
  "OS::Nova::Server": [
    {
      "name": "status",
      "options": [
        {
          "doc_count": 1,
          "key": "ACTIVE"
        }
      ],
      "type": "string"
    },
    ...
  ]
}

As with searches, administrators are able to request facet terms for all projects/tenants. By default, facet terms are limited to the currently scoped project; adding all_projects=true as a query parameter removes the restriction.

It is possible to limit the number of options returned for fields that support facet terms. limit_terms restricts the number of terms (sorted in order of descending frequency). A value of 0 indicates no limit, and is the default.

It is possible to not return any options for facets. By default all options are returned for fields that support facet terms. Adding exclude_options=true as a query parameter will return only the facet field and not any of the options. Using this option will avoid an aggregation query being performed on Elasticsearch, providing a performance boost.

Aggregations

Faceting (above) is a more general form of Elasticsearch aggregation. Faceting is an example of ‘bucketing’; ‘metrics’ includes functions like min, max, percentiles.

Aggregations will be based on the query provided as well as restrictions on resource type and any RBAC filters.

To include aggregations in a query, include aggs or aggregations in a search request body. "size": 0 prevents Elasticsearch returning any results, just the aggregation, though it is valid to retrieve both search results and aggregations from a single query. For example:

{
  "query": {"match_all": {}},
  "size": 0,
  "aggregations": {
    "names": {
      "terms": {"field": "name"}
    },
    "earliest": {
      "min": {"field": "created_at"}
    }
  }
}

Response:

{
  "hits": {"total": 2, "max_score": 0.0, "hits": []},
  "aggregations": {
    "names": {
      "doc_count_error_upper_bound": 0,
      "sum_other_doc_count": 0,
      "buckets": [
        {"key": "for_instance1", "doc_count": 2},
        {"key": "instance1", "doc_count": 1}
      ]
    },
    "earliest": {
      "value": 1459946898000.0,
      "value_as_string": "2016-04-06T12:48:18.000Z"
    }
  }
}

Note that for some aggregations value_as_string may be more useful than value - for example, the earliest aggregation in the example operates on a date field whose internal representation is a timestamp.

The global aggregation type is not allowed because unlike other aggregation types it operates outside the search query scope.

Freeform queries

Elasticsearch has a flexible query parser that can be used for many kinds of search terms: the query_string operator.

Some things to bear in mind about using query_string (see the documentation for full options):

  • A query term may be prefixed with a field name (as seen below). If it is not, by default the entire document will be searched for the term.

  • The default operator between terms is OR

  • By default, query terms are case insensitive

For instance, the following will look for images with a restriction on name and a range query on size:

{
  "query": {
    "query_string": {
      "query": "name: (Ubuntu OR Fedora) AND size: [3000000 TO 5000000]"
    }
  }
}

Within the query string query, you may perform a number of interesting queries. Below are some examples.

Phrases
\"i love openstack\"

By default, each word you type will be searched for individually. You may also try to search an exact phrase by using quotes (“my phrase”) to surround a phrase. The search service may allow a certain amount of phrase slop - meaning that if you have some words out of order in the phrase it may still match.

Wildcards
python3.?
10.0.0.*
172.*.4.*

By default, each word you type will match full words only. You may also use wildcards to match parts of words. Wildcard searches can be run on individual terms, using ? to replace a single character, and * to replace zero or more character. ‘demo’ will match the full word ‘demo’ only. However, ‘de*’ will match anything that starts with ‘de’, such as ‘demo_1’. ‘de*1’ will match anything that starts with ‘de’ and ends with ‘1’.

Note

Wildcard queries place a heavy burden on the search service and may perform poorly.

Term Operators
+apache
-apache
web +(apache OR python)

Add a ‘+’ or a ‘-‘ to indicate terms that must or must not appear. For example ‘+python -apache web’ would find everything that has ‘python’ does NOT have ‘apache’ and should have ‘web’. This may also be used with grouping. For example, ‘web -(apache AND python)’ would find anything with ‘web’, but does not have either ‘apache’ or ‘python’.

Boolean Operators
python AND apache
nginx OR apache
web && !apache

You can separate search terms and groups with AND, OR and NOT (also written &&, || and !). For example, ‘python OR javascript’ will find anything with either term (OR is used by default, so does not need to be specified). However, ‘python AND javascript’ will find things that only have both terms. You can do this with as many terms as you’d like (e.g. ‘django AND javascript AND !unholy’). It is important to use all caps or the alternate syntax (&&, ||), because ‘and’ will be treated as another search term, but ‘AND’ will be treated as a logical operator.

Grouping
python AND (2.7 OR 3.4)
web && (apache !python)

Use parenthesis to group different aspects of your query to form sub-queries. For example, ‘web OR (python AND apache)’ will return anything that either has ‘web’ OR has both ‘python’ AND ‘apache’.

Facets
name:cirros
name:cirros && protected:false

You may decide to only look in a certain field for a search term by setting a specific facet. This is accomplished by either selecting a facet from the drop down or by typing the facet manually. For example, if you are looking for an image, you may choose to only look at the name field by adding ‘name:foo’. You may group facets and use logical operators.

Range Queries
size:[1 TO 1000]
size:[1 TO *]
size:>=1
size:<1000

Date, numeric or string fields can use range queries. Use square brackets [min TO max] for inclusive ranges and curly brackets {min TO max} for exclusive ranges.

IP Addresses
172.24.4.0/16
[10.0.0.1 TO 10.0.0.4]

IPv4 addresses may be searched based on ranges and with CIDR notation.

Boosting
web javascript^2 python^0.1

You can increase or decrease the relevance of a search term by boosting different terms, phrases, or groups. Boost one of these by adding ^n to the term, phrase, or group where n is a number greater than 1 to increase relevance and between 0 and 1 to decrease relevance. For example ‘web^4 python^0.1’ would find anything with both web and python, but would increase the relevance for anything with ‘web’ in the result and decrease the relevance for anything with ‘python’ in the result.

Reserved Characters
python \(3.4\)

The following characters are reserved and must be escaped with a leading (backslash):

+ - = && || > < ! ( ) { } [ ] ^ " ~ * ? : \

Advanced Features

CORS - Accessing Searchlight from the browser

Searchlight can be configured to permit access directly from the browser. For details on this configuration, please refer to the OpenStack Cloud Admin Guide.